The Contraceptive Forecasting Handbook for Family Planning and HIV/AIDS Prevention Programs

Publication date: 2000

FPLM FPLM The Family Planning Logistics Management project is funded by the Commodities Security and Logistics Division (CSL) of the Office of Population and Reproductive Health (PRN) of the Bureau for Global Health (GH) of the U.S. Agency for International Development (USAID). Implemented by John Snow, Inc. (JSI) (contract no. CCP-C-00-95-00028-00), and subcontractors (The Futures Group International [TFGI] and the Program for Appropriate Technology in Health [PATH]), the FPLM project works to ensure the continuous supply of high-quality health and family planning products in developing countries. FPLM also provides technical management and analysis of two USAID databases, the contraceptive procurement and shipping database (NEWVERN) and the Population, Health and Nutrition Projects Database (PPD). This document does not necessarily represent the views or opinions of USAID. It may be reproduced if credit is given to FPLM/JSI. Recommended Citation Family Planning Logistics Management (FPLM). 2000. Contraceptive Forecasting Handbook for Family Planning and HIV/AIDS Prevention Programs. Arlington, Va.: FPLM/John Snow, Inc., for the U.S. Agency for International Development. Abstract The Contraceptive Forecasting Handbook for Family Planning and HIV/AIDS Prevention Programs is designed as a reference book for forecasting commodity needs for family planning and HIV/AIDS prevention programs. Topics include general methodological considerations, data sources and alternative techniques for preparing forecasts of consumption, special considerations in forecasting for new programs and HIV/AIDS prevention programs, methods for validating the forecasts, procedures for calculating quantities of contraceptives required based on the consumption forecast, and methods for monitoring the forecast over time. Second printing. 2005. John Snow, Inc./DELIVER, for USAID (contract no. HRN-C-00-00-00010-00). For further information, contact— Project Director John Snow, Inc./DELIVER 1616 North Fort Myer Drive, 11th Floor Arlington, VA 22209 USA Phone: 703-528-7474 Fax: 703-528-7480 Email: deliver_project@jsi.com Internet: deliver.jsi.com Project Officer Commodities Security and Logistics Division (CSL) Office of Population and Reproductive Health (PRN) Bureau for Global Health (GH) U.S. Agency for International Development (USAID) 1300 Pennsylvania Avenue, NW Washington, DC 20523 USA Phone: 202-212-4539 Fax: 202-216-3404 Internet: www.usaid.gov Note to Readers The Contraceptive Forecasting Handbook for Family Planning and HIV/AIDS Prevention Programs is designed as a reference book for forecasting commodity needs for family planning and HIV/AIDS prevention programs, not as a book to be read cover-to-cover. The Forecasting Handbook follows the sequence of steps required to produce and validate a complete forecast, and then to estimate procurement requirements and monitor progress and performance over time. However, each chapter is written to be as independent as possible of other chapters without being unnecessarily repetitive. You can turn to particular chapters for help and guidance whenever you need to accomplish a particular forecasting task—preparing a forecast using one or more of the different forecasting methodologies, validating or reconciling forecasts made by different forecasting methods, or calculating quantities of particular commodities to procure based on your forecast of consumption. The exceptions to this general rule are chapters 2 and 3 (Extrapolation from Historical Data and Corrections for Missing or Erroneous Data), which describe techniques applicable to all forecasts made from historical data. All readers should review the Preface, which describes the purpose and intended audiences for the Forecasting Handbook. A reader new to forecasting should study chapter 1 (Introduction) carefully to understand the basic concepts of forecasting and requirements estimation; all readers should skim this chapter to learn the terminology used throughout the handbook. Anyone who needs to prepare a forecast based on historical data (i.e., logistics data or service statistics) should review chapters 2 and 3 (Extrapolation from Historical Data and Corrections for Missing or Erroneous Data) carefully. These chapters describe the essential techniques for such forecasts. Chapters 4 through 7 describe techniques for preparing forecasts based on four different data sources—logistics data, service statistics, population data, and distribution system capacity. Readers who want to make one or more forecasts using these sources should study the appropriate chapter or chapters. Chapter 8 (Estimating Consumption for New Programs) describes special considerations for forecasting in new programs. This chapter assumes that the reader is familiar with the basic forecasting techniques described in chapters 4 through 7. Chapter 9 (Estimating Consumption for HIV/AIDS Prevention Programs) describes special considerations for forecasting condom needs for HIV/AIDS programs. This chapter also assumes that the reader is familiar with the forecasting techniques described in chapters 4 through 7. iii The Forecasting Handbook Chapter 10 (Validating and Reconciling the Forecast[s]) describes techniques for validating forecasts by comparing two or more forecasts made by different techniques. All readers who prepare forecasts should study this chapter. Chapter 11 (Requirements Estimation) explains how to calculate quantities of product that must be procured or obtained from donors after the forecast has been made. Readers who must make such calculations should study this chapter. Chapter 12 (Monitoring the Forecast and the Distribution Cycle) describes the process of monitoring progress over time, so that procurement quantities and future forecasts can be adjusted as circumstances change. In many cases, this task does not fall to the forecaster who prepared the original projections. However, he or she should make sure that someone undertakes this monitoring function; otherwise, the forecasting job is not complete. The appendices to the handbook provide detailed additional information for specific topics covered in the text. Of particular interest to general readers is appendix 1 (Related Publications), which lists additional references and contact information for organizations that can provide further information or assistance. For readers who need a thorough understanding of all the techniques described in the text, appendix 6 (An Example Forecast Using All Data Sources) provides a complete example of a forecast and requirements estimate prepared using all of the techniques. iv Table of Contents Page Preface .xi 1. Introduction . 1 1.1. Overview and Methodological Considerations. 1 1.1.1. Why Forecasts Are Prepared . 1 1.1.2. Short- versus Long-Term Forecasting . 2 1.2. Definitions . 2 1.3. Steps in the Process. 4 1.4. Forecasting Methods and Data Sources. 6 2. Extrapolation from Historical Data . 9 2.1. Organizing Data in a Time Series. 9 2.2. Extrapolation Using Simple Averages .11 2.3. Extrapolation Using Linear Trends .12 2.4. Drawing a Line by Eye .14 2.5. Extrapolation Using the Procedure of Semi-Averages .16 2.6. Extrapolation Using a Straight Line of Regression .16 2.7. Extrapolation When the Data Show Non-Linear Trends .17 3. Corrections for Missing or Erroneous Data .21 3.1. Adjustment for Incomplete Reporting .21 3.2. Adjustment for Missing Time Periods .22 3.2.1. Where the Trend Is Reasonably Stable .23 3.2.2. Where the Trend Is Upward or Downward .23 3.2.3. Where the Trend Shows a Seasonal Pattern.23 3.3. Adjustment for Both Incomplete Reporting and Missing Time Periods .25 4. Estimating Consumption Using Trends in Logistics Data .27 4.1. Data Sources and Limitations .27 4.2. Evaluating the Quality of Logistics Data .28 4.3. Correcting for Flawed Logistics Data .30 4.3.1. When Consumption Data Do Not Exist .30 4.3.2. When Consumption Data Are Not Reported.32 4.3.3. When Consumption and Losses Are Not Distinguished .32 4.3.4. When There Were Stockouts .33 4.4. Completing and Adjusting the Logistics Data-Based Forecast.34 5. Estimating Consumption Based on Service Statistics .35 5.1. Data Sources and Limitations .35 v The Forecasting Handbook 5.2. Evaluating the Quality of Service Data. 37 5.3. Completing and Adjusting the Service Data-Based Forecast. 37 5.3.1. Adjusting the Projection Based on Program Plans . 38 5.3.2. Calculating Commodity Consumption from Visit Projections. 39 6. Estimating Consumption Using Population Data . 43 6.1. Manual versus Automated Projections. 44 6.2. Data Requirements and Sources. 45 6.3. Evaluating the Quality of Population Data. 46 6.4. Steps in Preparing a Population Data-Based Forecast Manually. 48 6.5. Gathering and Adjusting Data for the Beginning Year of the Forecast . 48 6.5.1. Choosing the Base Year for the Projection . 49 6.5.2. Estimating Women of Reproductive Age for the Base Year . 50 6.5.3. Estimating the Actual Population at Risk of Pregnancy . 51 6.5.4. Choosing the Appropriate Contraceptive Prevalence Rate for the Base Year. 51 6.5.5. Calculating the Method Mix. 52 6.5.6. Estimating the Brand Mix. 52 6.5.7. Estimating the Proportion of National Contraceptive Use Attributable to the Program (Source Mix) . 53 6.6. Estimating WRA, CPR, Method Mix, and Source Mix for the Final Forecast Year . 53 6.6.1. Estimating WRA for the Final Forecast Year . 54 6.6.2. Estimating CPR for the Final Forecast Year . 55 6.6.3. Estimating Method and Brand Mix for the Final Forecast Year . 58 6.6.4. Estimating the Proportion of National Contraceptive Use Attributable to the Program (Source Mix) for the Final Forecast Year. 58 6.7. Estimating Changes in WRA, CPR, Method Mix, and Source Mix over the Forecast Period . 59 6.7.1. Estimating Intermediate Values for WRA. 59 6.7.2. Estimating Intermediate Values for CPR, Method Mix, and the Source Mix. 59 6.8. Calculating Commodity Consumption for Future Time Periods. 60 6.8.1. General Calculation for Population Data Forecasts. 61 6.8.2. Using Couple-Years of Protection Conversion Factors to Estimate Consumption for Short-Term Contraceptive Methods . 61 6.8.3. Using CYP Factors for Estimating Consumption of Long-Term Contraceptive Methods . 63 6.9. Using Spectrum/FamPlan for Contraceptive Forecasting . 64 6.10. Steps in Preparing a Population Data-Based Forecast Using Spectrum/FamPlan . 65 6.11. Gathering and Adjusting Data for the Spectrum/FamPlan Base Year. 65 6.12. Estimating Inputs for the Final Year of the Spectrum Forecast . 66 6.13. Completing the Spectrum/FamPlan Forecast . 66 7. Estimating Consumption Based on Distribution System Capacity. 69 7.1. Data Sources and Limitations. 70 7.2. Completing the Distribution System Capacity-Based Forecast . 71 7.2.1. Calculating Storage Capacity Requirements for a Single Facility . 72 7.2.2. Calculating Transport Capacity for a Single Transportation Link. 74 7.2.3. Preparing the Aggregate Delivery Capacity Forecast . 76 vi Table of Contents 8. Estimating Consumption for New Programs.79 8.1. Characteristics of an Acceptable Program Plan .79 8.2. Evaluating the Validity of the Program Plan .80 8.3. Completing the Forecast(s) .81 9. Estimating Consumption for HIV/AIDS Prevention Programs.83 9.1. Estimating Consumption Using Logistics Data .83 9.2. Estimating Consumption Based on Service Statistics .84 9.3. Estimating Consumption Using Population Data.84 9.4. Estimating HIV/AIDS Condom Consumption Based on Demographic Surveys.88 9.5. Estimating Consumption Based on Distribution System Capacity .91 9.6. Maximum Rates of Growth for HIV/AIDS Prevention Programs.92 10. Validating and Reconciling the Forecast(s) .93 10.1. The Need for Forecast Validation .93 10.2. Evaluation of Individual Forecast Quality .94 10.3. Forecast Reconciliation.95 11. Requirements Estimation . 103 11.1. The Basic Calculation for Requirements Estimation. 103 11.2. Determining Current Stock on Hand . 106 11.2.1. Estimating Stock on Hand at All Program Locations . 106 11.2.2. Estimating Stock on Hand at the Beginning of the Forecast Period . 107 11.3. Determining Shipments Already Received/on Order . 108 11.4. Estimating Current and Future Losses. 109 11.5. Identifying Other Adjustments to Inventory . 109 11.6. Determining Desired Inventory Levels . 110 11.7. Determining Desired Shipment Frequency . 112 11.8. Preparing Multi-Year Requirements Estimates . 113 12. Monitoring the Forecast and the Distribution Cycle . 115 12.1. Monitoring the Distribution Cycle . 115 12.2. PipeLine. 117 12.3. Monitoring the Forecast . 119 vii The Forecasting Handbook Figures 1. Quantity of IUDs Consumed by Clinic 1 in 1999 and Forecasts for 2000 . 12 2. Quantity of IUDs Consumed by Clinic 2 in 1999 and Forecasts for 2000 . 14 3. Quantity of IUDs Consumed by Clinic 3 in 1999 and Forecasts for 2000 . 15 4. Quantity of IUDs Consumed by Clinic 4 in 1999 and Forecasts for 2000 . 18 5. CY 1999 Service Activity at Clinic 5 and Forecasts for 2000: Orals . 40 6. Comparison of Three Alternative Projections for Lo-Femenal for Anyland.101 7. Depo-Provera Stock Status .119 8. Logistics Data-Based Projection for Anyland.171 9. Service Statistics Data-Based Projections for Anyland .178 10. Population Data-Based Projection for Anyland .182 11. Comparison of Three Alternative Projections for Anyland .187 Tables 1. Common Forecasting Situations .8 2. IUD Consumption in Four Facilities in CY1999. 10 3. IUD Consumption in Clinic 5 in CY1998 and CY1999. 24 4. Forecasts Using Logistics Data. 28 5. Logistics Data: Problems and Solutions. 30 6. Forecasts Using Service Statistics . 36 7. Service Statistics Data: Problems and Solutions. 38 8. CY1999 Service Activity at Clinic 5 and Projections for CY2000 Oral Contraceptives. 39 9. Forecasts Using Population Data . 44 10. Population Data: Problems and Solutions . 47 11. Population Data for Anyland for 1999 Base Year Forecast . 49 12. Population Data for Anyland for Final Forecast Year (2002). 54 13. Annual Percentage Change in Contraceptive Prevalence by Level of Family Planning Program Effort (1982–1989) and Socioeconomic Setting (1985) . 56 14. Declines in TFR from 1975 to 1990 by Level of Program Effort (1982–1989) and Socioeconomic Setting (1985). 57 15. Couple-years of Protection Conversion Factors . 62 viii Table of Contents 16. Spectrum/FamPlan, Commodities by Method .67 17. Forecasts Using Distribution System Capacity.71 18. New Program Planning: Issues to Consider .81 19. Condom–Specific Prevalence Rates .86 20. Condom Requirements: Kenya 1998–2010 .91 21. Evaluating Logistics Data-Based Forecasts .96 22. Evaluating Service Statistics Data-Based Forecasts.97 23. Evaluating Population Data-Based Forecasts .98 24. Evaluating Distribution System Capacity-Based Forecasts .99 25. Alternate Forecasts of Contraceptive Needs: 2000 . 100 26. Net Supply Requirements for CY2000 (1,000s) . 105 27. Data Sources, Problems, and Solutions . 105 28. Estimated Stock on Hand at All Levels as of January 1, 2000. 108 29. Calculating Desired Stock at End of Period (In Months of Supply). 111 30. 2000 Contraceptive Procurement Table . 114 31. Procurement and Pipeline, Directorate of Family Planning . 118 32. Life Expectancy, Kenya . 143 33. Number of Women of Reproductive Age, Kenya . 144 34. Number of Women of Reproductive Age and Total Fertility Rate, Kenya. 145 35. Contraceptive Prevalence Rate and Method Mix, Kenya . 146 37. Contraceptive Source Mix, Kenya . 147 37. Total Fertility Rate, Kenya . 148 38. Percent of Women of Reproductive Age in Union, Kenya. 149 39. Postpartum Insusceptibility, Kenya. 150 40. Induced Abortion Rate, Kenya. 151 41. Sterility Rate, Kenya . 152 42. Trends in the Percentage of Married Women Currently Using Contraception, by Country. 154 43. Summary Logistics Data for Region 1: Lo-Femenal . 167 44. Summary Logistics Data for Region 2: Lo-Femenal . 168 45. Summary Logistics Data for Region 3: Lo-Femenal . 169 46. Summary Logistics Data for Anyland: Lo-Femenal. 170 47. Summary Service Data for Region 1: Lo-Femenal . 172 48. Summary Service Data for Region 2: Lo-Femenal . 175 49. Summary Service Data for Region 3: Lo-Femenal . 176 ix The Forecasting Handbook 50. Conversion of Service Data Totals into Consumption Estimates: Lo-Femenal.177 51. Population Data for Anyland for 1999 Base Year Forecast .179 52. Population Data-Based Projection for Anyland MOH (1999–2002) .180 53. Evaluating Anyland’s Logistics Data-Based Forecast .184 54. Evaluating Anyland’s Service Statistics Data-Based Forecast.185 55. Evaluating Anyland’s Population Data-Based Forecast .186 56. Final Consumption Forecast for Anyland.188 57. Anyland MOH In-Country Distribution System Structure.189 58. Evaluating Anyland’s Distribution System Capacity-Based Forecast.196 59. SDP Stock Data for Anyland: Lo-Femenal .198 60. Central, Regional, and SDP Stock Data for Anyland: Lo-Femenal.200 61. Lo-Femenal Shipments Received or Scheduled.201 62. 2000 Contraceptive Procurement Table .202 Appendices 1. Related Publications .121 2. Logistics Management Information System Assessment Guidelines.125 3. JSI/FPLM Spectrum Projection Preparation Guidelines .135 4. Levels and Trends of Contraceptive Use as Assessed in 1998.153 5. Weights and Volumes of Commonly Supplied Contraceptives .157 6. An Example Forecast Using All Data Sources .165 x Preface Family planning and HIV/AIDS prevention programs must manage their logistics systems properly if they are to be successful in meeting the demand for services. In particular, logistics managers must properly forecast the quantities of each method and brand of contraceptive (or condom) required, procure or arrange for the procurement of the required commodities, receive and clear products through customs as they arrive, distribute commodities through in-country distribution channels in a way that prevents stock imbalances, and dispense commodities in good condition to the clients who need them. This handbook describes forecasting procedures and techniques that are useful in forecasting contraceptive or HIV/AIDS condom needs, though, in fact, the methodologies are applicable to any health commodity. Topics include— ˆ general methodological considerations; ˆ data sources and alternative techniques for preparing forecasts of consumption; ˆ special considerations in forecasting for new programs and HIV/AIDS prevention programs; ˆ methods for validating the forecasts; ˆ procedures for calculating quantities of commodities required based on the consumption forecast; and ˆ methods for monitoring the forecast over time. Anyone who must prepare national-level forecasts of health commodity requirements can use this handbook. Thus the audience includes procurement and logistics management personnel in host-country family planning and HIV/AIDS prevention programs, national and international donor staff and expatriate advisors, and external technical assistance personnel. In an ideal situation, forecasting is not a periodic (annual or quarterly) activity, but is accomplished through constant monitoring of inventories, usage rates, and other informa­ tion that may affect future demand. If the logistics management information system (LMIS) of the program is properly designed and kept up-to-date, the needed information will be available to staff responsible for forecasting and procurement. This handbook does not, however, provide a complete description of appropriate LMIS forms and procedures or the many other components of the logistics management system that must also be in place. These related issues are covered extensively in the various documents listed in appendix 1. This handbook is the work of a number of people from USAID’s Family Planning Logistics Management (FPLM) project, both at John Snow, Inc. (JSI) and the Division of Reproductive Health, Centers for Disease Control and Prevention (CDC). xi The Forecasting Handbook As readers will quickly perceive, forecasting of contraceptive and HIV/AIDS condom needs (like forecasting for most other reasons) remains more of an art than a science. The techniques described here can help produce better forecasts. But constant monitoring of the supply situation, along with a willingness to modify shipment plans and the forecasts themselves, are the keys to ensuring that the right goods, in the right quantities, in the right condition are delivered to the right place, at the right time, for the right cost. xii 1. ] Introduction The Forecasting Handbook describes the process of planning for the acquisition of com­ modities that are needed for the successful operation of a family planning or HIV/AIDS prevention program. Chapter 1 discusses the methodological considerations of forecasting and the general processes of forecasting consumption, forecast validation, and commodity needs estimation. 1.1. Overview and Methodological Considerations Forecasting contraceptive consumption is as much an art as a science, especially in new programs with no historical data. For this reason, the Forecasting Handbook recommends using multiple approaches to forecast preparation, rather than a single approach. 1.1.1. Why Forecasts Are Prepared One important reason for attempting to predict future contraceptive or condom needs is the time that elapses between the request for a commodity and the arrival of that commodity at the location where it is to be used. Because preparing an order and then sending, proc­ essing, approving, dispatching, and ensuring that it reaches its destination takes time, it is essential to have advance knowledge of the quantities that must be purchased or produced. Other reasons for doing everything possible to determine product requirements in advance are the consequences of not having such forecasts available when they are needed. In the case of contraceptives, obvious consequences include forcing a couple to change from a method or brand with which they are satisfied to a new product; payment of higher prices for the same product; loss of time and money caused by unsuccessful visits to service centers; or discredit to the program when timely services are not provided. Such problems result in program dropouts and unwanted pregnancies. In the case of an HIV/AIDS preven­ tion program, a condom stockout may be fatal to the client. To all this must be added the additional costs resulting from underused services or emergency orders for commodities. 1 The Forecasting Handbook 1.1.2. Short- versus Long-Term Forecasting It is useful to distinguish between short- and long-term forecasting efforts. Although there is much overlap (indeed, it is difficult to get experts to agree on a definition), short- and long-term forecasts tend to be prepared by different program staff, for different purposes, using somewhat different techniques. Short- and medium-term projections of contraceptive or condom needs are made primarily to meet the immediate tactical objectives of any logistics system—obtaining appropriate amounts of each commodity to be issued throughout the distribution system and ultimately dispensed to clients. The output of this type of projection exercise is clear—quantities of contraceptives needed over a fixed period of time; schedules by which they should arrive; budgets and cost estimates where appropriate; and, if necessary, requests to donors for assistance in obtaining the products. This function tends to be the responsibility of middle managers, who must make these projections on a fixed timetable regardless of the quality of data available or the degree of specificity of their program’s short- and medium-term plans. This handbook presents ideas and procedures that can be applied immediately in such situations, using simple methods easily understood by anyone directly involved with the management of supplies. Long-term projections, which are more strategic than tactical in nature, require a greater knowledge of both the history and evolution of family planning programs worldwide and also the determinants of supply and demand for contraceptive services and materials in a particular society. Long-term projections may be prepared by local program managers or by a combination of program staff and outside consultants. Such forecasts are more complex to produce, requiring more extensive knowledge of forecasting techniques. However, because long-term projections are used for macro-level applications such as estimation of demo­ graphic trends and evaluation of program impact, they typically do not require the same precision as short-term forecasts used for procurement. Long-term forecasting issues are not covered explicitly in this handbook, although the mathematical techniques are basically the same as those used for short-term forecasting. 