While actually making a population projection is a routine application of a computer program, the complexity of the undertaking lies in the derivation of the input data. Gathering the base data, ensuring that they are of adequate quality, adjusting them as necessary using demographic techniques, and assessing their comparability among countries are all activities that ensure the success of the projection process. Once the base estimates are derived, the researcher also must make reasonable and consistent assumptions about the future course of fertility, mortality, and international migration. Regional and world populations are obtained by first projecting each country population separately and then combining the results to derive aggregated totals. This section (adapted from Arriaga and Associates 1995) briefly summarizes the process of preparing population projections by the cohort component method.
The cohort component population projection method follows each cohort of people of the same age throughout its lifetime according to its exposure to mortality, fertility, and migration. Starting with a base population by sex and age, the population at each specific age is exposed to the chances of dying as determined by projected mortality levels and patterns by sex and age. Once deaths are estimated, they are subtracted from the population, and those surviving become older. Fertility rates are projected and applied to the female population in childbearing ages to estimate the number of births every year. Each cohort of children born is also followed through time by exposing it to mortality. Finally, the component method takes into account any in-migrants who are incorporated into the population and out-migrants who leave the population. Migrants are added to or subtracted from the population at each specific age. The whole procedure is repeated for each year of the projection period, resulting in the projected population by age and sex, as well as birth and death rates, rates of natural increase, rates of population growth, and other summary measures of fertility, mortality, and migration for each year.
For many developed countries, base data on population are taken from population registers or are current official estimates prepared by national statistical offices based on a census for an earlier year. For developing countries, the base population for a projection is taken from the latest census, generally since 1980. However, census enumerations are not perfect, and reported data on a population age and sex structure may be affected by age misreporting and by underenumeration of persons in certain ages. If the projection starts with errors in the base year, such errors will be carried throughout the projection period and will have an impact on the projected number of births as well.
Consequently, before being accepted to serve as a base for the projections, a population must be evaluated to detect errors and adjusted as necessary to correct them. Various methods have been developed to detect age misreporting, including analysis of digit preference, age ratios, and sex ratios. Techniques have been developed for making any needed corrections. Depending on the country-specific data problems, slight smoothing or strong smoothing techniques may be recommended. The base population age and sex structures for most developing countries in this report are at least slightly smoothed for the population ages 10 years and over.
Special attention is given to possible underenumeration of the youngest age groups, 0 to 4 years and 5 to 9 years, because errors in these ages may have a significant impact on the total projection. Suppose, for example, that children ages 0 to 4 years were undercounted in the base population. In the projection, not only would the surviving cohorts of these children be smaller than they should be, but when the female cohorts reached reproductive ages, the number of births they had would also be underestimated. The completeness of enumeration of these youngest age groups is evaluated by checking for consistency between the number counted and the estimated levels of fertility and mortality during the 10-year period prior to the census date, as children of these ages represent the survivors of births during that period.
When vital registration data are available and complete (which is usually the case only in developed countries), it is easy to construct life tables using microcomputer programs, and to thereby derive both a level and an age pattern of mortality suitable for the projection process. For most developing countries, however, it is necessary to estimate mortality some other way. Various techniques have
been developed to evaluate and correct information on deaths by sex and age in relation to information on population. Data on deaths may be provided not only in vital statistics registers, but also in surveys or censuses that include questions concerning deaths during a specific period of time; for example, deaths of any household members during the past year. If registered deaths can be evaluated and adjusted for errors, they can be used to obtain valuable information about the level and pattern of mortality.
There are several techniques (for example, the Coale-Preston technique, the growth balance technique developed by Brass, and the Bennett-Horiuchi technique) for estimating underregistration of deaths. Some of them are based on the assumption that the population is "stable." A stable population is one in which there has been no migration, and neither fertility nor mortality has changed in the past. Other techniques, developed more recently, do not require the assumption of stability. Some methods (for example, the Brass technique and modifications developed by Trussell, Sullivan, and Feeney; the Palloni-Heligman technique, and the Johnson technique) may be applied to estimate mortality during the first years of life. They are based on data on children ever born and children surviving, by age of mother.
