SteveScott
commented
2 years ago Interpolation would require installing geospatial libraries onto the runner. This may not be feasible. An alternative is to take a sample, and impute missing blocks by going one step up in geography at the census tract level by taking the average over the tract to fill in the missing data. The county ID of the parent is contained in the GeoID.
In order to calculate Where New Yorkers are visiting (destination), you have to loop through all the home locations within every census block group in the nation. It is taking a long time to execute and may fail. We may have to resort to taking a sample of the dataset and using that to infer the other data geospatially.
For example, if we had a map of the country, and half the blockfaces were sampled and the other half were dropped, we would need to create a geospatial interpolation through the census block groups (as points) to infer the missing census block groups (as points). This would be very accurate because there is a high autocorrelation between the population of neighboring blackfaces.