1.2. Definitions Logistics is the branch of management that ensures that resources needed by a working group—or the products required by a group of consumers—reach their destination in the required amount, in the least possible time, and at the least possible cost. This objective is often described as the six rights. The logistics system ensures— ˆ the right goods, ˆ in the right quantities, ˆ in the right condition, 2 Chapter 1 ] Introduction delivered ˆ to the right place, ˆ at the right time, and ˆ for the right cost. To achieve this objective, logistics managers must quantify future consumer needs. Webster’s Ninth New Collegiate Dictionary defines forecasting as— … to calculate or predict some future event or condition, usually as a result of rational study and analysis of available pertinent data. The accuracy of the forecast is directly related to the inherent predictability of the event, and to the completeness and the quality of the information available about past and present activity. For the purposes of this handbook, forecasting means estimating the consumption and losses for each contraceptive that will be distributed by a family planning or HIV/AIDS prevention program during some future period of time. Webster’s definition of demand is— … the quantity of a commodity or service wanted at a specified price and time. In the context of family planning or HIV/AIDS prevention services, price includes not only monetary and program personnel costs, but also the cost in time and inconvenience for the client who wishes to obtain services. Most programs want to increase the number of people who demand family planning (or condoms for HIV/AIDS prevention), but, in fact, they may be prevented from meeting that demand by a variety of constraints. Therefore, logisticians not only must try to forecast true demand, but also must take into consideration the program’s ability to deliver the commodi­ ties and services. In logistics terminology, managers are ultimately interested in the amount dispensed to clients—the quantity actually given to clients at the clinic, dispensary, shop, or field level of the distribution system. This is carefully distinguished from the amount issued—the quantity that is issued from one level to another within the distribution system (for example, from the central store to the regional stores). For reasons discussed in chapter 4, it is very important that dispensed-to-client data, rather than issues data, be used for forecasting wherever possible. Losses are those quantities of product that leave the distribution system for any reason other than being dispensed to clients. Losses are usually classified as either system losses or client losses. System losses are those that occur within the logistics system, such as expiration, damage, or theft. Client losses are those that occur after the client takes possession of the product. Because client losses are extremely difficult to measure and, in any case, largely beyond the control of a program’s logistics system, the terms use, con­ 3 The Forecasting Handbook sumption, dispensed to users, and dispensed to clients are usually considered interchangeable for logistics purposes. They are used interchangeably throughout this handbook. Consumption forecasts, no matter how they were made, should always be validated before use. Webster’s definition of validate is— … to support or corroborate on a sound or authoritative basis. For our purposes, validation means comparing two or more forecasts made using different methodologies to determine whether the forecast results are consistent, and, where they are not consistent, identifying strengths and weaknesses of each forecast to arrive at a best forecast of anticipated consumption and losses. After a forecast of the amounts expected to be dispensed to clients during a particular time period is finalized, program staff must take account of stocks that may already be on hand or on order before deciding how much to purchase or request from donors. This calculation process is known as requirements estimation. With estimates of quantities needed in hand, program managers must acquire the necessary products in a timely fashion. Webster’s definition of procure is— … to get possession of; obtain by particular care and effort. For our purposes, procurement means acquiring the contraceptives (through purchase, donation, or other means) and scheduling the contraceptive shipments. Finally, Webster’s definition of monitor is— … to watch, observe, or check, especially for a special purpose. Our special purposes in monitoring are to ensure that products are available at all times and in sufficient quantities to meet the anticipated demand of the program’s clients, and to ensure that losses are kept to a minimum. 1.3. Steps in the Process The entire forecasting process can be inferred from these definitions. The forecaster must— 1. Forecast not only the true demand for commodities, but also quantities that the program will actually dispense to clients and quantities that will be lost in the process. 2. Validate the estimates by comparing forecasts made by several methodologies. 3. Estimate requirements for obtaining commodities that are not available in sufficient quantities to meet anticipated needs. 4 Chapter 1 ] Introduction 4. Procure the commodities required through purchase or donations. 5. Monitor commodity consumption over time to correct supply imbalances that inevitably will occur, and gather data that will be needed for the next forecast. The basic steps that should be followed in completing these tasks are to— 1. Prepare a preliminary written schedule of the work, including travel schedules, appointments with key officials, and a final report to the program director and other responsible officers. Revise this schedule as the process proceeds, and keep it for future reference. 2. Collect, review, and evaluate data sources and other documents. 3. Visit key locations to interview staff and collect data, preferably following the supply chain down several distribution channels. Use these visits to determine data quality and identify gaps that must be filled before a forecast can be prepared. 4. Visit other programs and private sector outlets to determine the effect their activities will have on future demand and service delivery. 5. Analyze the information collected, focusing on the relevance of the data to future contraceptive use, and take steps to fill in gaps and correct for errors or identified deficiencies. 6. Prepare one, two, three, or four forecasts as discussed in the following chapters, depending on the number of separate data sources available and the purpose and scope of the final forecast. 7. Validate the primary forecast by comparing it to at least one forecast made by another technique. 8. Discuss the forecast(s) with host-country program managers and, where appropri­ ate, with U.S. Agency for International Development (USAID) and other donor staff to obtain consensus on the selection of a reasonable forecast. 9. Calculate procurement requirements by comparing stocks on hand or already on order to usage and loss forecasts. 10. Assist program staff in identifying source(s) of supply for quantities required. 11. Prepare proposed shipping schedules (separately for each source of supply) for quantities that must be procured. 12. Assist program staff in preparing documentation that they (or appropriate donor or procurement agencies) may require for ordering commodities. 5 The Forecasting Handbook 13. Monitor procurements, shipments, and consumption during the period of the forecast, adjusting both forecasts and procurement/shipment schedules as necessary based on actual quantities dispensed to clients. These elements could form the basis of a job description for the person responsible for forecasting. For family planning programs, the forecasting effort might focus on contraceptive methods that require relatively large quantities for constant resupply to clients (sometimes called major supply methods)—usually oral contraceptives, condoms, intrauterine devices (IUD), implants, injectables, and/or vaginal foaming tablets. However, other methods, such as foams, jellies, diaphragms, voluntary surgical contraception (VSC), rhythm, traditional methods, and others, should not be ignored; changes in the use of these methods will surely affect the demand for the major supply methods. In any case, early and frequent meetings with program staff are vital to the forecasting effort. Time can be wasted and dissatisfaction can result when all the players do not understand the “rules of the game.” Once everyone involved agrees on the procedures, the technical work of the forecast can begin. 1.4. Forecasting Methods and Data Sources Most of this handbook is devoted to quantitative methods for forecast preparation. The most important parts of the definitions of forecasting and procurement are those that say “as a result of rational study and analysis of available pertinent data” and “by particular care and effort.” Without reliable data and careful analysis, forecasting and procurement are little better than guesswork. Those preparing the forecast should collect relevant program and logistics information from their own data systems and, if possible, from suppliers, other programs, and other sources of family planning commodities and services in the geographical area served by the pro­ gram. These documents include— ˆ Records from the central contraceptive logistics management information system (LMIS). ˆ Previous forecasts and requirements estimates. ˆ Previous contraceptive procurement records. ˆ Program policy statements on service delivery. ˆ Relevant program planning documents. ˆ Demographic data. ˆ Demographic and Health Survey (DHS) reports and/or other survey reports that contain information on contraceptive prevalence, method mix, target population, clients’ sources of contraception, and others. 6 Chapter 1 ] Introduction ˆ Other program documents that show past performance and planned future direc­ tions of the program. Reports of program evaluations are particularly useful. ˆ Correspondence and other documents related to procurement for contraceptives, shipments, and scheduled shipments. ˆ Lists of principal program and current supplier officials, and resident officials of other potential suppliers and donor agencies. The primary requirement for accurate forecasts and procurement schedules is a reliable logistics database on contraceptive procurement, use, losses, and inventory levels, over time, at all program locations. In addition to logistics data, the forecaster should analyze program performance data and program plans for the future as well as the program’s existing or planned service delivery capacity. Demographic data for the catchment area should also be examined. These data are usually available from surveys, program evalua­ tions, program plans, service statistics, and the program’s LMIS. The following chapters describe methods for forecasting contraceptive consumption (or, for HIV/AIDS prevention programs, condom consumption), based on four different data sources— ˆ Logistics data (i.e., program data on historical consumption or issues). ˆ Service statistics from the organization distributing the product (i.e., data on clients and visits). ˆ Population data (i.e., demographic surveys). ˆ Distribution system capacity of the organization. Each of these sources, discussed in subsequent chapters, has advantages and limitations. Many programs, especially new ones, will not have data from all four sources. The types of forecasts that should be prepared depend on the types of data available and on the quality and reliability of data from each source. Other criteria for deciding which type(s) of forecast to use are the time period covered by the projection (short-, medium-, or long-term); the scope (for a town, region, country, or group of countries); and the purpose of the projection (for purchasing, budgeting, planning, or evaluation of program impact). Table 1 shows the forecasting methodologies FPLM typically uses for different types of forecasts. When projections are prepared for an entire country, or a major portion of the country, or when considerable volumes of product or large quantities of money are involved, it is advisable to use two, three, or, ideally, all four methods of projection, comparing the results to arrive at the “best” estimate. This strategy will increase considerably the probability of producing an accurate forecast. It allows projections made by one method to be compared with those made by a different procedure, providing an opportunity to discover weaknesses in the basic data or strengths that can be used later to validate the results of the calcula­ tions. The additional effort required to prepare several different types of projections is always more than compensated by gains in accuracy and reliability, and by reduction in the risk of projecting excessive or insufficient quantities, thus avoiding the losses that either error would cause. 7 The Forecasting Handbook Table 1. Common Forecasting Situations Short- or Medium-Term Forecasting for Existing Programs ‰ Historical logistics (consumption) data ‰ Service statistics ‰ Population (demographic) data Short- or Medium-Term Forecasting for New Programs ‰ Service statistics ‰ Population (demographic) data ‰ Distribution system capacity Long-Term Forecasting for New or Existing Programs ‰ Service statistics ‰ Survey data ‰ Distribution system capacity Because good data are so important to the forecasting process, the quality and complete­ ness of the data should be evaluated as the forecast is being prepared, and steps should be taken to correct any deficiencies. This will make the next forecasting cycle much easier. For all these techniques, it cannot be stressed enough that the quality of the forecast depends entirely on the quality of the data used in making it. Data are almost always incomplete or inaccurate to some extent; therefore, the ability to assess and compensate for data flaws is the key skill of the forecaster. Unless such adjustments are made appropriately, the forecasts will also be flawed, despite the mathematical precision of the calculations described in the following chapters. As you work through the procedures or use the tools described here, you must remember the first rule of data processing— Garbage in . Garbage out. 8 2. ] Extrapolation from Historical Data The first two types of projections for family planning or HIV/AIDS prevention programs— those based on logistics data and service statistics—use an identical mathematical technique called extrapolation to prepare the forecast. Indeed, any forecast based solely or primarily on historical data uses some form of this technique, which assumes that there is a discernable pattern of change in the historical data, and that this pattern will continue in the future. This chapter describes several variations of the extrapolation technique that can be per­ formed manually. Readers with access to a personal computer can find many software pro­ grams that automate these calculations. Although logistics data are used in the examples, the same techniques and formulas are used in making forecasts based on service statistics. 2.1. Organizing Data in a Time Series The first step in preparing an extrapolation is always to organize historical data into a time series, which is simply a table of contraceptive consumption over time. The quantity of IUDs consumed in each of the 12 months of a year, for example, form a time series. Table 2 pres­ ents four time series representing quantities of IUDs consumed in four different clinic locations during calendar year 1999. These data will be used to illustrate the different extrapolation techniques for preparing a forecast for calendar year 2000. The purpose of organizing data in this fashion is to observe the trend of the data items in the series, the variability of these values around an average or median value, and any patterns or models of change that repeat themselves. The time series also establishes the starting point for the projection into the future. 9 The Forecasting Handbook Table 2. IUD Consumption in Four Facilities in CY1999 Month Clinic 1 (Series 1) Clinic 2 (Series 2) Clinic 3 (Series 3) Clinic 4 (Series 4) January 10 10 18 10 February 11 11 16 13 March 12 12 20 17 April 13 13 22 22 May 14 14 19 30 June 15 15 23 27 July 16 16 24 29 August 17 17 20 19 September 18 18 27 21 October 19 19 28 14 November 20 20 30 11 December 21 32 26 12 The data from table 2 are obviously not very helpful organized simply as lists of numbers. The first step in extrapolating from any time series should always be to graph the historical data. It is customary to show time on the horizontal (x) axis of the graph and the variable being projected on the vertical (y) axis, though this is not mandatory. However, the divi­ sions for each axis must be of equal value, and each axis must extend far enough to allow the future projection to be drawn in. The steps to be followed are— 1. Create a graph with time on one axis and the quantity to be projected on the other axis (the quantity axis should be as long as possible, so the projection can be read more accurately). 2. Aggregate, smooth, or adjust data if necessary, using the techniques described in this and the following chapters. 3. Plot the available historical data for each time period. 4. Identify any observable trend in the data (stable, downward, upward, or cyclical). 5. Identify the magnitude and variability of deviations from the trend and decide, based on careful examination of the data, what data points, if any, must be corrected or discarded. 6. Select the trend line that most closely represents the historical data. 7. Choose the starting value for the forecast. 10 Chapter 2 ] Extrapolation from Historical Data 8. Draw a line representing the most probable extrapolation of the historical data through the future forecast period. 9. Read the values of the projection from the graph. The method used to make the extrapolation depends on what you see when you draw the graph. Possible manual techniques are the simple average, linear trend, drawing a line by eye, the procedure of semi-averages, linear regression, and more sophisticated decompo­ sition techniques for non-linear trends. 2.2. Extrapolation Using Simple Averages The easiest mathematical technique for extrapolation—and, unfortunately, the least use­ ful—is the simple average. The forecaster assumes that the future values of the variable being projected are just the average of the available historical data. Mathematically, the formula is expressed as— Total quantity consumed Estimated use in past n periods for next = period n The letter n is the number of past periods being considered. For the IUD consumption data at Clinic 1 in table 1, for example, the calculation is— Total quantity consumed 2000 for January Estimated use = 12 in past 12 months 186 = 12 = 15.5 This same figure is used for the remaining 11 months of 2000. Thus, the IUD consumption projection for 2000 for Clinic 1 is 15.5 x 12 = 186. Figure 1 shows the graph of historical IUD consumption for Clinic 1, along with two possi­ ble extrapolations. It is clear from the graph that the simple averages technique does not provide an appropriate forecast—the line for the simple average forecast is nothing like the pattern of historical data. This is because the historical data show a uniformly increasing trend in consumption of IUDs at Clinic 1. In such situations, a different extrapolation tech­ nique is needed. If the consumption figures at Clinic 1 were stable instead of increasing, however, simple averages would provide an adequate extrapolation. 11 The Forecasting Handbook Figure 1. Quantity of IUDs Consumed by Clinic 1 in 1999 and Forecasts for 2000 2.3. Extrapolation Using Linear Trends In cases such as Clinic 1, where a reasonably consistent increase or decrease in the variable being projected is found, a technique called linear trend can be used for extrapolation. This technique uses historical data from the first and last period to calculate the slope of the historical trend, projecting a straight line based on this slope. This is done most easily with a ruler, by drawing a straight line through the first and last historical points on the graph. The line is extended into the future for the period of the forecast. Mathematically, the formula for linear trend is— period for next Estimated use = period most recent Use in the + periodspast in use over Average change n 12 Chapter 2 ] Extrapolation from Historical Data where— Average change Use in period n − Use in period 1 inuse over = past n periods n − 1 So, in the example of Clinic 1— Average change 21 − 10 11 in use over = = = 1 past 12 months 12 − 1 11 Thus, usage is increasing at the rate of one IUD per month. Therefore— for January = 21 + 1 = Estimated use 22 2000 Continuing this logic, the estimated use is 23 for February, 24 for March, and so on. This line best represents the projection for 2000, based on simple examination of the graph in figure 1. In real life, of course, historical data rarely fall so precisely onto a straight line. The data for Clinic 2 in table 2 show performance identical to that of Clinic 1, except during the month of December. In this case, the linear trend calculation is— Average change 32 − 10 22 inuse over = = = 2 past 12 months 12 − 1 11 Thus, usage is increasing at the rate of two IUDs per month. Therefore— for January = 32 + 2 = Estimated use 34 2000 with a projection of 36 for February, 38 for March, and so on. This projection, shown in figure 2, is probably incorrect, demonstrating the limitations of the linear trend technique. The forecast is entirely dependent on the first and last historical data points; if these do not follow the pattern of the other data, the forecast may be com­ pletely wrong. In the case of Clinic 2, the forecaster must determine why the data for December 1999 are so different from the clinic’s experience earlier in the year, and decide whether to use different end points for the extrapolation. Was it simply a reporting error? If so, the error should be corrected or the data should be omitted for purposes of extrapolation. Did the same thing happen in December 1998? If so, perhaps December data should not be used for preparing the extrapolation, but an allowance should be made for a similar jump in Decem­ 13 The Forecasting Handbook ber 2000. Was the sudden increase due to a staffing change at the clinic; a new informa­ tion, education, and communication (IEC) program; or some other permanent change? If so, perhaps the linear trend extrapolation is realistic, or even low. These judgments must be made in collaboration with management staff of Clinic 2. Figure 2. Quantity of IUDs Consumed by Clinic 2 in 1999 and Forecasts for 2000 2.4. Drawing a Line by Eye Figure 3 plots the data and several projections for Clinic 3. These historical data are more realistic. There is some consistency and a perceptible (upward) trend, but the data do not fall neatly into a line. In such cases, the simplest technique consists of drawing a straight line with a ruler through the historical data, attempting to leave the same number of data 14 Chapter 2 ] Extrapolation from Historical Data points on each side of the line, both at the beginning and end of the line if possible. The by eye line in figure 3 satisfies the first of these criteria, but not the second—there is no way to do both. In this case, the results of the by eye extrapolation seem plausible. If program managers at Clinic 3 are comfortable with these results, then this extrapolation could be used as the projection. If not, one of the mathematically more precise techniques discussed later in this chapter should be used. Figure 3. Quantity of IUDs Consumed by Clinic 3 in 1999 and Forecasts for 2000 40 30 20 10 0 Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Feb Apr Jun Aug Oct Dec Feb Apr Jun Aug Oct Dec 1999 2000 Actual Use “By Eye” Linear Regression Semi-Averages The major drawback to the by eye technique is not its accuracy—certainly the projection shown in figure 3 appears reasonable. However, a by eye projection may be difficult for the forecaster to defend, and it is not replicable; another forecaster’s eye may see a slightly different line. If these issues are a concern, a mathematical technique should be used instead. 