Like mortality in infancy and childhood, mortality in adult ages can be estimated indirectly when reliable data are not available to measure it directly. Two principal techniques have been developed to estimate adult mortality based on information collected in censuses or surveys. They are the orphanhood technique, based on the number of persons whose mother or father has died, and the widowhood technique, based on the number of persons whose first spouse has died. Both provide an estimate of survivorship levels between two adult ages for a period of time prior to the year of data collection. However, these techniques are seldom used for the base mortality patterns of the projections in this report because the reference period to which the estimated mortality pertains is not well defined.
As in the case of mortality, procedures for estimating fertility depend on the availability of data and on the detail of the information. For cases where vital registration is complete, fertility can be measured directly using classical procedures. Most developing countries, however, do not have reliable vital statistics, and so techniques have been developed to measure fertility indirectly based on census or survey information.
Using the age structure of the population, the crude birth rate is sometimes estimated by the rejuvenation technique, in which the population at the youngest ages is "reverse survived" to determine the number of births from which they are survivors. This technique is attractive because it does not require the collection of any data related specifically to fertility. However, the reliability of the estimate depends on the quality of both the census data on age and the survival ratios used for the rejuvenation.
Under certain circumstances, census data by age can be used to obtain not only a crude birth rate but age-specific fertility rates as well. This is done by using the own-children technique based on information on children and women by single years of age. This technique requires data linking individual children to their natural mothers.
Other techniques, such as the Rele technique, use census data by age to calculate the net reproduction rate or total fertility rate based on the relationship of children of specified ages to the number of women in childbearing ages.
Finally, and most importantly for many developing countries, many censuses and surveys include questions related specifically to fertility; for example, the number of children women have had and whether they had a birth in the year preceding the inquiry. Responses to such questions can be used to estimate fertility indirectly. Some techniques to do this include the P/F (Parity/Fertility) ratio developed by Brass, based on the average number of children ever born to women in 5-year age groups and women's age pattern of fertility derived from births in the year preceding the census or survey; the P1/F1 ratio technique, also developed by Brass, based on first births only; and the Arriaga technique, which is similar to the P/F ratio technique but links data for more than one date. All of these methods can be used to estimate the age-specific fertility rates required for making component population projections.
Although migration is sometimes an important component of population change, it is not generally well recorded except in some European countries, such as Sweden and the Netherlands, that maintain complete and detailed population registers. Some countries collect information on arrivals and departures of passengers at the official borders of the national territory, but such data are seldom processed in such a way as to render them useful for statistical purposes. Even in countries with otherwise excellent statistical systems, information on international migration is often unreliable.
The primary source of information on immigration for purposes of population projections is census data on place of birth of the foreign-born population. To detect emigration as well, in order to calculate the net movement in or out of a country, it is necessary to find data for the countries in which the emigrants have settled (since they are the foreign immigrants of that country). In addition, special migration flows, such as refugee movements, are incorporated by considering reported numbers of refugees from the United Nations High Commissioner on Refugees, country sources, and media reports. Thus, most data on international migration are educated guesses at best, especially since not only total numbers but also age and sex distributions of the migrants are required for the projection process.
Once levels of mortality, fertility, and migration have been determined for the base year of the projection, each component must be projected into the future. Although the procedure for doing this is mechanical, careful attention must be paid in determining projected levels, trends, and patterns by age. Not only must the assumptions be appropriate for the particular country in question, but consistent assumptions must be made when projections are being carried out for more than one country.
An expected increase in contraceptive prevalence is implicit in the assumptions about future fertility declines for most developing country projections. For many developed countries, future fertility levels are projected to experience only minor change, either slight decreases, or in some cases, slight increases.In general, mortality is expected to continue to decline in most countries, as development and health advances continue. A particular exception relates to the impact that acquired immune deficiency syndrome (AIDS) will have on the mortality of some countries, where mortality levels in the next decades are expected to increase. (For a description of the method used to incorporate the impact of AIDS mortality on selected populations, see the next section of this appendix.) While there is no single "right" way to make assumptions about the future, the following procedures are those recommended and generally used by the Bureau of the Census for the projections presented in this report.
The first step is usually to assign a target level of life expectancy at birth and total fertility rate for some intermediate year in the future or the last year of the projection period. Next, a trend of these measures is determined for the period between the base year and the last year. Then, an age and sex pattern of mortality and a female age pattern of fertility are determined for each projected level of life expectancy and total fertility rate, respectively.
