Group Quarters populations are notoriously difficult in sub-county geographies and this difficulty is magnified by the ever-changing boundaries. Consider this: Census Block Group (CBG) 060372652012 is located just outside of Los Angeles. (1/6)
Using @nhgis rectified boundaries, the population totals for the last 3 censuses are
1990: 499
2000: 2571
2010: 548
Did this CBG magically quadruple in population in 10 years and then immediately quarter in size in the 10 after that? Of course not! That would be ridiculous!(2/6)
1990: 499
2000: 2571
2010: 548
Did this CBG magically quadruple in population in 10 years and then immediately quarter in size in the 10 after that? Of course not! That would be ridiculous!(2/6)
So where exactly is this CBG? It's presently located right outside of UCLA but in 2000 it included part of UCLA! That huge growth between 1990 and 2000 is almost entirely due to the inclusion of Group Quarters population that was then drawn out of the CBG in Census 2010! (3/6) 
The GQ populations are
1990: 155
2000: 2208
2010: 202
When we remove the GQ population the residential population numbers are much more reasonable:
1990: 344
2000: 363
2010: 346 (4/6)
1990: 155
2000: 2208
2010: 202
When we remove the GQ population the residential population numbers are much more reasonable:
1990: 344
2000: 363
2010: 346 (4/6)
If you wanted to project this CBG and don't account for GQ, you'd end up with a very weird projection since the data -- even with rectified boundaries -- are, well, weird. When we remove GQ population we get a much more consistent time series. (5/6)
If you want to do small area population analysis, you should probably pay attention to GQ populations. (6/6)