15 The Forecasting Handbook 2.5. Extrapolation Using the Procedure of Semi-Averages Another option is to calculate an average for the first half of the series and another for the second half, using the formula for simple averages shown earlier. These two values are plotted on the graph at the midpoint of the appropriate half of the series. A line is then drawn between the two dots and extended forward for the time period of the projection. This is called the procedure of semi-averages. Using the example of Clinic 3— Total quantity consumed Estimated use in January − June in first half = of series 6 (18 + 16 + 20 + 22 + 19 + 23)= 6 6 118 = 19.7= And— of series in second half Estimated use = 6 Decemberin July ­ Total quantity consumed = (24 + 20 + 27 + 28 + 30 + )26 6 155 = = 25.8 6 The first of these points is plotted at the midpoint of the January–June series (i.e., be­ tween March and April), and the second is plotted at the midpoint of the July–December series (i.e., between September and October). As figure 3 shows, the extrapolation using semi-averages in this case gives a projection similar to the by eye technique. This projec­ tion also appears to be reasonable. Results obtained with this procedure are generally acceptable, and are usually better than the results of any of the techniques described earlier. 2.6. Extrapolation Using a Straight Line of Regression Another procedure—requiring more elaborate math and a larger number of historical data points—involves calculating a straight line of regression using the technique of least squares or another equivalent method. While the mathematics are complicated to describe, 16 Chapter 2 ] Extrapolation from Historical Data regression is essentially an automated version of the by eye technique. The regression technique draws a straight line through the data that minimizes the total of the differences between the actual data points and the values depicted by the regression line. Prior to the age of computers, this technique required considerable time and effort. Today, it can be managed easily with a personal computer and even with some calculators. The formulas can be found in any of the statistical references in appendix 1. Figure 3 also shows the results of a linear regression forecast for Clinic 3. A regression line is the most reliable projection of the trend of any series of data that can be represented by a straight line. When computer software or a calculator with regression capability is avail­ able, this technique should be used instead of the mathematically simpler techniques shown in the preceding examples. 2.7. Extrapolation When the Data Show Non-Linear Trends Time series do not always exhibit a straight-line trend. When this is the case, it is not appropriate to attempt a projection by any of the procedures described so far. Many phe­ nomena are represented more accurately by curved lines, and must be analyzed using more sophisticated statistical techniques. Such analyses are mathematically complex. This handbook presents a single manual procedure that may be used if computers are not available. The reader requiring more sophisticated techniques should review any of the statistical texts listed in appendix 1. Figure 4 is a graph of the data from table 2 for the quantity of IUDs dispensed by Clinic 4. This time series presents a different problem from the cases discussed earlier. When the values of the series are plotted in the same way as the previous examples, they produce a curved trajectory, with a peak in the months of May and June and lower levels at the beginning and end of the year. If this curved line pattern is repeated over several consecu­ tive years, the forecaster should suspect that consumption of IUDs in this location is affected by seasonal variations—factors related to the different seasons of the year. If this is the case, a similar, although not necessarily identical, curved line trajectory will probably occur in the following year. None of the above procedures will directly forecast this curved trajectory. In cases like this, automated techniques ranging from the very simple to the very complex are the best techniques for forecasting. Lacking these, it may be possible to make a satisfactory estimate manually, using a variation of the procedure of semi-averages. In the example of Clinic 4 in table 2, it is possible to calculate four quarterly averages using the simple average formula. For Clinic 4’s 1999 data, these would be 13.3, 26.3, 23, and 12.3. These quarterly averages are then represented by dots placed on the central month of each quarter (February, May, August, and November). The four dots are joined by lines, giving a trajectory which represents the trend of the preceding year better than a straight regression line, as figure 4 shows. 17 The Forecasting Handbook Figure 4. Quantity of IUDs Consumed by Clinic 4 in 1999 and Forecasts for 2000 40 30 20 10 0 Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Feb Apr Jun Aug Oct Dec Feb Apr Jun Aug Oct Dec 1999 2000 Actual Use Semi-Averages Regression Decomposition These quarterly averages could be used directly as the projection for 2000. Note, however, that the trend indicated by the straight regression line shows that consumption is decreas­ ing, although relatively slowly. Actually, between the first and second semesters there was a decrease of about 10 percent. In this case, a more accurate forecast for the coming year could be calculated by decreasing the value of each quarterly average by 10 percent. Such a correction should definitely be made if the increase from one year to the next were greater. Figure 4 also shows this projection. This is an elementary example of decomposition, a statistical term that simply means calculating different aspects of a time series separately. Here, the seasonal variations and the trend were calculated in this fashion. More compli­ cated patterns can be extrapolated using more sophisticated variations of the decomposi­ tion technique, but the assistance of a statistician (or a computer program) is needed to do so. Depending on the purpose and scope of the forecast, a projection that ignores the non­ linear data pattern may still give a satisfactory result. Suppose, for example, that the purpose of the forecast is to estimate an aggregate annual consumption figure for procure­ 18 Chapter 2 ] Extrapolation from Historical Data ment purposes. Rather than plotting monthly data and quarterly averages, as shown in figure 4, the forecaster might plot annual consumption totals for several years and then see if the annual trend can be projected using one of the linear methods. This strategy might produce a sufficiently accurate annual estimate for Clinic 4, though it does not allow the forecaster to estimate monthly or quarterly shipping schedules. 19 The Forecasting Handbook 20 3. ] Corrections for Missing or Erroneous Data Reporting of logistics and service data is rarely complete. There might be missing or incom­ plete reports; consolidated groupings, such as several brands of oral contraceptives being grouped together and reported as pills; or other problems. In some cases, current reports may be missing, but earlier reports may be available. In these cases, the reported quantities must be adjusted to account for the missing and/or consolidated data. Making these adjustments requires good judgment and an understanding of the trend patterns shown by the data that are available. The adjustment techniques described in this chapter are equally applicable to logistics and service data-based fore­ casts, or to any extrapolation made from a historical time series. 3.1. Adjustment for Incomplete Reporting When good-quality data are available, but reports from some facilities are missing, the values can be estimated by increasing the reported quantities by the percentage of missing reports. If the forecaster accepts that quantities were dispensed from the nonreporting outlets at the same rate as from the outlets that did report, the quantities should be increased by using the formula— Estimated Quantity reported used use = during period Proportion of outlets reporting 21 The Forecasting Handbook For example, if 85 percent of outlets reported 850,000 Lo-Femenal orals dispensed last year, then— Estimated 850,000 use = during period 0.85 = 1,000,000 However, it is frequently (though not always) the low-producing outlets that do not report. Moreover, program locations that miss a reporting period might make up for the missing quantities on the next report. For these reasons, the formula above may be too simplistic. For example, program managers may determine (or estimate) that the 85 percent reporting rate represents 90 percent of the contraceptives dispensed. In such cases, this proportion should be used instead of the proportion of outlets reporting— Estimated 850,000 use = during period 0.90 = 944,000 This type of adjustment is not practical in situations where the reporting is very incomplete. Confidence in the accuracy of forecasts decreases as the proportion of facilities reporting decreases. If the level of nonreporting is very high, the program manager should turn to other forecasting techniques described in later chapters. 3.2. Adjustment for Missing Time Periods In some programs, data are reasonably complete for some time periods, but non-existent for others. This situation occurs when facilities report routinely but sometimes neglect to report, or when reports are occasionally lost in transit. It might also occur when there is no service activity to report, either because of stockouts or some other program problem. The first step in such cases is to identify why data are missing and to determine whether it is likely that service activity during the missing time period(s) differed greatly from activity during the periods for which data are available. If significant differences in service activity during the time periods for which data are missing are suspected, corrections must be made based on expert judgment or the missing data must be collected. If great differences in service activity are not suspected, mathematical adjustments to correct for missing periods can be made. The form of such adjustments depends on the trend pattern observed in the data that do exist. 22 Chapter 3 ] Corrections for Missing or Erroneous Data 3.2.1. Where the Trend Is Reasonably Stable The easiest mathematical correction for a missing time period is the simple average of time periods for which data do exist. This is done in the same way as in the extrapolation example of chapter 2— Total quantity used Estimated use in other n periods for each = missing period n If the existing data show a reasonably stable pattern over time, then this technique will work well. Of course, if multiple time periods are missing, the accuracy of the estimate produced using these corrections will be less certain. 3.2.2. Where the Trend Is Upward or Downward If existing data show an increasing or decreasing trend over time, it may be more accurate to correct for a missing period by using the average of the period before and the period after the one for which data are missing— Quantity used in Quantity used in Estimated use previous period + following period for each = missing period 2 Returning to the example of Clinic 3 (shown in table 2), if the forecaster had found June’s data missing, he or she could have estimated consumption as— Estimated Quantity used in + Quantity used in May July use for = June 2 19 + 24 = = 21.5, rounded = 22 2 Note that the actual consumption for June shown in table 2 was 23; thus, this technique produced a reasonable correction in this case. 3.2.3. Where the Trend Shows a Seasonal Pattern If the data that do exist show a seasonal pattern such as Clinic 4 (table 2 and figure 4), then a mathematical correction can be made only if complete data are available for a previous cycle. Such a case is shown in table 3: Clinic 5 had exactly the same seasonal consumption pattern in 1998 as Clinic 4’s pattern for 1999, and a similar seasonal pattern (but with higher consumption rates overall) in 1999. Unfortunately, Clinic 5’s report for May 1999 is missing. 23 The Forecasting Handbook Table 3. IUD Consumption in Clinic 5 in CY1998 and CY1999 Month CY1998 CY1999 January 10 12 February 13 16 March 17 20 April 22 26 May 30 ?????? June 27 32 July 29 35 August 19 23 September 21 25 October 14 17 November 11 13 December 12 14 Total 225 233 + ????? In this case, the forecaster can correct for the missing data point by assuming that the proportion of total consumption for 1999 represented by May 1999 is the same as the proportion of total consumption for 1998 represented by May 1998. The proportion is calculated simply by dividing the figure for the time period in question by the total— Proportion of use Use in single period represented by = a single period total consumption So— Proportion of use 30 represented by = = 0.133 May 1998 225 It is probably reasonable to assume, therefore, that May 1999 represents 0.133 of 1999’s total consumption. Of course, 1999’s total consumption is not known, because May 1999’s data are missing. However, the 1999 total figure can be estimated using the process for incomplete reporting discussed earlier. The formula is— Estimated Quantity reported used total = use Proportion of total use reported 24 Chapter 3 ] Corrections for Missing or Erroneous Data The proportion of 1999’s use that was reported is 0.866 (i.e., everything but the 0.133 assumed to represent May 1999) and the 1999 consumption excluding May’s figure was 233, as shown in table 3. Thus— Estimated 233 total use = = 269.05 for 1999 .866 Finally, this estimated total is multiplied by May’s estimated proportion to arrive at the correction for the missing data point. The formula is— Estimated use Proportion of use for = Estimated total x represented by period missing period use in previous year So— Estimated use for = 269.05 x 0.133 = 35.7, rounded = 36 May 1999 Of course, data for multiple time periods may be missing, in which case there will be no complete data sets to which these formulas can be applied. In such instances, the fore­ caster can go back to the graphical representation of the data and fill in the missing points by eye—or make further field visits to collect the missing reports. 3.3. Adjustment for Both Incomplete Reporting and Missing Time Periods It is also possible that reporting is incomplete and that data for some periods are missing. The above formulas can be applied sequentially to make such corrections. When this is necessary, the correction for incomplete reporting should be made first; then the appropri­ ate formula for correcting for missing data should be applied. 25 The Forecasting Handbook 26 4. ] Estimating Consumption Using Trends in Logistics Data All the preceding examples are projections based on quantities of product dispensed to clients (and assumed to be consumed). In programs where data from the LMIS are complete and of good quality, a logistics data-based forecast can be prepared by simply following the procedures described in chapters 2 and 3. Because these procedures require very few assumptions on the part of the forecaster, logistics data-based projections normally provide the best basis for short- and medium-term forecasting of future product use. Of course, brand new programs do not have historical data of any kind and, so they cannot use this projection method. 4.1. Data Sources and Limitations Historical consumption data are drawn from the program’s LMIS, and may be referred to as distribution, sales, dispensed-to-user, or dispensed-to-client data. Consumption data are also sometimes called issues data, though, as mentioned earlier, logisticians prefer to reserve this term for quantities issued from higher levels of the distribution system to intermediate levels. Regardless of the terminology, reports of quantities dispensed to clients at the lowest level in the distribution system should be used for forecasting wherever possible, because historical trends in consumption are the single best predictor of future consumption. When data from the lowest level are incomplete or incorrect, the forecaster must use distribution data from the lowest level for which there are reasonably complete and accurate data. In practice, it is often necessary to strike a balance between complete­ ness of reporting and nearness to the lowest level. Use great caution, however, in substi­ tuting issues data from a higher level for dispensed-to-client data; if the lowest-level facilities are stockpiling commodities (or letting them expire or be lost), issues data may be completely unrelated to actual consumption. 27 The Forecasting Handbook Some of the advantages and disadvantages of logistics data for forecasting are summarized in table 4. Table 4. Forecasts Using Logistics Data Advantages Disadvantages ‰ Based on the quantity you are trying to predict—consumption. ‰ Requires few assumptions. ‰ Automatically takes distribu­ tion/service delivery constraints into account. ‰ Easy to understand and prepare. ‰ Requires little knowledge of forecasting. ‰ Easy to systematize and institutionalize. ‰ Assumes that the future will be similar to the past. ‰ Assigns equal value to old and new experience. ‰ Incorrect if there have been instances of over- or undersupply (or stockouts) in the past. ‰ Often ignores losses. ‰ May confuse distribution (issues) with consumption (dispensed-to­ client) data. ‰ Does not take into account changing program plans. In general, the data sources for consumption figures include— ˆ The reporting system or LMIS (where contraceptives dispensed to clients are reported directly from outlets on a monthly or quarterly basis). ˆ Receiving records. ˆ Records of contraceptives issued and/or dispensed. ˆ Records used for stock accounting and/or monitoring (e.g., inventory control cards, stores registers). ˆ Financial records, including budgets, records of payment, and others. ˆ Suppliers’ shipping records. ˆ Records of physical inventories. 4.2. Evaluating the Quality of Logistics Data Unfortunately, many programs lack complete and accurate dispensed-to-client data. The initial activity in these cases is evaluation of the available data. The quality of any data source depends on three factors— ˆ Design of the data collection system. ˆ Accuracy of the data. ˆ Completeness of the data. 28 Chapter 4 ] Estimating Consumption Using Trends in Logistics Data The appropriateness of the system design can be evaluated by the following— ˆ Determining if all the data required (in this case, consumption or dispensed-to­ client data) are collected and reported by the system. ˆ Assessing the difficulty of data collection, data entry, and processing. ˆ Assessing the completeness and format of the output. ˆ Determining if the instructions for operating and maintaining the system are clear and complete. ˆ Determining if “nice to know” but unnecessary data overburden the system. ˆ For automated systems, determining whether data are collected, entered, and processed in a timely manner. If consumption data are not gathered or are gathered by a system so flawed that its output is clearly unreliable, the program’s routine reporting system cannot be used to prepare the forecast. The accuracy of MIS or LMIS reports should be checked in a representative sample of facilities by comparing reports from higher levels with local records. Even if done in only a small sample of sites, such verification must be completed before the forecaster can confi­ dently prepare a logistics data-based forecast. It may be that some sites maintain no records, or that their records are inaccurate. For example, losses and/or contraceptives borrowed from another outlet might not be recorded or reported. The verification effort will reveal such problems. If the MIS is automated, the accuracy of the data entered into the computer system must also be checked by comparing data reported at higher levels with corresponding data recorded at lower levels. Completeness of the data can be verified by the following— ˆ Determining whether reporting is up-to-date. ˆ Counting the number of reports submitted and comparing this number with the number required. ˆ Determining if all required data are contained in each report. Appendix 2 contains more complete guidelines for assessing an LMIS. Common problems with LMIS data, with possible solutions, are summarized in table 5. 29 The Forecasting Handbook Table 5. Logistics Data: Problems and Solutions Problems Typical Solutions Issues data versus dispensed-to-client data ‰ Use data from lowest level available. ‰ Beware of double counting. ‰ Check inventory control and data collection processes. Incomplete data ‰ Compensate through extrapolation and interpolation. ‰ Adjust for factors such as volume and seasonality. Timeliness Same as for incomplete data. Data quality/reliability ‰ Are data recording and reporting procedures understood and followed? ‰ Crosscheck— ƒ dispensed-to-client records with stock records; ƒ stock records with actual stock levels; ƒ records from different levels on movement of same stock. ‰ Validate with other forecasts. Incorrect data collected ‰ Collect correct data. ‰ In cases where records/reports are kept by method, not brand: ƒ limit number of brands to one; ƒ do a clinic survey to determine brand mix. Stockouts and stock imbalances ‰ Compensate through extrapolation or interpolation, if not severe. ‰ If severe or prolonged, don’t use. Losses not distinguished from consumption Reduce consumption by estimated losses. 4.3. Correcting for Flawed Logistics Data At the conclusion of these investigations, the forecaster must decide whether the quality of available data is sufficient to proceed with a logistics data-based forecast. A number of techniques beyond those discussed in chapters 2 and 3 can be used to compensate for common flaws in logistics data, as discussed below. 4.3.1. When Consumption Data Do Not Exist The scarcity of reliable stock balance and consumption data at the service delivery level of many programs has led logisticians to seek substitutes or surrogates for consumption data. Unfortunately, there are no adequate substitutes. However, because commodity planning 30 Chapter 4 ] Estimating Consumption Using Trends in Logistics Data must proceed (usually under severe time constraints), issues data are often used as a proxy for consumption. For example, it is common to assume that everything the central medical store has issued to the districts during a particular time period has been dispensed, and that this figure can be taken to represent the level of total consumption. In a perfectly operating maximum-minimum (max-min) inventory control system, stock is issued on a replacement basis, and the quantities issued from any program level closely approximate consumption.1 However, a max-min system can only operate perfectly if its information system operates perfectly, because the control procedures are based on con­ sumption data. Thus, where consumption data are not available, it must be presumed that the inventory control procedures do not work well. In such cases, using issues data as a surrogate for consumption data will only perpetuate current and past stock imbalances and forecasting errors. The higher the level in the distribution system from which issues data are used, the greater the possibility of grievous error. In these situations, it is usually still possible to prepare a forecast using logistics data. In small countries, site visits can be made to most or all higher-level facilities and a sample of service delivery points (including those with the largest volume of contraceptive distribu­ tion or, perhaps, all of them). In larger countries, only a sample of outlets and higher-level facilities can be visited. In either case, the review should follow one or more distribution channels from top to bottom (or bottom to top). At each level, physical counts of stock on hand should be taken and compared to inventory records, receipts data should be cross­ checked with issues data from the level above, and issues data should be crosschecked with receipts data from the level below. These data can be used to prepare a reasonably good estimate of consumption. In some countries where information systems are quite inefficient, such a review is con­ ducted annually, not only to collect information on service delivery and contraceptive supply, but also to supervise, evaluate, and gradually improve the information system.2 This special effort is feasible in most situations, and is even more important when the quantities of contraceptives donated or distributed are large. Logistics data gathered in this fashion can be complemented with small client surveys, taken to determine how much those clients normally purchase (or consume) in a given period of time—for example, one month or one week. Such data can be used to estimate annual or quarterly consumption. If these mini-surveys are conducted annually, comparisons can be made against the results obtained in earlier years to establish consumption trends. 