In setting target levels for both mortality and fertility, available data on past trends are taken into consideration. If estimates are available for more than one date in the past, a logistic function can be fitted to these data, since this function approximates expected changes in life expectancy at birth and total fertility rate. The results of the logistic function must be carefully scrutinized, however, to ensure that they yield an acceptable future target for the individual country circumstances.Recent population and socioeconomic trends and policies of each country are taken into account to determine if the projected trends are plausible. For example, for mortality, information concerning programs of public health are considered in judging the results. For fertility, factors such as trends in age at marriage, the proportion of women using contraception, the strength of family planning programs, and any foreseen changes in women's educational attainment or in their labor force participation in the modern economic sector are considered.
In some instances, no data on past trends are available to which a logistic curve can be fitted. In such circumstances, life expectancies can be projected based on increases related to the general level of mortality. The United Nations has recommended such increases based on countries with available data. For fertility, when trend data are not available for estimating future changes using a logistic function, the past experience of other countries serves as a guideline to determine the pace of future change.
Once levels of life expectancy at birth and total fertility rate have been set for the base year and some future year or the last year of the projection, a logistic function is often used to determine the trend. For developed countries with little expected change in fertility, intermediate levels are often determined linearly rather than logistically.
The next task is to determine an age pattern of mortality and fertility for each of the projected values, since these patterns tend to vary as overall levels change. For each level of projected life expectancy at birth, a set of central death rates is estimated using an iterative interpolation process. The interpolation is logarithmic and uses a set of central death rates for the base year and a "limit" set of rates with very low mortality. Life tables constructed with the interpolated rates correspond to the life expectancies at birth projected previously. Age-specific fertility rates for each projected level of total fertility rate are interpolated between the set for the base year and "model" sets derived from empirical data for populations at various levels of total fertility.
Once mortality and fertility have been tentatively projected for each country according to its particular circumstances, the estimates are compared with projected values for other countries in the same region and with those for other regions. Differences are evaluated to make sure they exist for valid reasons that can be explained by known peculiarities of the particular countries.
Finally, in recent years the Bureau of the Census has concluded that distinctive mortality assumptions must be made for selected countries in this report because of the death risk due to AIDS. Using methodology that takes into account the effect of AIDS, country projections have been prepared that assess its impact on future populations in countries where the infection is significant.
Assumptions about future migration are generally much more speculative than assumptions about fertility and mortality. International migration may occur as a result of changing economic conditions, or as a result of political unrest, persecutions, famines, and other extreme conditions in the countries of origin. Thus, individuals may feel rejected by stagnated economies and attracted by industrialized societies, or refugees may flee in large numbers looking for better or more stable lives elsewhere.Due to the unpredictability of conditions such as crop failure, emerging violence, and bellicose activities, migration forecasts are subject to large errors. If migration is known to have a negligible impact on a country's current growth rate, future migration is often assumed to be nil. If a country's migration is known to be significant, the estimated number of migrants during the past is frequently held constant in projecting to the near future. Projected migration is usually assumed to diminish, reaching zero at some year in the medium- to long-term future. The age and sex composition of international migrants depends on the situation in each country. If information is not available, model patterns by age and sex are sometimes used.
Due to the differing nature of the base data for each country, there is no standard starting date for each country's projection. The projection period for a few countries started as recently as 1990 when the base information was current to that date. In contrast, the projection period for many African countries (and a few countries in other regions as well) started as long ago as the 1970's, or even before, although information for a later date on one or more of the variables may have been taken into account for the early years of the projection. "New" information for such a country may pertain to 1980 as opposed to a 1970 figure available for the previous round. Thus, total populations in the revised projections may change for any year in the past.
When the projected population for any individual country changes, so does the aggregated total for the corresponding region and for the world. New aggregations are made for world regions and world totals, combining the latest projected data for all countries, and superseding previously projected world and regional totals given in previous reports.
The differing starting dates complicate aggregations not only of total population but of vital rates and other measures as well. For this reason, regional and global aggregations of crude birth and death rates, life expectancy at birth, infant mortality rates, and age-sex distributions of the population generally can be presented only for the latest year for which all countries have a projected estimate for each variable. In this report, such measures are usually shown for 1996.
