Population

[dsweber2]

We use population at various geo levels to aggregate some signals. @RoniRos has suggested that we make these publicly available, both for reproducibility and as a useful signal in their own right.

We have some geo_type's population in the geo_mapper utility already, and this is about making those available. Unclear whether we should use version or reference_date to record populations changing over time (at the moment, the geo_mapper has only 2 values). reference_date would more accurately reflect the meaning, but version would allow compression to only the actually present values.

[RoniRos]

Unclear whether we should use version or reference_date to record populations changing over time (at the moment, the geo_mapper has only 2 values). reference_date would more accurately reflect the meaning, but version would allow compression to only the actually present values.

I'm not sure what the right answer is, but let me share what I think would be the use cases: I expect population sizes to be revised once a year, by importing extrapolation estimates from the US Census Bureau. Then different signals will use the population numbers that are relevant to their respective reference_date. So it feels at least like 'reference_date' is very appropriate here.

Another use case is in deriving what I call a "micro density" signal, which is people density per land area, but calculated at the very fine level of census block or census track, then geo-aggregated by weighing by population size (so you get the "average density" that a person in the country experiences). It's basically #people/land area, but at a very fine geo level. This will require population sizes at these fine geo-levels, The land areas at census level (excluding e.g. bodies of water) are available and unlikely to change for a given geo unit (unless the geo units themselves change). So "micro density" is also expected to be revised annually, with the population revision.