1 For a full discussion of maximum/minimum inventory control systems, see JSI/DELIVER. 2004. The Logistics Handbook: A Practical Guide for Supply Chain Managers in Family Planning and Health Programs. Arlington, Va.: JSI/DELIVER, for the U.S. Agency for International Development. 2 Procedures for taking these physical inventories can be found in JSI/DELIVER. 2004. The Logistics Handbook: A Practical Guide for Supply Chain Managers in Family Planning and Health Programs. Arlington, Va.: JSI/DELIVER, for the U.S. Agency for International Development. 31 The Forecasting Handbook All of these are very simple techniques, within the reach of any country or organization. In all countries, even the least developed, there are firms that specialize in marketing, adver­ tising, and survey-based market studies. It is usually cheaper and more efficient to contract with one of these local agencies to conduct such surveys, rather than to train the staff of family planning organizations or use foreign consultants. 4.3.2. When Consumption Data Are Not Reported If consumption data are not gathered or reported, but other logistics data are, stock records can be used to estimate historical consumption patterns. These records must include receipts, issues, and inventory levels for (at least) the most recent two years, preferably for the lowest level in the distribution system. Ideally, these data would be available for all program locations over a longer period of time. Where these data exist, consumption can be estimated by adding receipts during the year to the beginning inventory and subtracting the ending inventory. The formula is— Estimated use at lowest level = Beginning inventory + Receipts - Ending inventory during period at lowest level at lowest level at lowest level If receipt data from the lowest level are not available, issues data from the next higher level can be used. There are dangers in using this technique. First, the formula makes no distinction between consumption and losses or expiration. This problem is discussed further below. Substituting issues data from the next higher level assumes that no product is lost in transit or storage, which may also be untrue. When lower-level logistics data are not available, it is common to apply this technique with data from the central store and, perhaps, the next level down. This is even more dangerous, because the above problems may exist at every level, thus compounding errors of estima­ tion. Worse, inventory fluctuations at lower levels are ignored by this technique. For example, applying the formula at the district level corresponds to the assumptions that district-level issues equal service delivery-level consumption, and that lower-level inventory control systems are completely functional. These assumptions are questionable at best, and should be verified by site visits to selected outlets. It may be that commodities are simply piling up at the lowest level unused, or worse, that the outlets have used everything and have run out of stock. It is preferable, in such cases, to take a physical inventory at a sample of outlets, as described earlier. In any event, field inventories collected during this investigation should be recorded and preserved, because they will be needed when prepar­ ing estimates of quantities to be procured. 4.3.3. When Consumption and Losses Are Not Distinguished Even where consumption data are reported from the lowest level to the highest, there may be no data that indicate the quantities of a product that were used and the quantities that were lost, making it necessary to assume that everything not still in storage was consumed. 32 Chapter 4 ] Estimating Consumption Using Trends in Logistics Data This assumption overestimates use by the amount of losses that actually occurred, with the result that the next forecast overestimates consumption, perhaps leading to oversupply and additional losses in the future. An essential task for the forecaster is to investigate whether this is happening, and, if so, to reduce reported consumption figures by the estimated amount of losses. Chapter 11 discusses procedures for estimating losses. 4.3.4. When There Were Stockouts It is important to understand that even when logistics records accurately reflect true consumption, they might not reflect true demand. This happens when some contraceptives are out of stock for extended periods. If stockouts were a problem during the period covered by the historical data, it may be possible to adjust the consumption data to reflect true demand using a formula similar to the one used for missing reports. That is— Estimated Quantity reported consumed consumption = during period Proportion of time stocks were available For example, if outlets reported that 850,000 cycles of Lo-Femenal orals were dispensed last year, and it is known that they were stocked out, on average, 25 percent of the time, then— Estimated 850,000 consumption = = 1,133,333 during period 0.75 This calculation assumes a more or less even distribution of stockouts during the period. In adjusting actual consumption data in a time series, it is important to verify this assump­ tion, as discussed in chapters 2 and 3. If, for example, all the outlets were stocked out during the last quarter due to a missed shipment, and if consumption had been rapidly rising until then, this formula underadjusts the consumption figures. It is also important to note that facilities may be effectively stocked out even though inventory records do not show zero stock balances. In undersupply situations, it is common for staff to hoard small (or large) quantities for emergency use or other reasons. The fore­ caster must be alert to this possibility, looking both for actual stockouts and for situations in which stock is not moving. Stockouts are not the only reason that consumption may underestimate (or overestimate) true demand. Other constraints at the service delivery points, as well as the program’s service delivery policies, may significantly influence quantities dispensed to clients. For example, consumption records may under-represent true demand for Depo-Provera® if staff at a significant proportion of facilities are not trained or authorized to provide injections. 33 The Forecasting Handbook 4.4. Completing and Adjusting the Logistics Data-Based Forecast When the necessary corrections to historical consumption figures have been made, the forecaster can complete an initial logistics data-based forecast by direct extrapolation, as described in chapter 2. An example of such a forecast for the fictitious country of Anyland is included as appendix 6. Remember, however, that extrapolation by definition assumes that the future will repeat the patterns of the past. This assumption is not always valid. Future program plans, such as opening a new clinic or beginning an intensive education and distribution program, may mean that future consumption patterns will be quite different than the past. The forecaster, working with program managers, must take these differences into account when adjusting the logistics data-based forecast. If new clinics are opening, it may be possible to understand their likely growth patterns by examining historical consumption data from old clinics with similar catchment areas. Similarly, consumption patterns around the time of past IEC programs may indicate the likely result of new IEC programs. Where such data exist, they should be used to quantify the anticipated effect of new program plans on the consumption forecast. Perhaps the most difficult adjustments are those required when a new contraceptive method is introduced; such changes may cause decreases in consumption for other methods, rather than increases. In any case, the forecaster should not simply add a fixed percentage to the estimate for each commodity to account for program growth. The “last year plus 10 percent” method of projection is perhaps the most common and least accurate of forecasting methods; programs almost never show such a historical pattern of growth. Increases in the use of some methods are often accompanied by decreases in the use of other methods. Other forecasting methods that more readily reflect such changes in program plans and operating environments are discussed in the following chapters. 34 5. ] Estimating Consumption Based on Service Statistics Consumption projections based on service statistics are made using the extrapolation techniques described in chapter 2. An additional step, multiplying estimated numbers of clients by the estimated quantities of supplies required by each client, is needed to convert the resulting service projections into consumption estimates. This conversion requires additional assumptions that may reduce the accuracy of the service data-based forecast. 5.1. Data Sources and Limitations Service data are taken from regular management reports at service delivery sites. In some programs, such data are not available. Other programs may measure many aspects of serv­ ices provided, including demographic characteristics of clients, data on quality of services, and so forth. The most commonly counted service statistics are— ˆ New acceptors or new clients. The number of persons visiting a program and accepting a method for the first time. (There are many variations of this indicator, including clients new to modern contraception, clients new to the particular facil­ ity, and clients new to a particular contraceptive method.3) ˆ Revisits. The number of repeat visits made by all clients during a particular time period. ˆ Users or current users. The number of individuals served by the program who are using a particular method at a particular point in time, whether or not they have actually made a visit during the reporting period. For details, see the USAID EVALUATION Project’s Handbook of Indicators for Family Planning Program Evaluation (reference in appendix 1). 35 3 The Forecasting Handbook Note that these data items measure quite different things. Service statistics systems that gather visit data count how many visits are made to their facilities; programs that collect continuing user data count how many clients they consider to be active, regardless of how many times each client has visited. New acceptors or new clients are the number of unique individuals new to family planning (or new to the specific program, clinic, or method, depending on the definition being used) during the reporting period. The sum of new acceptors plus revisits equals the total number of visits made during the time period. If a single individual made three visits, that person is counted three times. Measurements of users or current users, on the other hand, count this individual only once. Accurate counts of users or current users are very difficult to obtain except in programs with sophisticated computer systems that track individual clients through time. The trend today is to encour­ age programs to track numbers of visits rather than numbers of clients because of the difficulties associated with tracking individuals who may change clinics, switch from a public to a nongovernmental organization (NGO) provider, or change names due to marriage. If new acceptor or new client and revisit data are available, they can be used to forecast commodity requirements. Some advantages and disadvantages of service statistics for forecasting purposes are shown in table 6. The most important limitation of these data for commodity forecasting is that a specific assumption must be made regarding the amounts dispensed at each client visit. Table 6. Forecasts Using Service Statistics Advantages Disadvantages ‰ Automatically takes distribution and service delivery constraints into account. ‰ Easy to modify forecast to account for program service targets. ‰ Focuses discussion on client services. ‰ Easy to understand and not difficult to prepare. ‰ Requires little knowledge of forecasting. ‰ Easy to systematize and institutionalize. ‰ Requires critical assumption regarding dispensing protocols. ‰ Assumes that the future will be similar to the past. ‰ Assigns equal value to old and new experience. ‰ Incorrect if services have been interrupted for any reason. ‰ Definitions of critical data items (e.g., new client, revisit, new user, user, continuing user) likely to be unclear or applied inconsistently. ‰ Brand-specific data (or even method- specific data) may not be gathered. ‰ Produces unrealistic forecast if service targets are unrealistic. Basic service statistics data should be available from an organization’s MIS, and dispensing policies should be found in the organization’s policies and procedures or training materials. Unfortunately, service statistics reporting systems are frequently very weak, and dispensing practices may not be standardized (or standards may not be enforced). 36 Chapter 5 ] Estimating Consumption Based on Service Statistics 5.2. Evaluating the Quality of Service Data This data source, like logistics data, may suffer from inaccuracy because of poor reporting system design, nonreporting, and/or inaccurate reporting. Comments in chapter 4 regarding appropriateness, accuracy, and completeness of logistics data apply equally to service statistics. The same processes of information system evaluation and field verification of data should be applied before a decision is made to prepare a service data-based forecast. This evaluation can be completed simultaneously with the logistics information system analysis. In addition to these concerns, there are often problems with service statistics definitions. Different programs, sometimes within the same country, may use different definitions for new acceptors, continuing users, active users, or any of the other measures. Even within a single organization, program managers may redefine the terms without giving proper training to staff who record the data at the clinic level. When this occurs, service data are unreliable, and only marginally useful for forecasting contraceptive needs. Further compli­ cating the process, service statistics may categorize clients by method but not by brand. Knowing the total quantity of oral contraceptives required is not sufficient for forecasting if the program offers four different brands. In such cases, the forecaster is required to deter­ mine the brand mix from other data. Table 7 summarizes common problems with service statistics data sources, and offers common solutions. Any of these difficulties may preclude preparation of a service data-based contraceptive forecast. It is suggested that service data-based forecasts be prepared only in cases where the following criteria are met— ˆ The service statistics reporting system reports data on visits (either in total or broken down by new acceptors or new clients and revisits). ˆ Method mix can be determined, either because visit data are reported by method or brand, or because other data exist that allow the forecaster to reasonably estimate a method mix breakdown to apply to total visits. ˆ Prescribing protocols are reasonably standardized throughout the program, so that commodity use can be calculated from the service data-based forecast, as described below. 5.3. Completing and Adjusting the Service Data-Based Forecast If the decision is made to proceed with a service data-based forecast, then graphs of historical visit data should be prepared, as described in chapter 2, and the appropriate extrapolation technique should be used to make the forecast. If service data reporting is incomplete, the formulas shown in chapter 3 can be used to adjust for missing data. Depending on the prescribing protocols of the program, it may be necessary to make 37 The Forecasting Handbook different consumption calculations for different types of visits, or at least for new acceptors and revisits; reasons for this are discussed at the end of this chapter. Table 7. Service Statistics Data: Problems and Solutions Problems Typical Solutions Definitions ‰ Unclear. ‰ Reporting and recording procedures not understood or followed. ‰ If not defined clearly or correctly collected, don’t use. ‰ Define each statistic clearly. ‰ Ensure that service statistics recording and reporting procedures are understood and followed. Visits ‰ Do clinic personnel know and follow guidelines for number of units to dispense to clients? ‰ If no guidelines, or guidelines not followed uniformly, don’t use. ‰ If guidelines are known and followed, multiply number of visits by standard quantity dispensed to convert to quantities of contraceptives. Users Don’t use for commodity forecasting. Timeliness/incomplete data ‰ Compensate through extrapolation and interpolation. ‰ Adjust for such factors as volume and seasonality. Lack of brand-specific data ‰ Use method-specific data as upper boundary for aggregate brand estimates made by another method (e.g., logistics data-based). ‰ Do a field-based study of brand mix. 5.3.1. Adjusting the Projection Based on Program Plans Some programs frame their plans in terms of service delivery targets—for example, antici­ pated numbers of new acceptors and revisits, either in total or broken down by method. Such plans must be accommodated in the forecasting process, but unrealistic targets should not simply be incorporated into the forecast. Future targets should be analyzed in light of current trends, and discussions should be held with program planners and managers to help them develop realistic targets. A major service that the forecaster can provide in this process is comparing service targets to extrapolations based solely on historical data. Since planning targets are frequently very optimistic, a compromise between such plans and historical reality is usually required. Another frequent error is the assumption that use of all methods will increase uniformly over time—the last year plus 10 percent method of forecasting. As stated in chapter 4, this assumption is almost always wrong. While a new program may initially experience growth in every method, more mature programs usually experience a gradual shift from less effective traditional methods to more effective modern methods, and from temporary modern methods to permanent methods. As this transition occurs, consumption for some methods may actually 38 Chapter 5 ] Estimating Consumption Based on Service Statistics decrease. These trends can also be identified through service data-based forecasts made using actual historical data. 5.3.2. Calculating Commodity Consumption from Visit Projections After the forecasts of service levels are prepared and agreed upon, commodity consumption for the projected service levels can be estimated. Table 8 and figure 5 show sample service statistics data for 1999 and an extrapolation for 2000, which are used to illustrate these calculations. Table 8. CY1999 Service Activity at Clinic 5 and Projections for CY2000 Oral Contraceptives 1999 (Actual) 2000 (Projected) Month New Acceptors Revisits New Acceptors Revisits January 2 10 2 16 February 3 10 2 16 March 2 11 2 17 April 1 11 2 17 May 2 12 2 18 June 3 12 2 18 July 2 13 2 19 August 1 13 2 19 September 2 14 2 20 October 3 14 2 20 November 2 15 2 21 December 1 15 2 21 Most programs have stated policies on the quantities of contraceptives to be dispensed when clients visit an outlet. For example, the policy might be, “Clients who accept oral contraceptives for the first time are to be given one monthly cycle (mc) at the first visit and three mc on subsequent visits, providing there are no contraindications.” If all outlets follow this policy and there are rarely any contraindications, pill use is estimated by adding the number of first (or initial) visits by pill clients to three times the number of subsequent visits (or revisits). That is, for each type of visit and method, the following formula applies— Estimated use Estimated Quantity of product for next = total visits x given at each visit period of this type of this type 39 The Forecasting Handbook Figure 5. CY 1999 Service Activity at Clinic 5 and Forecasts for 2000: Orals Using the above dispensing policy and the data for clinic 5 from table 8— ⎛Estimated ⎞ ⎛Estimated ⎞Estimated use = ⎜ new visits x 1 mc ⎟ + ⎜ revisits x 3 mc ⎟ for 2000 ⎜⎝ for 2000 ⎟⎠ ⎜⎝ for 2000 ⎟⎠ = (24 new visits x 1 mc ) + (222 revisits x 3 mc ) = 690 monthly cycles In this example, commodity usage for initial visits and revisits must be calculated sepa­ rately, because the program policy for contraceptive distribution differs by visit type. If program policy does not distinguish among visit types (“All condom clients are given twelve condoms at each visit.”), only one calculation is required. 40 Chapter 5 ] Estimating Consumption Based on Service Statistics This calculation is accurate when the policy regarding quantities dispensed at each visit is followed faithfully in all or almost all service outlets. However, dispensing policies often are not followed faithfully. In programs where contraceptives are sold, the amount purchased differs from one client to another. Even where commodities are provided free of charge, the norm may vary from time to time, or from one site to another. If there have been commod­ ity shortages, quantities dispensed at each visit may have been reduced. Conversely, if contraceptives are nearing their expiration dates, quantities dispensed at each visit may have been increased, so that the products can be used before they expire. Quantities dispensed at each visit might also be higher if service delivery points are under pressure to increase overall distribution levels. When such difficulties are encountered, the forecaster and program managers must agree on necessary adjustments to the figures used for the quantity of product given at each visit. Of course, the degree of confidence in the final forecast depends on the accuracy of these adjustments. At a minimum, dispensing practices should be investigated explicitly at a sample of service delivery points before these decisions are made. The formulas for estimating consumption presented here can be applied equally well to service targets or to estimated service levels extrapolated from historical data. Thus this forecasting method can be used in new programs that have no historical data on either services or consumption. A complete example of a service data-based forecast for the fictitious country of Anyland is included in appendix 6. 