In a county of declining population, the losses would diminish each year with my method. For example, consider a county with a 10%/year loss rate that starts with 1000 people. In the first year it would lose 100 people bringing it to 900. In the next year it would lose 90 people bringing it to 810. In the third year it would lose 81, and so on. The absolute number lost decreases each year.
Got it. Then your method is probably better for population loss to be sure. With Georgia, one problem that I had is that for some racial subgroups (mainly Native Americans and Asian, and Other) was that in some cases, I was getting negative numbers. This could happen in counties where there was a very small population and then the population projections had it declining significantly. This happened mainly given the inaccuracy/high margin of error for estimates of population loss/growth for very small populations. In those cases I simply set the population for that particular racial subgroup to a minimum of 0. But your method would, I think prevent that from happening, so I may try that for a future state, in particular if I try to make race data projections again.
Since the Census Bureau produces racial estimate, why not use those directly?
I am using the census race estimates as the basis for projections, but I am not quite sure what you mean by using them "directly." My guess is that maybe you mean something like "why not just take the most recent race estimates from the American Community Survey and project them forward to 2020?"
I assumed you used the Census Estimates, and not the ACS.
The Census provides annual estimates for counties, including breakdowns for race, ethnicity (Hispanic), and separately for age. It does not provide combined populations by race and age.
But you could project 2020 populations based on race and age (VAP). Then assume that the ratio of VAP/total is constant for each race to get projected VAP by race, and correct these totals to match the projected total VAP.