A TECHNIQUE FOR MAKING MORE ACCURATE PROJECTIONS OF MIGRATION AGE DETAIL

Population projections are frequently required for many substate areas and must be prepared with maximal computer and minimal analytical effort. Yet at the same time realistic age detail forecasts call for a flexible means of treating age specific net migration. To do this a simplified version of Pittengers model was used. A major problem in adapting Pittengers model entailed simplifying the computation algorithm. This algorithm is complicated and requires considerable analytical effort in assigning parameter values and calibration testing. When projecting a few populations this causes no special problems but it presents significant difficulty if the projections are to be mass produced which is the concern of this discussion. Another problem had to do with the assignment of migration pattern types to individual populations. An automated assignment procedure had to be developed to replace an essentially judgmental task. The implementation of an automated version of Pittengers model requires essentially 2 steps: defining a set of typical direction flow patterns; and developing an algorithm for assigning a particular pattern to each population group. The model contains a file of directional flow patterns that when correctly selected and properly scaled yield in- and out-migration rates that can be combined to mimic closely the historical age specific net patterns. Changes in scale allow flexibility in pattern shape and for forecasts. The migration patterns used by the model are defined only for ages 15-64. The source for analyzing age specific direction migration flows was data published for State Economic Areas (SEAs). In defining migration patterns the age group at which the peak occurs and the height of the peak are 2 important factors. Also it appears that the downward slope of the curve after the peak varies with the level of migration. Since migration flows in a particular area are apt to be considerably different for each race sex group the procedure is used once for each group. The steps involved are outlined. Once the slope has been determined for a race sex group in a particular area a decision must be made as to which of the 12 patterns within the slope grouping best describe the character of this area. The comparisons were made using data for counties in the states of California New York and Washington. The migration model performed better in 86% of the cases where mean error was the yardstick and in 82% of the cases where the worst fitted cohort was used. The state specific range was 81-93% in the former group and 79-86% in the latter.