41 The Forecasting Handbook 42 6. ] Estimating Consumption Using Population Data Both of the preceding forecast methodologies use trends in historical data to predict future patterns of contraceptive consumption. A different forecasting technique—population data- based forecasting—uses demographic data from the Demographic and Health Surveys (DHS) and other sources of population and family planning data to estimate future contraceptive demand. Because these population surveys are conducted at odd, infrequent intervals, and because different questions are asked from one survey to the next, there are rarely enough comparable historical data points to apply the extrapolation methodologies discussed in chapter 2 to a population data-based forecast. However, it is possible to prepare a forecast using population data by setting a goal for the total fertility rate (TFR) or contraceptive prevalence rate (CPR) for the ending year of the forecast and determining how many contraceptive users are required to reach this goal. Numbers of contraceptive users are then converted into estimates of consumption using the couple-years of protection (CYP) conversion factors, which are simply the estimated quanti­ ties of contraceptives required to protect a couple from unwanted pregnancy for one year. A major consideration in making a population data-based forecast is that the quality of the forecast depends on the accuracy of the TFR or CPR goal used. Setting an appropriate goal requires familiarity with the individual program and country, and an understanding of historical precedents regarding rates of TFR or CPR change in developing countries. Goal setting is complicated further by the fact that many family planning programs have already established overly optimistic goals. Significant errors in selecting the ending-year goal will cause large errors in the forecast. On the other hand, this methodology has a significant advantage over logistics- and service statistics-based forecasts because it does not require historical program data. Consequently, population data-based forecasts are particularly appropriate when historical service or logistics data are unavailable or inaccurate, and for new programs that lack historical data. 43 The Forecasting Handbook However, the assumptions underlying population data-based forecasts may significantly affect their accuracy. Because they are not based on program performance data, popula­ tion-based forecasts do not take into account limitations of the service delivery or logistics systems. Furthermore, even if a population-based forecast accurately reflects demand by the population at large, important assumptions must be made regarding the portion of that demand that will be fulfilled by a particular program. While population-based forecasts may give an accurate estimate of need, such estimates may not be indicative of quantities clients will actually demand, or of quantities that they will finally consume. These issues require additional assumptions that may further reduce the forecast’s accuracy. For these reasons, population-based projections are more often used for long-range fore­ casting, or for validation of short-term forecasts made by other methods. Table 9 sum­ marizes some of the advantages and disadvantages of population-based forecasts. Table 9. Forecasts Using Population Data Advantages Disadvantages ‰ Independent of existing service delivery system(s). ‰ Does not suffer data limitations of other methods (missing reports, incomplete data, low quality of recording/reporting). ‰ Where survey design is sound and data processing rigorous, may provide more accurate estimates of consumption. ‰ Provides estimate of maximum need or demand. ‰ Usable in new programs with no historical data. ‰ Does not automatically take account of service delivery or logistics system limitations. ‰ Requires critical assumption for client usage rates of each method (CYP factor). ‰ Since contraceptive source data are frequently inadequate, requires critical assumption regarding portion of total demand to be met by program. ‰ Often requires critical assumptions regarding trends in total fertility rate, contraceptive prevalence rate, and method mix, which may be overly optimistic. ‰ Subject to sampling and non- sampling errors (particularly for methods with very low prevalence). ‰ Data rarely broken down by brand. ‰ Survey data on condom use as backup method or for HIV/AIDS prevention frequently inadequate. ‰ Survey data frequently old. ‰ Losses cannot be estimated from survey data alone. 6.1. Manual versus Automated Projections As with the logistics- and service data-based forecasting techniques presented earlier, population data-based forecasting can be done manually. The manual method is presented 44 Chapter 6 ] Estimating Consumption Using Population Data in full here, and is also the basis for the Anyland population data example shown in appen­ dix 6. Because the manual technique requires a very large number of mathematical calcula­ tions, and because an excellent software program (Spectrum/FamPlan) for carrying out the same tasks is available, this chapter concludes with a brief explanation of the Spectrum software. Appendix 3 presents a more in-depth description of Spectrum/FamPlan and the required input data. Data requirements and sources, considerations in evaluation of data quality, and issues in selecting ending-year values for method mix, brand mix, CPR, and source mix are the same for preparing either a manual or a Spectrum/FamPlan forecast. 6.2. Data Requirements and Sources Population and family planning data are the results of survey, census, or operations research studies of a geographical area or specific population group. The key demographic and program data required for contraceptive forecasting are— ˆ Number of women of reproductive age (WRA). Number of women in their reproductive years (15–49). ˆ Percentage of WRA married (MWRA) or in union. An estimate of the percentage of WRA who are potentially at risk of pregnancy. ˆ Contraceptive prevalence rate (CPR). Percentage of the base population (WRA or MWRA) using a contraceptive method, frequently disaggregated by modern versus traditional methods and by individual contraceptive methods. ˆ Method mix. Mix of contraceptive methods used by the population, expressed as the percentage that each method constitutes of all contraceptives used. ˆ Total fertility rate (TFR). Average number of live births a woman would have if she survived to age 49 and had births at the prevailing age-specific rates. ˆ Source mix. Source of supply for contraceptives, as reported in the DHS. This is needed because most prevalence surveys report on all national use, whereas most forecasts are prepared for a specific program, such as a ministry of health (public sector) program. ˆ Population growth rate. Annual rate of population growth, measured as births minus deaths plus migration, or the more commonly available rate of natural increase, which is simply births minus deaths. It should be noted that these rates measure the growth of an entire population and may differ somewhat from the rate of growth for MWRA. These population data are typically available from one of several sources— ˆ Demographic and Health Surveys (DHS). A regular worldwide series of surveys, including such indicators as total fertility rate (TFR), percent of women in union, contraceptive prevalence rate (CPR), source of family planning services, and method mix. DHSs are ultimately the property of the host country (through the 45 The Forecasting Handbook MOH) but are often published jointly with Macro International, Inc., under contract to USAID or the United Nations Population Fund (UNFPA). ˆ Reproductive Health and Family Planning Surveys. A series of national surveys, similar to the DHS, coordinated and published by the U.S. Centers for Disease Control and Prevention (CDC), Division of Reproductive Health. ˆ National censuses. Complete population counts taken by national census bureaus every 10 years. They detail the age and sex structure of a national population and various subpopulations, providing figures for women of reproductive age (WRA) and percent married (MWRA) or in union. ˆ Intercensal surveys. Sample surveys conducted between national censuses also provide data on WRA or MWRA. ˆ Other local and national population or family planning surveys. Additional surveys made for a variety of reasons by national or local governments, foreign donors, or others; these report population, family planning, and/or HIV/AIDS data. ˆ World Population Prospects. Population projections published every other year by the United Nations Department of International Economic and Social Affairs Statistical Office. They provide estimates of future levels of demographic variables, including TFR and WRA. ˆ Levels and Trends of Contraceptive Use. Historical contraceptive prevalence information by country, published every four years by the United Nations Department of International Economic and Social Affairs Statistical Office. This is the essential data source for setting CPR targets. ˆ World Contraceptive Use (wall chart). A summary of CPR trends and other data from Levels and Trends of Contraceptive Use, published annually by the United Nations Department for Economic and Social Information and Policy Analysis, Population Division. ˆ International Data Base. Population projections, by age and sex, for all developing countries, published by the Center for International Research of the U.S. Census Bureau; particularly useful for obtaining yearly estimates of WRA (see www.census.gov). ˆ Population Reference Bureau (PRB) World Population Data Sheet. Current worldwide estimates of family planning and demographic data, including total population, TFR, CPR, and annual rate of natural increase. 6.3. Evaluating the Quality of Population Data Of the above sources, the DHSs and CDC surveys are the most useful for making a population data-based forecast. However, even if recent DHSs are available, it is necessary to use other sources for estimates of WRA and for guidance in setting goals for the ending year of the forecast. Though it is desirable to use local survey data when preparing contraceptive 46 Chapter 6 ] Estimating Consumption Using Population Data estimates, it is essential that the forecaster assess such local surveys for methodological flaws that may make the data unusable. Some countries have none of these surveys, and in other countries the most recent survey is so old or flawed that it cannot be used with confidence. In such cases, a population data- based forecast should not be made. If population data appear serviceable, it still may be necessary to adjust data from older surveys to obtain current estimates. In any case, the forecast must be based on the same population as the surveys from which the input data are drawn. Table 10 summarizes common problems encountered in using population data for forecasting, and offers possible solutions. Table 10. Population Data: Problems and Solutions Problems Typical Solutions Old data If more than five or six years old, don’t use them. Unreliable data due to sampling or non- sampling errors Don’t use them. Lack of program-specific data ‰ Get program-specific data. ‰ Estimate proportion from national data, if possible. ‰ Use national data as upper boundary for forecasts made by other methods. Biases in survey data ‰ Be aware of them. ‰ Percentage of WRA married often underestimates percentage of WRA at risk; may need to adjust up. Lack of brand-specific data ‰ Use method data as upper boundary for aggregate brand estimates made by another method (e.g., logistics data-based). ‰ Do a field-based study of brand mix. CYP conversion factors ‰ Use country-specific CYP factors, whenever possible; if none, use global recommendations. Data do not account for multiple method use (e.g., condoms as backup) ‰ Adjust CYP factor to compensate or, preferably, prepare separate projection for condom use for HIV/AIDS prevention. Overly optimistic TFR targets ‰ Use established guidelines regarding program growth. (continued on next page) 47 The Forecasting Handbook Problems Typical Solutions Partial data (e.g., regional operations research data) ‰ Decide how representative they are of country as a whole; if they are, extrapolate. Survey data do not account for losses ‰ Estimate separately and adjust requirements estimate. 6.4. Steps in Preparing a Population Data-Based Forecast Manually Manual preparation of a population data-based forecast requires conversion of population data into estimates of commodities needed for the various time periods covered by the forecast. In particular, the forecaster must— 1. Gather the necessary demographic and programmatic data (WRA, percent in union, method mix, brand mix, source mix, and CPR) for the beginning year of the fore­ cast from the sources listed previously, adjusting, if necessary, for out-of-date data. 2. Forecast changes in these variables over the time period of the forecast. Interpo­ late between beginning- and ending-year values, and calculate the numbers of users of each method for each year of the forecast. 3. Convert numbers of contraceptive users to quantities of contraceptives required using couple-years of protection (CYP) factors. Each of these steps is discussed in the following sections. 6.5. Gathering and Adjusting Data for the Beginning Year of the Forecast Data for the fictitious country of Anyland are summarized in table 11. Two common difficul­ ties with population data-based forecasting are immediately apparent: no single source contains all the necessary data for the forecast, and the data from different sources are likely to be from different time periods. In this example, it was necessary to use five different sources—the country DHS, the International Data Base of the U.S. Census Bureau, the PRB World Population Data Sheet, local logistics data, and the UN’s Levels and Trends in Contraceptive Use. While the DHS and PRB data are both from 1999, the U.S. Census WRA figure is from 1996, three years out of date. 48 Chapter 6 ] Estimating Consumption Using Population Data Table 11. Population Data for Anyland for 1999 Base Year Forecast Data Item Source Value Beginning (Base) Year: 1999 Ending Year: 2002 Women of reproductive age (WRA) U.S. Census Bureau International Data Base (1996) 4,940,447 Annual rate of population increase PRB World Population Data Sheet (1999) 3% WRA in union DHS (1999) 100% (See following discussion) Contraceptive prevalence rate (CPR) —all methods DHS (1999) 16% Annual CPR increase (percentage) UN Levels and Trends of Contraceptive Use (1998) 1.0 Method mix: Condoms Orals Other DHS (1999) 9.0% 45.4% 45.6% Brand mix (orals): Lo-Femenal Other Anyland logistics MIS (1999) 50% 50% Source mix (all methods): Public sector Other DHS (1999) 65% 35% CYP conversion factors: Condoms Orals USAID Defaults 120 15 6.5.1. Choosing the Base Year for the Projection Adjusting old data to obtain current estimates of population parameters is time-consuming and problematic, potentially requiring the forecaster to make assumptions about trends in many of the variables shown in table 11. Such additional assumptions may introduce significant error into the forecast. To minimize the number of such adjustments, the date of the survey used as the major data source for the projection should be chosen as the base or starting year of the forecast. In table 11, the 1999 DHS is the source of most of the data items, so 1999 should be the first year of the projection, even if 2000 is the first year for which a forecast is actually desired. 49 The Forecasting Handbook 6.5.2. Estimating Women of Reproductive Age for the Base Year Both the U.S. Census Bureau and the United Nations have invested substantial effort in preparing demographically sound estimates of population growth. The U.S. Census Bureau Center for International Research’s International Data Base provides estimates of WRA for all developing countries for each year between 1995 and 2005, and for the year 2010. The UN World Population Prospects also publishes estimates of WRA, divided by country. If possible, one of these two sources should be consulted to obtain a current estimate of WRA. If the forecaster does not have access to these data sources, it is possible to adjust older census enumerations of WRA, using an annual population growth rate to obtain an estimate for the beginning year of the projection. It should be noted that subpopulations (such as WRA) usually have different growth rates from national populations. However, given the inherent imprecision introduced by other assumptions that must be made in preparing the forecast, and the relatively short timeframe of projections made for procurement purposes, this approximation can be used when the forecaster cannot obtain U.S. Census Bureau or UN estimates. The formula is— Estimated Estimated ⎛Estimated Annual rate ⎞ WRA for = WRA for + ⎜ WRA for x of population⎟ year n + 1 year n ⎜⎝ year n increase ⎟⎠ Population growth rates are reported in one of two ways—the annual rate of population increase, which takes into account estimated births, deaths, and effects of migration; or the annual rate of natural increase, which takes into account only births and deaths. The first figure is more appropriate and should be used when it is available. If it is not, then the annual rate of natural increase can be substituted. In table 11, the 1996 WRA figure is 4,940,447, and the annual rate of population increase is 3 percent. If U.S. Census Bureau or UN projections are not available, then the 1999 base year WRA figure would be calculated by using this formula repeatedly— Estimated WRA for = 4,940,447 + (4,940,447 x 0.03) 1997 = 4,940,447 + 148,214 = 5,088,661 Similarly— Estimated WRA for = 5,088,661 + (5,088,661 x 0.03) 1998 = 5,088,661 + 152,660 = 5,241,321 50 Chapter 6 ] Estimating Consumption Using Population Data And finally— Estimated WRA for = 5,241,321 + (5,241,321 x 0.03) 1999 = 5,241,321 + 157,240 = 5,398,561 6.5.3. Estimating the Actual Population at Risk of Pregnancy The data sources discussed earlier will provide estimates of the percentage of WRA currently married and/or the percentage of WRA in union. However, neither of these figures is a good estimate of the number of women at risk of pregnancy. In many societies, substantial numbers of women and men are sexually active without being married or “in union.” Active teenage women, in particular, are often underreported, though more recent surveys in some countries have targeted teenagers specifically. Accordingly, it is frequently necessary to adjust reported figures to give a more accurate estimate of the population actually at risk. Adding the percentage of WRA reported in the DHS as living together to the percentage of WRA currently married yields an estimate of WRA in union; this figure should always be used instead of WRA currently married when data are available. Although WRA in union gives a more accurate picture of the population at risk, it may still seriously underestimate the target population in some countries. When the forecaster has reason to believe that the WRA in union significantly underestimates the actual population at risk of pregnancy, a different strategy can be used to prepare the forecast—all women of reproductive age can be considered at risk, and the CPR for all women, as opposed to currently married women, is then used for the projection. Because it is never true that all women are at risk of pregnancy, this assumption is obviously illogical. Mathematically, however, using the CPR for all women, which is lower than the CPR for WRA currently married, compensates for the assumption that 100 percent of women are at risk, allowing the forecaster to complete the projection without making a potentially incorrect guess about the number actually at risk. 6.5.4. Choosing the Appropriate Contraceptive Prevalence Rate for the Base Year The initial CPR estimate is best obtained from the most recent DHS table, “Current use of contraception, by age.” This table provides CPR figures for both WRA currently married and all women. If the projection is to be made using WRA in union or WRA currently married, the first CPR figure should be used; if the projection is to be based on total WRA, then the second CPR figure should be used. If a survey other than the DHS is used, the same principle should apply. The forecaster must be certain that the base population for WRA and the base population for CPR are the same. 51 The Forecasting Handbook 6.5.5. Calculating the Method Mix Most surveys provide data on contraceptive users by method. If the DHS is used, for exam­ ple, the desired figures are contained in the table, “Current use of contraception, by method.” This table provides data for all women and WRA currently married. As with CPR, the choice of which figure to use depends on which WRA figure has been selected as the basis for the projection. The DHS and most similar surveys present data for each method as a percentage of all women or WRA currently married, including women who are not contracepting. For projec­ tion purposes, the forecaster needs the method mix expressed as percentages relative to all women or WRA currently married who are using any method of contraception. This percent­ age can be obtained by dividing the DHS figure for women using each method by the percentage of women using any method— Method mix Percentage using method for = a method Percentage using any method For example, if all women are being used for the projection, and the survey reveals that 8 percent use the pill and 30 percent use any method, then— Method mix 8 for = = 26.6% orals 30 In other words, orals represent 26.6 percent of overall contraceptive use. This calculation is repeated for all other methods. Where surveys other than the DHS are used, it is likely that this same calculation will be needed. 6.5.6. Estimating the Brand Mix If the forecast is being made for procurement purposes, it is usually not sufficient to know just the number of users of each method. Unless the program provides only a single brand of each method, the forecaster needs a further breakdown into individual contraceptive brands to complete the projection. Very few surveys tabulate data by brand. The most obvious source of information on brand mix within a method is the program’s LMIS. If consumption data or low-level issues data are available, a brand mix can be calculated and used to disaggregate method figures into brands. If there is no functional LMIS, then a representative sample of service sites should be visited, records should be reviewed, and service providers and program managers should be interviewed to determine the brand mix. As with the overall method mix, the desired figure is the percentage that each brand represents of the total use of that method. Of course, for new programs, no such data are available. In these cases, program targets or historical experience of other established programs should be used to make brand estimates. 52 Chapter 6 ] Estimating Consumption Using Population Data 6.5.7. Estimating the Proportion of National Contraceptive Use Attributable to the Program (Source Mix) Most surveys on which population data forecasts are based are national in scope. These data can be used directly to prepare national-level consumption estimates. However, consump­ tion estimates are often needed for individual service delivery programs (i.e., for an NGO program or for the public sector program only). To prepare program-specific estimates, the forecaster must estimate the source mix—the proportion of national contraceptive use attributable to the particular program for which the forecast is being prepared. This proportion may be very different for different methods. For example, an HIV/AIDS prevention program may contribute a significant portion of the national condom CPR, but nothing at all for other contraceptive methods. Thus, the source mix will likely have to be estimated separately for each method. As shown in appendix 3, DHSs include a table called “Source of supply for modern contra­ ceptive methods.” In many cases, this table can be used to estimate the source mix. In other cases, however, the DHS breakdowns may be too general, or the survey may be too old. A special problem in preparing forecasts for social marketing programs is that many survey respondents report the source of their supply as “pharmacy,” making it difficult to distinguish between social marketing and purely private sector activities. When such problems are encountered, the forecaster and program managers should estimate source mix based on local data or program managers’ experience. However, errors in these estimates yield proportional errors in the consumption forecast. If data are lacking and managers are unsure of their estimates, less confidence can be placed in the final forecast. In the Anyland example in table 11, the DHS showed that the public sector accounts for 65 percent of contraceptive use for every method. To prepare a projection for the MOH, the forecaster should use a source mix of 0.65 for each method, assuming the MOH is the only public sector program that provides contraceptives. 6.6. Estimating WRA, CPR, Method Mix, and Source Mix for the Final Forecast Year The above formulas and procedures provide the necessary population parameters for a single year—the base year of the projection. Before consumption estimates can be made, it is necessary to project how these parameters will change over the forecast period. The three population parameters most likely to change significantly over the course of a short- or medium-term forecast are WRA, method mix, and CPR. Consequently, at least these parame­ ters require estimates for the future years of the forecast. The most common technique is to estimate values for these data items for the last year of the projection and then calculate intermediate values using the formulas for linear trend projections shown in chapter 2. In a mature and relatively stable program, it may be reasonable to assume that both method mix and CPR will remain constant for a period of up to four years, so that the linear 53 The Forecasting Handbook trend calculation is needed only for WRA. In a less stable environment, the CPR and method mix must be adjusted as well, and the brand mix may also differ. A similar judgment must be made regarding the source mix. In such situations—and in the case of longer-range forecasts covering more than four years—manual calculations are extremely tedious. In these instances, automated techniques (preferably Spectrum/FamPlan) should be used to make the projection. Regardless of whether a manual or automated projection is made, the forecaster needs to present the estimated future values for WRA, CPR, method mix, source mix, brand mix, and the other parameters in a form similar to that shown in table 12, ensuring that program managers and other concerned individuals (e.g., donors) agree that the figures are reason­ able. Table 12. Population Data for Anyland for Final Forecast Year (2002) Data Item Value Women of reproductive age (WRA) 5,899,153 Annual rate of natural increase 3% WRA in union 100% Annual CPR increase (in percentage points) 1% Target contraceptive prevalence rate (CPR)—all methods 19% Method mix: Condoms Orals Other 9.0% 45.4% 45.6% Brand mix (orals): Lo-Femenal Other 50% 50% Source mix (all methods): Public sector Other 65% 35% CYP conversion factors: Condoms Orals 120 15 6.6.1. Estimating WRA for the Final Forecast Year The WRA figure for the final year of the forecast should be taken from the source used for the base year, again preferably either the U.S. Census Bureau’s International Data Base or the UN’s World Population Prospects. If neither of these sources is available, the formula given above for adjusting older WRA data for the base year can be used repeatedly to calculate WRA figures for the forecast years, though this methodology is much less desirable. 