Interpreting Results

Interpreting the results of a cohort-survival analysis would appear to be straightforward. The model provides you with estimates of population by sex for the various age groups. The results are, of course, estimates and there will be some margin of error, but that is expected in trying to forecast the future. But this "margin of error" covers a multitude of sins, and under certain conditions the error could be quite large.

The population projection model assumes that all past forces remain the same. It is, essentially, a straight-line projection of past trends in migration and in survival. Events may intervene to make this a highly unreasonable assumption. A war will affect the survival rate of the young adult male population, and that effect will be carried through the population pyramid over time. A shift of economic forces may change the migration rate, and radically alter the profile of the future population. It is possible to modify the straight-line assumption built into the model in this chapter; it is not possible to modify the model to account for "discontinuous" (unforeseen or radically different) change. The further you project into the future, the greater the likelihood that the driving forces will change, even reversing themselves. A five-year projection will probably be fairly accurate; a twenty-year projection is frequently only roughly accurate.

There is also a degree of error introduced by the number of age-groups used in the analysis. The more age-groups, the fewer people there will be in each group. The smaller the "sample" in each age-group, the greater the error due simply to rounding and measurement. In larger towns and cities, measurement error will not be great enough to be a concern. Collapsing the number of age-groups will avoid measurement error, but at the price of a cruder analysis, not only in terms of population composition but also in terms of the time-interval which is considered.

Nor does the cohort-survival method of projecting population predict other important covariants of age and sex. Often planners and managers are as interested in the income or class characteristics of the population as they are in its size. It is possible (using a mainframe computer) to develop population projection models which incorporate such considerations. But such models require much more data, incorporate many more assumptions, and need a very large population base if they are to avoid the problems of measurement error. They are used more frequently on the regional or the State level.

609