54 Chapter 6 ] Estimating Consumption Using Population Data 6.6.2. Estimating CPR for the Final Forecast Year The CPR estimate for the final forecast year is the crucial assumption in a population data- based forecast. A special danger derives from the fact that programs and governments often set national targets for increases in CPR that are very optimistic. The forecaster is likely to come under pressure to use these in the population data-based forecast. Except for the UN’s World Contraceptive Use wall chart and Levels and Trends of Contraceptive Use there are few generally recognized data sources for historical rates of change in CPR that can be used to confirm realistic goals or refute unrealistic ones. Given the relatively short time period of forecasts made for immediate procurement pur­ poses, major CPR increases or decreases during the forecast period are unlikely. For refer­ ence, appendix 4 contains the most recent data from Levels and Trends of Contraceptive Use. Historical changes in CPR are summarized and categorized in table 13. It is immediately apparent from these figures that the most successful family planning programs in the world have increased contraceptive prevalence by only one or two percent­ age points per year. In countries with lower levels of prevalence and less commitment to family planning, the change is closer to one-half of one percentage point growth per year, ranging all the way down to negative growth. One rational way to set the target CPR for the country or program in question is to check the rates of change in countries with similar programmatic and cultural settings, using the data from table 13 and appendix 4. Another strategy is to prepare the population data-based forecast based on estimates of the trends in the TFR rather than the CPR, using Spectrum/FamPlan to calculate CPR changes as described later in this chapter. A great deal of effort has been invested in the study of TFR trends throughout the world. For example, there are country-specific and average figures for TFR decline by strength of program effort and level of development of the country, as shown in table 14, as well as UN and World Bank estimates of future levels of TFR, as shown in appendix 4. 55 The Forecasting Handbook Table 13. Annual Percentage Change in Contraceptive Prevalence by Level of Family Planning Program Effort (1982–1989) and Socioeconomic Setting (1985) Program Effort, 1982–1989 Socioeconomic Setting, 1985 Strong Moderate Weak Very Weak High Mexico 3.3 Colombia 0.9 Korea Rep. 2.0 Mauritius -0.8 Singapore 1.6 Average 1.4 Jamaica 1.0 Panama 0.5 Trinidad & 0.1 Tobago Average 0.5 Jordan 0.7 Brazil 1.1 Costa Rica 0.8 Average 0.9 Iraq -0.1 Upper Middle Thailand 1.0 Indonesia 1.0 Sri Lanka 1.0 China 1.3 Average 1.1 Tunisia 1.8 Botswana 1.2 Ecuador 1.4 Dominican 1.3 Republic El Salvador 0.6 Egypt 1.6 Philippines 1.0 Malaysia 1.1 Average 1.3 Algeria 1.9 Peru 1.6 Zimbabwe 1.0 Syria 1.1 Iran 5.8 Turkey 1.2 Guatemala 0.5 Paraguay 1.3 Average 1.8 Lower Middle India 0.5 Morocco 2.2 Vietnam 1.9 Average 2.0 Honduras 1.5 Kenya 1.7 Zambia 2.1 Tanzania 1.7 Pakistan 1.0 Haiti 1.0 Cameroon 1.0 Nigeria 0.1 Lesotho 1.3 Ghana 0.8 Average 1.2 Bolivia 1.8 Cote d’Ivoire 0.6 Average 1.2 Low Bangladesh 2.3 Nepal 1.4 Average 1.8 Rwanda 1.3 Senegal 0.1 Mali 0.2 Uganda 1.5 Average 0.8 Mauritania 0.3 Sudan (North) 0.3 Malawi 1.2 Benin 0.5 Average 0.6 Source: Contraceptive prevalence rate increases are based on data from the UN Department for Economic and Social Affairs, Population Division, World Contraceptive Use 1998. The format of table 13 and the categorization of countries is taken from W. Parker Mauldin and John Ross, unpublished analysis (see Spectrum/FamPlan manual). 56 Chapter 6 ] Estimating Consumption Using Population Data Table 14. Declines in TFR from 1975 to 1990 by Level of Program Effort (1982–1989) and Socioeconomic Setting (1985) Program Effort, 1982–1989 Socioeconomic Setting, 1985 Strong Moderate Weak Very Weak High Mexico 1.7 Taiwan 1.5 Colombia 1.3 Korea Rep. 1.3 Mauritius 0.7 Singapore 0.3 Average 1.1 Jamaica 1.7 Korea PDR 1.4 Panama 1.1 Trinidad & 0.8 Tobago Cuba 0.6 Chile 0.5 Average 1.0 Jordan 1.5 Brazil 1.2 Lebanon 1.1 Venezuela 1.0 Costa Rica 0.7 Average 1.1 Kuwait 2.4 Iraq 0.7 Average 1.5 Upper Middle Thailand 1.8 Indonesia 1.5 Sri Lanka 1.2 China 1.1 Average 1.4 Tunisia 2.0 Botswana 1.8 Ecuador 1.6 Dominican 1.5 Republic El Salvador 1.3 Egypt 1.1 Philippines 0.9 Malaysia 0.6 Average 1.4 Algeria 2.6 Peru 1.7 Zimbabwe 1.4 Guyana 1.3 Syria 1.2 Iran 1.0 Turkey 1.0 Guatemala 0.8 Paraguay 0.6 Congo 0.0 Average 1.2 Libya 0.8 Saudi Arabia 0.7 Average 0.7 Lower Middle India 1.0 Morocco 2.0 Vietnam 1.4 Average 1.7 Honduras 1.5 Kenya 1.4 Zambia 0.8 Tanzania 0.7 Papua New Guinea 0.6 Pakistan 0.5 Haiti 0.5 Cameroon 0.5 Nigeria 0.5 Lesotho 0.4 Ghana 0.4 Madagascar 0.3 CAR 0.2 Average 0.6 Bolivia 1.2 Myanmar 1.0 Liberia 0.0 Cote d’Ivoire 0.0 Lao PDR -0.1 Congo -0.2 Cambodia -0.6 Average 0.2 Low Bangladesh 2.0 Nepal 0.8 Average 1.4 Rwanda 1.7 Senegal 0.6 Afghanistan 0.2 Mali 0.0 Guinea 0.0 Burundi 0.0 Togo 0.0 Mozambique 0.0 Sierra Leone 0.0 Burkina Faso 0.0 Guinea Bissau -0.2 Uganda -0.4 Niger -0.5 Average 0.1 Mauritania 0.9 Sudan 0.7 Malawi 0.2 Chad 0.0 Somalia 0.0 Benin 0.0 Ethiopia -0.2 Average 0.3 Source: W. Parker Mauldin and John Ross, unpublished analysis (See Spectrum/FamPlan manual). 57 The Forecasting Handbook In any case, it is essential that forecasts be prepared using multiple data sources, as discussed in chapter 1. This strategy will highlight excessively optimistic (or excessively pessimistic) assumptions for CPR and other data items. 6.6.3. Estimating Method and Brand Mix for the Final Forecast Year Unfortunately, few comprehensive studies on trends in method mix have been completed, and most of the rules of thumb on method mix changes apply to long-term changes as a program moves from its initial launch to a more mature stage. This empirical record clearly demonstrates the shift over time from traditional to resupply to more permanent methods, but provides little guidance in estimating short-term method mix changes. It is best to be conservative when estimating change in method mix over a four- or five-year period. Although method mix can be affected immediately by stockouts of a method, major shifts in the overall method mix have usually progressed more slowly. Without an aggressive program to introduce or expand the use of specific methods (backed by training for service providers and an IEC campaign to orient clients), it is unlikely that there will be significant changes in the method mix during the forecast period. Where proper launch activities were not undertaken, attempts to change even the brand mix for a single method have been unsuccessful. On the other hand, demand for some methods can increase rapidly with only word-of-mouth promotion among clients. This has happened for injectables and Norplant® in countries as different as Tanzania and Haiti. The forecaster should review all program plans to launch or expand the use of particular methods, as well as program budgets for IEC, service delivery training, procurement, and distribution. If historical logistics or service statistics data (or multiple population surveys) are avail­ able, these should be studied to discern trends in method mix. In the absence of quantita­ tive data, knowledgeable service providers should be interviewed, and their best estimates of trends evaluated for reasonableness. If neither of these approaches yields a satisfactory result, it may be best to assume no change in method mix over the short-term forecast period. These same comments apply to estimation of brand mix. Where these decisions must be made without hard data, the program’s information systems should be strengthened quickly. 6.6.4. Estimating the Proportion of National Contraceptive Use Attributable to the Program (Source Mix) for the Final Forecast Year In the rare cases where two or more successive DHSs are available, and the DHS table “Source of supply for modern contraceptive methods” is sufficiently detailed, the extrapola­ tion techniques shown in chapter 2 can be used to estimate changes in the source mix. Frequently, however, no hard data are available, and the forecaster and program managers will have to use their best judgment in estimating changes in the source mix. As with the other parameters, it is best to be conservative. Unless specific program interventions aimed 58 Chapter 6 ] Estimating Consumption Using Population Data at changing the market share of individual programs are planned, these percentages are likely to remain relatively constant over the time period of a short-term forecast. 6.7. Estimating Changes in WRA, CPR, Method Mix, and Source Mix over the Forecast Period Once agreement has been reached on the population parameters for the base year and the final forecast year, values must be computed for each intermediate forecast year. For a base year of 1999 and a final forecast year of 2002, for example, values must be computed for 2000 and 2001. 6.7.1. Estimating Intermediate Values for WRA The U.S. Bureau of the Census’s International Data Base and the UN’s World Population Prospects provide annual WRA estimates; as discussed above, these should be used where possible. If it is impossible to obtain access to either of these sources, the formula for adjusting WRA data given above can be used. The formula is— Estimated Estimated ⎛Estimated Annual rate ⎞ WRA for = WRA for + ⎜ WRA for x of population⎟ year n + 1 year n ⎜⎝ year n increase ⎟⎠ Returning to the Anyland example, WRA for 1999 had been estimated to be 5,398,561, and the annual rate of population increase is 3 percent. The estimate for 2000 would be— Estimated WRA for = 5,398,561 + (5,398,561 x 0.03) 2000 = 5,560,518 The 2001 estimate is calculated the same way. 6.7.2. Estimating Intermediate Values for CPR, Method Mix, and the Source Mix Unless there is good reason to think otherwise, the forecaster should assume that the year- to-year change for each of these parameters is linear, and simply interpolate a line between the first and last values. Interpolation is really the same procedure as the extrapolation technique using linear trends described in chapter 2, except that the points being esti­ mated are between the first and last point, rather than beyond the last point (hence interpolation instead of extrapolation). The formulas are— Averagechange Target value in final period - Value inbase year over = forecast period Number of years in forecast 59 The Forecasting Handbook And— Estimate for Estimate for Average change period n + 1 = period n + over forecast period In the Anyland example, the CPR in 1999 was 16 percent, and the target CPR for 2002 is 19 percent. So— Average change 19% - 16% over = = 1.0% forecast period 3 And— Estimate for = 16% + 1.0% = 17%2000 Applying the formula again gives a CPR estimate of 18 percent for 2001. Method mix changes for the interim time periods are calculated similarly. If changes in the brand mix within one or more methods are expected during the time period of the forecast, the same procedure can be used to calculate brand mix values for each interim period. Note that the sum of the method mix percentages for each time period must equal 100 percent, as must the brand mix percentages for each method. It may be necessary to round individ­ ual interpolated figures to reach 100 percent. Finally, if significant changes in the source mix are expected during the forecast period, this same interpolation technique should be used to estimate the intermediate values. 6.8. Calculating Commodity Consumption for Future Time Periods Once the forecaster and program planners have come to agreement on the above data items and projections, the consumption forecast for each method and brand can be completed. 60 Chapter 6 ] Estimating Consumption Using Population Data 6.8.1. General Calculation for Population Data Forecasts The general formula for this calculation is— Estimated consumption Estimated CPR for Method mix Brand mix Source mix of a method of a brand = year WRA for n x Year n for year for this methodx for year for this brandx for year for this methodx nnn in year n CYP conversionx factor Although at first glance this formula appears complex, it is straightforward. Multiplying estimated WRA times the CPR, the first two factors, simply gives the total number of women at risk of pregnancy who are estimated to be contracepting. (Again, remember to be consis­ tent in using all women or WRA in union for these factors.) Multiplying this result by the method mix and brand mix percentages, the third and fourth factors, gives the number of users being protected by a particular commodity. Multiplying further by the source mix, the fifth factor, provides an estimate of the number of those users being protected by the program. With this figure in hand, it is necessary to estimate the quantity of commodities needed to protect those women throughout each time period. This is done using the last factor in the equation, the CYP conversion factor. 6.8.2. Using Couple-Years of Protection Conversion Factors to Estimate Consumption for Short-Term Contraceptive Methods The quantity of commodities required to protect each user is normally estimated for a one- year period. Again, this quantity is called the couple-years of protection (CYP) conversion factor—the amount of a particular contraceptive needed to provide one couple with protec­ tion for one year. Table 15 shows the CYP standards established by USAID’s EVALUATION Project. Although both the CYP concept and CYP conversion factors have been used for program evaluation for decades, there is still controversy regarding application of the factors for certain purposes. In particular, establishing the exact quantity of a given contra­ ceptive required to protect one couple for one year from an unwanted pregnancy is prob­ lematic for many contraceptive methods. Use of the standard factors from table 15 is likely to be acceptable for orals and injectables, because the CYP factor is very closely associated with the menstrual cycle, and there is little variation from woman to woman or country to country. 61 The Forecasting Handbook Table 15. Couple-years of Protection Conversion Factors4 Method CYP Oral contraceptives 15 cycles/CYP Condoms 120 pieces/CYP CuT 380A IUDs 3.5 CYP/insertion Injectables Depo-Provera® Noristerat Cyclofem 4 doses/CYP 6 doses/CYP 12 doses/CYP Vaginal foaming tablets 120 tablets/CYP Norplant® 3.5 CYP/implant For barrier methods, however, there is great variability from person to person and place to place, and few hard data on which to base CYP factors. The standard for condoms and foaming tablets assumes 120 acts of vaginal, protected intercourse per year, and that the two methods are not used together. It further assumes an unspecified level of client wast­ age. These are difficult assumptions to prove or disprove. It is known that the frequency of sexual relations varies from culture to culture and individual to individual. Moreover, many users combine condoms or spermicides with the rhythm method or withdrawal, or with sexual techniques other than vaginal coitus. Condoms used to prevent HIV/AIDS or other sexually transmitted diseases are often estimated separately. Moreover, more than one method may be used simultaneously, either for backup or, in the case of condoms, for disease prevention. The standard CYP factors do not automatically account for such varia­ tions. For these reasons, the authors of the CYP methodology indicated explicitly the need to conduct country-specific user surveys to establish CYP factors. Use of local data for conver­ sion factors can improve the quality of the forecast for barrier methods. Unfortunately, such data are rarely available, so the forecaster normally has to use the standard factors from table 15. In the Anyland example of table 11, for example, the calculation for condoms for the public sector for the 1999 base year (presuming there is only one brand of condoms in the program) is— Stover, John, Jane T. Bertrand, Susan Smith, Naomi Rutenberg, and Kimberly Meyer-Ramirez. 1997. Empirically Based Conversion Factors for Calculating Couple-Years of Protection. Chapel Hill: The EVALUATION Project. Carolina Population Center, Tulane University, and The Futures Group International. 62 4 Chapter 6 ] Estimating Consumption Using Population Data Estimated of condoms consumption = 5,398,561 x 0.16 x 0.09 x 1 x 0.65 in 1999 x 120 = 6,063,664 Note that this calculation uses the adjusted WRA estimate for 1999, not the 1996 survey figure shown in table 11. 6.8.3. Using CYP Factors for Estimating Consumption of Long-Term Contraceptive Methods CYP conversion factors are even more problematic in forecasting for longer-acting methods (IUD, implants), because these methods provide protection that extends beyond the time period of the forecast. This means that not all of the women being protected by these methods in a particular year need a device that year—many of them will be protected by IUDs or implants received in previous time periods. The CYP factors for IUDs and implants take into account the extended use life of these methods, with an allowance for discon­ tinuation. Mathematically, the inverse of the CYP factor for these methods gives the portion of the program’s users who need devices in a given year. If no local data are available, the fore­ caster can just use the inverse of the factors for IUDs and Norplant from table 15 in the general formula for estimated consumption to obtain quantity estimates for these methods. If Norplant represents 2.0 percent of Anyland’s method mix, for example, the required quantity for the public sector is— Estimated consumption of Norplant = 5,398,561 x 0.16 x 0.02 x 1 x 0.65 in 1993 1 x 3.5 = 3,208 63 The Forecasting Handbook For IUDs, a more refined estimate is sometimes possible. It is generally true that women continue using the first device for many years or discontinue relatively quickly, switching to another method or discontinuing contraception. For commodity forecasting purposes, the simplest and most accurate assumption for IUDs is that a single device is required for each new user. If program estimates of new users can be obtained or calculated for the forecast period using the extrapolation techniques described in chapter 2, IUD quantities for each time period of the forecast should be estimated by the following formula, rather than the general formula for estimated consumption— Estimated consumption Estimated number of new usersof a brand = of the brandof IUD in year n in year n A complete population data-based forecast for the fictitious country of Anyland, illustrating all the manual techniques described here, is included in appendix 6. 6.9. Using Spectrum/FamPlan for Contraceptive Forecasting Because population-based forecasts require so many demographic and family planning parameters, and because the calculations are long and complicated, it is much easier to use an automated tool to complete the forecast. Spectrum/FamPlan, developed by The Futures Group International (TFGI), is the most appropriate tool for this purpose. Appendix 3 describes Spectrum in more depth, and shows the data sources for the most important inputs to the model. Complete documentation of Spectrum/FamPlan, available through TFGI, provides step-by-step instructions for installation and operation of the program. The primary advantages of Spectrum are ease and accuracy. The model requires a number of additional parameters beyond those presented in this chapter, but the program includes a database of many of the most important variables, disaggregated by country and year, which can be selected as defaults. For example, once the country and forecasting years are selected, the software presents the user with complete estimates of WRA, by year; TFR for the initial year; and high, medium, and low assumptions for the ending year. Spectrum/ FamPlan allows the user to make changes to the default values and then calculates the impact of the changes automatically, freeing the forecaster from hours of manual calcula­ tions. This allows the forecaster to focus on verifying the accuracy of the input data. Because the software performs all the forecasting calculations (using formulas similar to the manual procedures discussed above), potential math errors are eliminated. Moreover, Spectrum takes into account the effects of mortality on the base population and the impact of the proximate determinants of fertility on the numbers of contraceptive users required to achieve the TFR goal. These variables have a major impact on the ultimate accuracy of the forecast, but are too complex to be included in manual calculations. 64 Chapter 6 ] Estimating Consumption Using Population Data 6.10. Steps in Preparing a Population Data-Based Forecast Using Spectrum/FamPlan The general steps in preparing a Spectrum/FamPlan forecast are as follows— 1. Gather the data required for the beginning year of the forecast (WRA, percent in union, TFR, CPR, method mix, source mix, sterility, total abortion rate, and post­ partum insusceptibility) from the various sources listed above. 2. Validate the data by checking them against additional data sources where possible, and then input them into Spectrum/FamPlan for the base year. Even when default values of the variable are available in the program (e.g., WRA, TFR), it is important to check the defaults and make changes as required. 3. Enter a goal for TFR in the final year of the forecast. This goal should be based on program plans and trends, estimates of future TFR from the data sources listed earlier, and knowledge of historical trends in TFR decline shown in table 14. 4. Input ending year values for any of the other variables that are likely to change over the period of the forecast. 5. Run the model. Spectrum/FamPlan calculates the level of contraceptive use that resulted in the base year TFR, given the other demographic and program charac­ teristics entered. It then calculates the number of users and the quantities of contraceptives these users require to achieve the TFR goal in the ending year of the forecast. 6. Use the software to produce output tables on contraceptive commodity needs, by year, for the period of the forecast. 7. Divide the quantities of contraceptives estimated by the software for each method into quantities for each brand using the techniques described earlier in this chapter. (Future versions of the software are expected to calculate brand-level data automatically.) 6.11. Gathering and Adjusting Data for the Spectrum/FamPlan Base Year The data requirements for making a population data forecast with Spectrum/FamPlan are detailed in appendix 3, which includes an annotated example taken from actual source data for Kenya. Except for the additional proximate determinants of fertility, these inputs are the same data that were used for making the manual population data-based forecast described earlier in this chapter. With Spectrum, the forecaster is not required to adjust the figures for WRA, or to calculate the source mix or method mix by hand; the software performs this math. 65 The Forecasting Handbook The forecaster is required to collect the best data available and to input them accurately into the program. Of the four required proximate determinants, two (postpartum insuscepti­ bility, percentage of WRA in union) are presented clearly in DHS tables. Another (sterility) can be estimated by proxy from the DHS (see appendix 3). The fourth (abortion) is often entered as the default (0) due to lack of credible data. 6.12. Estimating Inputs for the Final Year of the Spectrum Forecast Spectrum/FamPlan requires the user to enter values for all inputs for the ending year of the projection, and then interpolates between the beginning and ending-year values to produce values for the intermediate years. Considerations in selecting the ending values of the most important variables are described in depth earlier in this chapter. Spectrum/FamPlan also allows ending-year values to be entered for marital status, age at first use for sterilization, and abortion. Though these are unlikely to change significantly over the time period of a short- or medium-term forecast, the forecaster should also be aware of trends in these parameters. One parameter that will obviously change during the period of the forecast is the popula­ tion of women of reproductive age. Fortunately, the EasyProj component of Spectrum auto­ matically calculates future estimates of WRA based on assumptions of fertility, mortality, and migration. The TFR estimate for the final forecast year is the crucial assumption in the automated version of population data-based forecasting. A major advantage of using Spectrum is that the forecaster can set TFR targets, rather than using CPR, as the manual technique requires. There is a good record of historical changes in TFR throughout the world, and there are several credible databases of TFR projections by country and year. Thus, TFR targets are easier to set and defend than CPR targets. The EasyProj component of Spectrum/FamPlan will automatically produce three scenarios (high, medium, and low variants) for ending-year TFR, using data developed by the UN and reported in World Population Prospects. The forecaster might simply choose the medium variant TFR for the projection, but—as with CPR targets—all data sources, as well as the official program goals, should be evaluated care­ fully before a TFR target is selected. 6.13. Completing the Spectrum/FamPlan Forecast Once all these data are verified and entered, Spectrum/FamPlan will produce a wide range of reports. Table 16 is a Spectrum/FamPlan “Commodities by Method” report taken from the Kenya example of appendix 3. As mentioned above, future versions of Spectrum/FamPlan will disaggregate these data by brand. 66 Chapter 6 ] Estimating Consumption Using Population Data Table 16. Spectrum/FamPlan, Commodities by Method 67 The Forecasting Handbook 68 7. ] Estimating Consumption Based on Distribution System Capacity New family planning or HIV/AIDS prevention programs (and less well-managed older pro­ grams) do not have historical data on trends in either contraceptive consumption or serv­ ices provided, so they will not be able to use these data sources for making projections. Population data-based forecasts may or may not be feasible. Even in programs that have these data, rapid expansion of service delivery or introduction of additional products con­ fuses historical trend analysis. In forecasting commodity requirements for new or rapidly expanding programs, it is essential to consider the capacity of the distribution system to handle the estimated volume of supplies—even if the forecast is accurate, clients will not receive goods unless the program has adequate storage, transport, and staff. In these situations, a forecast based on distribution system capacity should be prepared. Ideally, such a forecast would take into account three types of programmatic constraints— ˆ The realistic level of demand for services in the program’s target population. ˆ The quantity of services that can realistically be provided by existing staff and facilities. ˆ The quantity of contraceptive commodities that can be stored and moved through the distribution system. This chapter discusses only the third type of constraint—the capacity of the in-country distribution system. Even where projections using other data sources are feasible, a distribution system capacity forecast may be needed. Forecasts based strictly on historical data (either logistics or service statistics) automatically take into account the capacity limitations of a program’s storage and distribution systems, because they reflect rates of growth (or decline) that the 69 The Forecasting Handbook program has proven it can achieve. Population data-based forecasts do not. Thus, popula­ tion data-based forecasts—or logistics- or service-based forecasts that have been adjusted to reflect future program plans—may or may not be consistent with the program’s ability to actually deliver the required commodities. In these cases, an explicit review of the pro­ gram’s storage and transport capacity is also in order. A distribution system capacity forecast can take one of two forms— 1. The forecaster might simply calculate the maximum throughput of the program’s existing storage and transport systems, using this constraint to set a ceiling for forecasts prepared by the other methods. This approach is appropriate in situations where constraints of time, funding, or human resources make it difficult or impossi­ ble for the program to increase its logistics capacity. 2. The forecaster might begin with one or more forecasts prepared by the other meth­ ods and calculate the storage and transport capacity required to achieve the pro­ jected levels of commodity distribution. This approach is appropriate for longer-term forecasts, because the program’s ability to increase its logistics capacity should be greater over the long term. In these cases, the distribution system capacity projec­ tion quantifies the program implications of targets set by one of the other forecast­ ing methods. 7.1. Data Sources and Limitations There are several determinants of distribution system capacity— ˆ Contraceptive requirements. The quantity of each product needed to achieve a particular level of service. ˆ Method mix. The actual (or desired) percentage represented by each method in a facility’s service output. ˆ Staff time. The amount of time properly trained procurement, warehouse, and transportation staff are available (or needed). ˆ Storage capacity. The amount of space available (or required) to store and manage products at each warehouse or storage facility. ˆ Transport capacity. The amount of space available (or needed) on public or private transport for shipping the required commodities down through the distribution system. ˆ Cost or budget. Anticipated costs or maximum budget levels available (or needed) for all of these determinants of capacity. Many family planning programs and almost all HIV/AIDS prevention programs are expanding rapidly. In such situations, the level of all program inputs may be changing quickly, and it is important to try to quantify and understand the implications of such changes. Even in 70 Chapter 7 ] Estimating Consumption Based On Distribution System Capacity mature family planning programs where contraceptive prevalence is no longer growing at a rapid rate, the method mix continues to evolve, following worldwide patterns and trends. Traditional methods gradually are replaced by modern contraceptives, and users of reversible contraceptives gradually switch to permanent methods. These changes have profound implications for the volume of commodities that must be procured, stored, transported, and dispensed. In theory, data on all the determinants of distribution system capacity are available from a program’s administrative management and budgeting systems. In practice, large programs (and less well-managed small programs) may not know the dimensions of every storeroom in the country, or whether all the district trucks are functioning, or the volumes of commodi­ ties distributed by different facilities. In such cases, site visits or surveys may be used to obtain the required infrastructure data. It also may be possible to make simplifying assump­ tions described later in this chapter. Note that the calculations for distribution system capacity-based forecasts are predicated on the assumption that the program has—or can create—a properly functioning distribution system. If this is not the case, a valid distribu­ tion system capacity-based forecast cannot be prepared. Table 17 summarizes some of the advantages and disadvantages of forecasting based on distribution system capacity. Table 17. Forecasts Using Distribution System Capacity Advantages Disadvantages ‰ Closely associated with the idea of a service delivery plan. ‰ Based on very concrete assumptions understandable to program managers. ‰ Essential for determining physical infrastructure and budgetary implications of service or demographic targets. ‰ Usable in new programs with no historical data. ‰ Requires large quantities of local data from all levels of the service delivery system. ‰ Requires properly functioning max­ min inventory control system. ‰ Requires tedious and extensive mathematical calculations. ‰ May require drastic simplifying assumptions. 7.2. Completing the Distribution System Capacity-Based Forecast In preparing the distribution system capacity-based projection, the forecaster and program managers must first decide whether it is possible to change the program’s capacity during the time period covered by the forecast. If changes are possible, then the forecaster’s task is to quantify the capacity changes required to meet program targets prepared by one of the other forecasting methods. If changes are not possible, then the task is to calculate the maximum throughput of commodities that can be achieved with existing capacity. In either 71 The Forecasting Handbook case, it is necessary to calculate capacity requirements for individual transport and storage links at each level in the pipeline and for the pipeline as a whole. 7.2.1. Calculating Storage Capacity Requirements for a Single Facility To calculate the storage space required at a single facility, the forecaster must know the maximum quantity of each commodity to be stored and the unit volume that each com­ modity occupies in storage. Storage volumes per carton for commonly supplied contracep­ tives are listed in appendix 5. In a properly functioning distribution system, the maximum stock level for each product, stated in terms of months of supply, is set by program policy. For example, the policy might be that the maximum stock level at the central store is nine months of supply. These levels, and the quantities of stock they imply, are not the same as the consumption estimates produced by the forecasting methods described earlier. This is because storage facilities must also maintain safety stocks as a buffer against uncertainty of demand (or inconsis­ tency of supply), and quantities sufficient to cover distribution during the time between ordering and receiving new supplies. If max-min policies do not exist, they must be created before the distribution system capacity-based forecast can be prepared.5 The maximum stock level for each facility (in months of supply) can be converted to quanti­ ties to store using the following formula— Maximum quantity Maximum Average quantity to store = stock level x dispensed to clients (in units) (in months) per month The average quantity dispensed may be based either on historical logistics data or on projections made by one of the methods described in previous chapters. After the maximum quantity to store has been calculated, the conversion from quantities to be stored to volume to be stored for any product is— ⎛ Maximum quantity to store Quantity per carton ⎞ ⎟⎟⎠ Cubic meters x per carton Cubic meters ⎜⎜⎝= of storage space 5 For a full discussion of maximum/minimum inventory control systems, see JSI/DELIVER. 2004. The Logistics Handbook: A Practical Guide for Supply Chain Managers in Family Planning and Health Programs. Arlington, Va.: JSI/DELIVER, for the U.S. Agency for International Development. 72 Chapter 7 ] Estimating Consumption Based On Distribution System Capacity Few storekeepers or program managers think of their storerooms in terms of cubic meters of usable storage. They are more likely to think in terms of available floor space. Cubic meters of storage space can be converted to square meters of actual floor space by using two common assumptions— ˆ The maximum height commodities should be stacked (to prevent damage to the product and injury to the storekeepers) is 2.5 meters (8 feet). ˆ In a properly organized storeroom, the floor space needed for aisles, packing and handling areas, ventilation, and so on is equal to at least 100 percent of the floor space actually devoted to product storage. If these assumptions are accepted, then the floor space needed for commodity storage is— Square meters Cubic meters of storage space= of storage space 2.5 meters And the total area needed for the commodity is— Square meters Square metersof storage and = of storage space x 2 handling space Smaller storerooms at lower-level facilities may require less handling space because volumes to be handled are smaller. Managers may decide that 50 percent extra space for handling is sufficient at the clinic level, for example. In this case, the forecaster would multiply by 1.5 instead of 2 in the equation above. On the other hand, large facilities at the central level may require more than 100 percent handling space, particularly if mechanical handling equipment, such as a forklift, is employed. In such cases, a larger multiplier should be used. As an example, suppose that an HIV/AIDS prevention program intends to rent a new central warehouse for condoms, and that program policy specifies that a maximum of nine months of supply should be maintained at the central level. Also, assume a forecast has been prepared that suggests the average quantity dispensed to clients over the time period of the forecast is one million condoms per month. The product is supplied by USAID, which sup­ plies condoms in cartons of 6,000, each with a volume of 0.11 cubic meters, as indicated in appendix 5. The required storage space is— Maximum quantity to store = 9 x 1,000,000 = 9,000,000 units (in units) Cubic meters ⎛ 9,000,000 ⎞ 3 of storage space = ⎝⎜ 6,000 ⎠⎟ x 0.11 = 165 m Square meters 165 m3 2 of storage space = 2.5 meters = 66 m 73 The Forecasting Handbook And, finally— Square meters of storage and = 66 m2 x 2 = 132 m2 handling space These calculations should be repeated for each product to be stored, and the answers summed to obtain the total required size of the warehouse or storeroom. If the grand total is greater than the storage space actually available, several options can be considered— ˆ Alternatives for increasing storage capacity can be examined. In many places, it is possible to rent additional temporary or permanent space instead of building new stores. In most programs, significant additional storage capacity can be gained by simply reorganizing existing space and disposing of unusable commodities. ˆ The maximum quantity to store can be lowered if the delivery schedule into and out of the facility can be changed. The more frequently deliveries are made, the lower the stock levels can be. It may also be possible to reduce safety stock levels by changing the type of max-min inventory control procedures used.6 ˆ The projected quantities to be distributed can be lowered by reducing the projec­ tion for one or more products, repeating the previous calculations until the fore­ casted amounts fit within the storage constraints. A decision to reduce projected distribution because of storage capacity constraints must be made in close cooperation with program managers. From a logistics viewpoint, the easiest way to reduce storage (or transportation) volumes in a family planning program is to reduce projected quantities of condoms, because these are the bulkiest contraceptive commodities. However, in a country with a serious HIV/AIDS problem, this strategy would be absolutely wrong. 7.2.2. Calculating Transport Capacity for a Single Transportation Link In distribution systems that have policies for maximum and minimum stock levels at each facility or each type of facility, stock balances vary between the max and min, and each facility generally needs to be resupplied with the amount of product issued or dispensed in the previous period. To ensure that the system stays in balance, however, resupply decisions should not be made on the basis of quantities issued from higher-level storage facilities to lower-level facilities, but rather on the basis of quantities dispensed to clients at the service level of the system. Thus, for any higher-level storage facility, the monthly quantity that needs to be issued is calculated as— See JSI/DELIVER. 2004. The Logistics Handbook: A Practical Guide for Supply Chain Managers in Family Planning and Health Programs. Arlington, Va.: JSI/DELIVER, for the U.S. Agency for International Development. 74 6 Chapter 7 ] Estimating Consumption Based On Distribution System Capacity Average quantity Average quantity Number to issue = dispensed to clients per month x of SDPs per month at an SDP served Thus a regional medical store that serves 30 SDPs which each dispense an average of 6,000 condoms per month should expect to issue— Average quantity to issue = 6,000 x 30 = 180,000 pieces per month The volume that this amount of product will occupy is calculated using the formula shown previously. If the 180,000 condoms are USAID product, for example— Cubic meters ⎛ 180,000 ⎞ 3 of storage space = ⎝⎜ 6,000 ⎠⎟ x 0.11 = 3.3 m This is the amount of space in the vehicle that will be needed for condoms for a regular replenishment trip if all 30 service delivery points (SDP) are to be restocked in a single monthly trip from the regional medical store. Of course, the calculation illustrated here is based on average quantity dispensed. In actual practice, individual SDPs will need amounts above or below the average, and there are likely to be significant differences in service volumes between SDPs. In such cases, it may be necessary to calculate quantity requirements for each SDP individually. This example assumes that delivery is made by the higher-level storage facility to the SDPs in a single monthly trip. If multiple trips are made (or if lower-level facilities are responsi­ ble for picking up supplies from the higher-level facility), the volume that must be trans­ ported on a single trip is correspondingly reduced. Also, the calculation shown is for monthly distribution; if the delivery schedule were quarterly, for example, the volume estimate must be multiplied by three. As with the storage capacity calculation, this procedure must be repeated for each com­ modity to be transported, with the results summed to obtain the total shipping volume for a single trip. Note that this calculation computes the regular replenishment quantity for a facility. For new programs or facilities, initial stocks equal to the maximum quantity to store must be provided, because the facilities will have no stock on hand at the outset. That is— Initial Maximum quantity to store= quantity (in units) If the stock policy at the SDP specifies a maximum stock level of three months and a mini­ mum stock level of two months, for example, then a new clinic initially needs a three-month supply—18,000 condoms in the previous example. 75 The Forecasting Handbook Analogous calculations can be made to determine the weight of commodities that must be transported. Appendix 5 provides product weights needed to make the computation. Contra­ ceptives are light relative to their volume, so weight is rarely the limiting factor, except where products must be carried by porters or animals. In cases where transport is by air, both weight and volume may need to be considered because of the cost. If these calculations show that existing transport capacity is insufficient, three options are available— ˆ Alternatives for increasing transport capacity can be examined. These include using commercial transport (which may be less expensive than program-owned transport) and procurement of additional program-owned vehicles. ˆ The projected quantities to be distributed can be lowered by reducing the projec­ tion for one or more products. As with storage space constraints, such decisions must be made in close cooperation with program managers. ˆ The frequency of deliveries can be increased. Obviously, if it is possible to deliver monthly instead of quarterly, three times as much product can be transported. But a more frequent delivery schedule is feasible only if vehicles and drivers are available and if fuel, per diem, and other added expenses can be covered. Note that maximum and minimum levels and delivery schedules are interrelated. Specifi­ cally, deliveries must be made at least as frequently as the difference between max and min. That is— Resupply ≤ Max stock level - Min stock levelinterval where the max stock level and min stock level are expressed in months. In the example above, SDPs must be resupplied on a monthly basis (because 3 – 2 = 1). Thus, decisions to change delivery schedules may also require changes to max and min levels, which affect the amounts that the facilities themselves must store.7 7.2.3. Preparing the Aggregate Delivery Capacity Forecast The above calculations produce the capacity requirements (or limitations) for a single facility or a single transportation link. Ideally, this analysis should be performed for each facility and transportation link in the distribution system to identify individual bottlenecks in the various supply chains. To understand the reasons for this interrelationship, see JSI/DELIVER. 2004. The Logistics Handbook: A Practical Guide for Supply Chain Managers in Family Planning and Health Programs. Arlington, Va.: JSI/DELIVER, for the U.S. Agency for International Development. 76 7 Chapter 7 ] Estimating Consumption Based On Distribution System Capacity The total capacity of the existing distribution system can be computed as the sum of the maximum quantities that can be moved through all of the individual supply chains. Of course, each supply chain is only as strong as its weakest link—lower-level facilities cannot move more product than higher-level facilities in the supply chain can provide, even if the lower-level capacity is greater. If the forecasting exercise is undertaken with the assump­ tion that the program’s logistics capacity can be increased during the forecast period, this constraint does not apply. These calculations can only be completed manually in very small distribution systems that manage a small number of products and a few facilities. In larger systems, it may be possi­ ble to use standard max and min levels and averages for transport capacity, storage space, and consumption to perform the analysis for each type of facility rather than for each individual facility. This approach is dangerous in programs where facilities of each type vary greatly in size, because a distribution system may have enough capacity to serve the aver­ age facility, but not enough to serve the largest facilities. In such cases, automated tech­ niques for calculating service capacity must be used to produce a more refined projection. This analysis may produce several alternative strategies for meeting the program’s commod­ ity throughput requirements. For example, capacity can be increased by obtaining addi­ tional storage space and more vehicles; changing max-min policies and increasing the frequency of resupply with existing vehicles might accomplish the same thing. These two strategies have different cost implications and different staffing and management implica­ tions. The forecaster should be prepared to assist program managers in quantifying these implications. A complete example of a service distribution system capacity-based forecast for the ficti­ tious country of Anyland is included in appendix 6. 77 The Forecasting Handbook 78 8. ] Estimating Consumption for New Programs Because new programs lack historical data, they cannot use the logistics data-based fore­ casting method to produce a forecast. However, it should be possible to prepare forecasts using some or all of the other methods, depending on the external data available and the planning process used in establishing the program. In any event, the program’s develop­ ment plan must be the basis for any new program’s forecast. 8.1. Characteristics of an Acceptable Program Plan When new programs are established, they tend to plan based on serving a certain percent­ age of some target population. The population data-based forecasting method can be applied directly to such goals. Some program plans include expected numbers of acceptors, perhaps from a well-defined population subgroup such as factory workers, or from the general population of a limited area of the country. In such cases, the service statistics data-based forecasting technique can be used to obtain consumption estimates. A complete discussion of program planning is beyond the scope of this handbook. For logistics management purposes, however, a new program plan must include at least the following elements— ˆ Estimates of the target population to be served, preferably based on an analysis of unmet need and services already provided by other programs. ˆ A realistic, phased schedule for increased acceptance in the target population (which does not assume that the entire population will be served in the first year of the program). ˆ A specific list of where and when the service delivery outlets will be established. ˆ Numbers of trained service delivery staff now available. 79 The Forecasting Handbook ˆ Projections of when and how many staff will be trained to provide services in the future. ˆ Details of the distribution system to be established, including numbers and capaci­ ties of warehouses and storerooms, types and capacities of transportation links between facilities, and inventory management policies and procedures. ˆ Details of the number and types of personnel available for logistics management activities and the training they already have or will require. ˆ An explanation of how well equipped the program is and/or when needed equip­ ment will be available, as well as details of other resources available for logistics management. ˆ A detailed plan for instituting the data collection and reporting mechanisms required for obtaining logistics and service statistics data for future use in com­ modity forecasting and other program management activities. 8.2. Evaluating the Validity of the Program Plan As mentioned earlier, new programs tend to be overly optimistic. Though the new program has no historical data of its own, it is possible to compare program plans and targets with the experience of similar programs operating in similar environments. The literature on evaluation of family planning programs is filled with information on what can be expected from a given level of program effort. Table 13 on page 56 and table 14 on page 57 provide just two examples of the data available on rates of program growth. Though HIV/AIDS prevention programs are far younger than family planning programs, research efforts are underway to gather similar data on program expansion. Forecasting supply needs for new programs is frequently complicated by incomplete program plans. While most new programs establish service delivery targets by year, few actually detail their current service delivery and logistical capacities or plans for future expansion. To make a distribution system capacity-based forecast, the forecaster has to gather these data. Table 18 summarizes additional points to consider when evaluating a program’s development plan. A common approach in setting or evaluating a program’s targets is to gather a group of experts with experience in the field, and ask them—in either a structured or unstructured format—to assess the program’s prospects based on the likely evolution of the external environment, political events, economic changes, and so forth. This qualitative method of forecasting is commonly used in many fields, and may be a useful adjunct to the more quantitative methods described in this handbook. When a new program plans to offer all methods, the proposed method mix should be com­ pared with other programs serving the same or similar target populations and, if possible, with the method mix and prevalence achieved by the private sector. When they are avail­ able, historical trend data from such programs also should be examined and used as a basis for further quantifying the new program’s likely pattern of growth. 80 Chapter 8 ] Estimating Consumption for New Programs Table 18. New Program Planning: Issues to Consider The most successful national family planning programs have increased contraceptive prevalence one to two percentage points per year between 0 percent and 20 percent prevalence, and two to three percentage points per year between 20 percent and 50 percent prevalence. These are national averages, so annual growth rates that are twice as high may be found in urban areas. Very successful clinics may experience even more rapid growth over some period of time. A good estimate of potential demand for contraception can be made from rates of unmet need if a recent contraceptive prevalence survey is available. Major concentrations of actual or potential clients will likely be located in areas with the lowest fertility rates. It is here that the highest rates of knowledge, positive attitude, and current use almost certainly will be found. Sectors of the population or regions of the country with high rates of prevalence of traditional family planning methods or high abortion rates are more likely to be ready to adopt family planning on a large scale. Programs with strong private sector components tend to grow faster than programs based purely in the public sector. Single-purpose family planning programs tend to grow faster than integrated programs, especially if the single-purpose program is market oriented. Programs that concentrate a smaller number of service points in areas of high demand tend to grow faster than comprehensive programs that have many diverse or dispersed facilities. An effective CBD distributor should produce a minimum of 1.5 CYP per month at the beginning of the program and between 3 and 10 CYP per month when the program is mature. A successful, mature social marketing project that distributes only condoms should generate 0.5 percentage point of additional prevalence per year. If multiple contra­ ceptives are distributed, one percentage point is not unrealistic. It is doubtful that HIV/AIDS prevention programs can develop new markets for condoms faster than the most successful social marketing projects, at least in the general population. 8.3. Completing the Forecast(s) In short, estimated demand must be considered, along with the program’s budget, capacity to deliver family planning services, and anticipated performance of other programs serving all or part of the same population group. Ideally, all these issues will be reflected in the program plan. If the plan has been prepared properly, projected service levels should be stated either in population coverage terms or more specifically in terms of new acceptors and revisits. If the appropriate population data are available, a population data-based forecast based on program goals should be prepared, using the methodology described earlier. In addition to consumption estimates, such a forecast can be used to assess the reasonableness of popula­ tion coverage targets included in the plan (or to generate them if they do not exist). 81 The Forecasting Handbook If the planning process has produced new acceptor and revisit targets, these can be used instead of actual historical data to prepare a service statistics data-based forecast, again using the methodology described earlier. Because such a forecast is not based on historical reality, it is very important to compare it to forecasts made by other methods. In particular, it is essential to compare forecasts with commodity budgets set forth in the program plan or elsewhere, and to highlight discrepancies between the two for consideration by program managers. A distribution system capacity-based forecast should always be completed for new pro­ grams, using either actual or planned facility and transport information, if only to force quantification of the program’s logistics management capabilities. If program planners are unable to produce the necessary data for completing a distribution system capacity-based forecast, it is very unlikely that the program can be implemented successfully. When it is impossible to obtain details of storage, transport, and service delivery capacities at lower levels, projected supply quantities should be compared at least against central-level storage and transport capacities, which are easily assessed. Although the forecaster’s mandate is to look at the situation realistically, care must be taken to avoid discouraging those who are establishing the new program. When in doubt, the forecaster should err on the high side, to ensure that the forecast itself does not become the limiting factor in the program’s expansion. 82 9. ] Estimating Consumption for HIV/AIDS Prevention Programs HIV/AIDS prevention programs are generally younger than family planning programs, so they lack the wealth of historical data on program growth and commodity consumption that family planning programs enjoy. Of the data that do exist—whether for family planning or HIV/AIDS—condom consumption figures are frequently the least reliable, because survey questions often are not comprehensive. However, the severity of the epidemic requires an urgent response, even if historical data for program planning are not available. In spite of these constraints, all of the forecasting methods discussed earlier, with some modifications, are applicable to HIV/AIDS prevention efforts. The forecasting strategies for new programs outlined in chapter 8 are appropriate for new HIV/AIDS programs as well. In all cases, the forecaster should prepare projections using as many data sources as possible, for the same reasons that multiple data sources should be used for family planning program forecasts. 9.1. Estimating Consumption Using Logistics Data HIV/AIDS prevention programs fortunate enough to have historical data on quantities dispensed to clients should prepare a logistics data-based forecast using exactly the same methods described in chapter 4. For most HIV/AIDS prevention programs, it is even more important that logistics data from the lowest possible level of the distribution system be used for the projection, because distribution system malfunction is more likely in newer programs. It is essential to avoid basing forecasts and subsequent shipments only on previous shipment histories, since many national and local HIV/AIDS prevention programs are stocked initially on an emergency 83 The Forecasting Handbook basis, with little or no analysis of actual consumption patterns. Such a strategy is essential for getting distribution systems up and running. In subsequent resupply periods, however, program managers must ensure that shortages due to underestimation of demand do not occur. They must also ensure that condoms do not expire unused at SDPs or remain in intermediate storage facilities until they deteriorate, lest increased condom breakage rates threaten clients’ health and destroy the program’s credibility. To prevent such supply imbalances, HIV/AIDS prevention programs should move quickly to establish LMISs that gather lowest-level consumption data. Many HIV/AIDS programs dispense condoms in non-clinical settings, in which gathering data on individual clients is neither desirable nor feasible. Nonetheless, it should always be possible to gather data on quantities dispensed in aggregate from service delivery points by comparing stock balances at the beginning and end of each resupply period. 9.2. Estimating Consumption Based on Service Statistics The service statistics data-based forecasting method is most applicable to clinic-based HIV/AIDS prevention programs, though the technique can be applied in any setting in which clients are counted and dispensing protocols are standardized. As with family planning programs, separate estimates must be prepared for each different dispensing protocol used by the program. In the family planning example in chapter 5, commodities for new visits and revisits were computed separately, because quantities dispensed to a new client differed from quantities dispensed to old clients. HIV/AIDS prevention programs might or might not have different protocols for new and old clients, but probably will have separate protocols for different target population segments. A pro­ gram that dispenses to commercial sex workers, military recruits, and sexually transmitted infection (STI) clinics, for example, may need to prepare separate projections for each group, because each group’s usage pattern is likely to be quite different. 9.3. Estimating Consumption Using Population Data All HIV/AIDS prevention programs, new or old, should be able to prepare forecasts based on data about the population(s) to be served. However, population data-based estimates must not be made using overall population size or seroprevalence in the general population, as some forecasting methods recommend. These data can be used to determine the severity of the epidemic and the total potential need for condom protection. However, need is not the same as demand for condoms. Even in countries where the scope of the epidemic is well understood, clients’ demand is usually substantially less than the actual need. Though individuals may know they are at risk of contracting HIV or other STIs, they may not use condoms consistently (or at all) to prevent infection. This gap between knowledge and attitudes, and between attitudes and practice, has been well documented in family planning and other health programs. HIV/AIDS programs face the same issues. Though the conse­ quences of becoming infected with HIV far outweigh the consequences of pregnancy, 84 Chapter 9 ] Estimating Consumption for HIV/AIDS Prevention Programs experience with other STIs indicates that fear of adverse consequences rarely causes sustained behavior change among members of high-risk groups. Table 19 shows condom-specific prevalence rates found in selected developed and develop­ ing countries. These figures are taken from the UN’s World Contraceptive Use 1998 wall chart. Unfortunately, these sources have traditionally interviewed only women. When cou­ ples were interviewed, partners’ reports of condom use frequently differed from one another. For this reason, condom prevalence figures must be interpreted with caution. Nonetheless, in these surveys only a few countries—developing or developed—have reached a condom prevalence rate of 10 percent or higher, and the majority have rates of less than 5 percent. Thus forecasts based on an assumption of 20 percent prevalence of condom use in the general population—or even on 10 percent or 5 percent—may vastly overstate likely con­ dom consumption. Instead of assuming that the entire general population will be reached, the forecaster should prepare projections for the specific high-risk groups to be served. Forecasts prepared in this manner require separate estimates for segments of the target population that differ in their sexual behavior—including frequency of sexual relations, coital frequency, sexual practices, and prevalence of condom use. For example, condom use rates vary drastically between monogamous married couples and commercial sex workers. The first step in preparing a population data-based projection for HIV/AIDS programs should be to define the target segments and estimate the size of each segment. For exam­ ple, the target population might be commercial sex workers. One segment might be female brothel workers in a lower-income group, working in the hotel district of a capital city. Another might be women between the ages of 15 and 20 who work for escort services that cater to clientele staying in higher-priced hotels in the tourist area of the capital. After the target population has been segmented into appropriate risk/practice groups, it is necessary to establish quantities of condoms that are required to protect a member of each group over a specific time period. It is usually a dangerous simplification to apply the standard CYP factor from table 15 to an HIV/AIDS prevention program, because HIV/AIDS programs are rarely designed to serve the same general population as family planning programs. Rather, it is necessary to define separate consumption factors, here called seg­ ment member-year of protection or SYP factors, for each different target population seg­ ment. The definition of SYP is the functional equivalent of CYP—the number of condoms required to protect a full time user of condoms for one year. Few data upon which to base SYP factors for particular target segments are actually avail­ able. Initially, programs may have to choose SYPs on the basis of local experts’ estimates, or on interviews with small samples of clients from each segment. For some segments, the standard CYP factor may be sufficient. Whatever factors are chosen initially, the program should move swiftly to design and implement small client surveys to establish more accu­ rate figures. 85 The Forecasting Handbook Table 19. Condom–Specific Prevalence Rates Country Rate Country Rate Country Rate WORLD 4 AFRICA 1 ASIA (excl. Japan) 3 OCEANIA 1 LATIN AMERICA 4 MORE DEVELOPED 14 Japan 46.0 Hong Kong 35.0 Finland 32.0 Sweden 25.0 Singapore 24.0 Denmark 22.0 Grenada 22.0 Slovakia 21.0 Czech Republic 19.0 United Kingdom 18.0 Jamaica 17.0 Costa Rica 16.0 Poland 14.0 Norway 13.0 Mauritius 13.0 Italy 13.0 Spain 12.0 Trinidad & Tobago 12.0 New Zealand 12.0 Canada 10.0 USA 10.0 Republic of Korea 10.0 Bahrain 10.0 Switzerland 8.0 Netherlands 8.0 Barbados 7.0 Hungary 7.0 St. Vincent 7.0 Turkey 7.0 St. Kitts & Nevis 6.0 Guadeloupe 6.0 Iran 6.0 Antigua 6.0 Malaysia 6.0 Dominica 6.0 Paraguay 6.0 Portugal 6.0 St. Lucia 6.0 Venezuela 5.0 Belgium 5.0 Martinique 5.0 Puerto Rico 4.0 Germany 4.0 Australia 4.0 Austria 4.0 France 4.0 Romania 4.0 Bangladesh 4.0 Colombia 4.0 Peru 4.0 Pakistan 4.0 Mexico 4.0 Vietnam 4.0 Zambia 4.0 Brazil 4.0 Guyana 3.0 Montserrat 3.0 Reunion 3.0 Honduras 3.0 Nicaragua 3.0 Ecuador 3.0 Haiti 3.0 Sri Lanka 3.0 El Salvador 2.0 Bahamas 2.0 Panama 2.0 Cuba 2.0 Yugoslavia SFR 2.0 Kuwait 2.0 Qatar 2.0 Bulgaria 2.0 Ghana 2.0 Malawi 2.0 India 2.0 Tunisia 2.0 Guatemala 2.0 Belize 2.0 Nepal 2.0 Zimbabwe 2.0 Thailand 2.0 Oman 2.0 Botswana 1.0 Bolivia 1.0 Dominican Republic 1.0 Swaziland 1.0 Philippines 1.0 South Africa 1.0 Iraq 1.0 Lesotho 1.0 Benin 1.0 Cameroon 1.0 Morocco 1.0 Burkina Faso 1.0 Indonesia 1.0 Kenya 1.0 Jordan 1.0 Tanzania 1.0 Algeria 1.0 Egypt 1.0 Senegal 1.0 Côte d’Ivoire 1.0 Uganda 1.0 Madagascar 0.5 DR of Congo 0.5 Gambia 0.4 Nigeria 0.4 Togo 0.4 Mali 0.4 Syria 0.3 Namibia 0.3 Yemen 0.3 Rwanda 0.2 Mauritania 0.1 Burundi 0.1 Ethiopia 0.1 Sudan -- Liberia -- Niger -- Source: United Nations. World Contraceptive Use, 1998. 1999. Once SYP factors have been selected, the total need for condoms by a segment for any year of the forecast can be calculated simply as— Total need = Population segment size x SYP 86 Chapter 9 ] Estimating Consumption for HIV/AIDS Prevention Programs For example, assume that the population of commercial sex workers is estimated to be 10,000, and 2,000 are found to be in the first segment (brothel workers). It is estimated that each of these brothel workers has sexual encounters in which a condom should be used, on average, six times a night, five nights a week, 50 weeks of the year. This gives a total of 1,500 sexual encounters that require a condom. Thus, the number of condoms to protect a user in this segment—the SYP—is 1,500. The total need is then simply— Total need = 2,000 x 1,500 = 3,000,000 As mentioned earlier, need does not equal either demand or consumption. The total need is the maximum number of condoms that would be used if every member of the segment used a condom with every sexual encounter. However, not all members of a population segment do so. Total current need, defined as the quantity needed to fully protect those members of the segment who are actually using condoms, can be calculated as— Total current = Total need x Percent of segment need using condoms For example, if 40 percent of the population segment members report that they are current users of condoms, then— Total current = 3,000,000 x 0.40 need = 1,200,000 This figure represents the quantity of condoms required to protect the members of the segment if members of the segment who report using condoms use them consistently in all sexual encounters. No country boasts a 100 percent usage rate of condoms, even among current users. Therefore, the consumption estimate for each population segment should be further adjusted by the condom consistency of use rate reported by current users— Total Total current = current x Consistency of use rate demand need 87 The Forecasting Handbook If, on average, self-reported current users of condoms state that they use condoms about 50 percent of the time, then— Total current = 1,200,000 x 0.50 demand = 600,000 Most countries lack data on consistency of use rates for specific target segments. In the short term, the best guesses of local experts can be used to make the projections, but operations research studies should be instituted quickly to obtain more objective data. It may be necessary or appropriate simply to use this total current demand as the estimate of condom consumption. However, not all condoms dispensed to clients are used for disease or pregnancy prevention. Some may be lost, given to friends, used to practice or familiarize oneself with condoms, or simply not used. If the forecaster believes that these quantities are significant, the estimate should be adjusted upward, using the following formula— Total Total ⎛ Total User ⎞ current = current + ⎜ current x wastage⎟ consumption demand ⎜⎝demand rate ⎟⎠ For example, if current clients report that they only use about 8 out of every 10 condoms they are given, corresponding to a user wastage rate of 20 percent, then— Total current = 600,000 + (600,000 x 0.20) consumption = 720,000 An HIV/AIDS prevention program that serves the general population can use this same methodology for estimating consumption. In this case, the population segment to be targeted is all males of reproductive age, and the SYP factor is just the CYP estimate for the general population. The formula for total need is— Total Males of need = reproductive age x CYP In this case, the forecaster should use the CYP factor from table 15 or, preferably, local data. The other formulas are the same as shown earlier. 9.4. Estimating HIV/AIDS Condom Consumption Based on Demographic Surveys Some recent DHSs (primarily in Africa) have consistently asked males about their contracep­ tive behavior, though the male samples are smaller than the female samples. One of the 88 Chapter 9 ] Estimating Consumption for HIV/AIDS Prevention Programs findings of these newer surveys is that males consistently report higher condom use than females, sometimes by a factor of 10 or more. When such surveys are available, their results can be used to generate a population data-based consumption estimate for condoms for HIV/AIDS prevention. This methodology can be used either in addition to or instead of the technique described immediately above; as always, it is better to do both. To prepare a forecast based on DHS sample results for condom use, the number of females in union using condoms is subtracted from the number of males who report using condoms. The difference is taken to be the number of males using condoms with a partner other than their regular partner, and it is assumed that they use these extra condoms primarily for HIV/AIDS prevention (though obviously they may be using them primarily or solely for family planning). This group is considered to be the target population for the HIV/AIDS condom projection. The required data items are— 1. Number of women of reproductive age (WRA) in union. 2. Family planning condom prevalence for WRA in union. 3. Number of men of reproductive age (MRA)—usually age 15–59 years. 4. Reported condom prevalence for MRA. Again, these data are available only in countries where a recent DHS has included a male component. In other countries, this methodology cannot be used. The number of women using condoms is calculated as— Women Women of Condom using = reproductive age x prevalence reported condoms inunion by women Thus if there are 1,000,000 women of reproductive age in union (15–49), and the preva­ lence rate for women in union using condoms from the DHS is 1.0 percent— Women using = 1,000,000 x 0.01 = 10,000 condoms The number of men using condoms is estimated similarly— Men Condom using = Men of x prevalence reported condoms reproductive age by men 89 The Forecasting Handbook For example, if there are 1,100,000 men between the ages of 15 and 59, and 5.0 percent respond that they are condom users— Men using = 1,100,000 x 0.05 = 55,000 condoms The difference between these figures provides an estimate of the number of men using condoms with women other their regular partners. As stated above, this methodology assumes these condoms are used mainly for HIV/AIDS prevention. Thus— HIV/AIDS condoms for Men using = condoms Men using - condoms using Women In this example— Men using condoms for = 55,000 - 10,0

View the publication