A function to return smallest available geography about tidycensus

[walkerke]

You bring up some interesting points that I'll keep in mind. I do like having users be deliberate about the geography they choose - especially given the trade-offs (block groups are more refined, but often have very large MOEs). However I would like to allow for users to get block group data by state. Block group data are only available from the API by county, so the logic within the package at the moment makes a call to the Census API for each county specified and then combines the result for the user.

To get all block groups for a state, the user would need to supply all the counties for that state at the moment. This can be done with the built-in dataset fips_codes :

or_counties <- filter(fips_codes, state == "OR") %>%
  pull(county_code)

or_hhinc <- get_acs(geography = "block group", variables = "B19013_001", 
                  state = "OR", county = or_counties)

However that is not entirely intuitive and I haven't documented it, plus it isn't very efficient if geometry == TRUE. I think that is logic that could be taken care of internally, which I'll put on the agenda for version 0.4.

[dkyleward]

Actually, this could potentially be an avenue to restructure the basic logic of the package. I was having a hard time getting "block group" info. A little googling, and I realized that I had to specify the county variable in order to get block groups.

I started looking at your code for get_decennial. I noticed that you are handling nesting cases in a one-off approach. For example:

if ((geography %in% c("block group", "block") & length(county) > 
        1) | (geography == "tract" & length(county) > 1)) {

and:

if (geography == "tract" & length(state) > 1) {

If you established a hierarchy for geometries inside the function, then you'd be able to handle most use cases with the same logic. Here's some pseudo code with just states, counties, and tracts. In short, it loops over the largest geographies first and then moves into smaller geographies - getting data as needed. It's not exactly right, but it conveys the idea.

if is.null(state) {state <- get_all_states()}
if geography == "state" {result = get_state_data(state)}
for each state:
  if is.null(county) {county <- get_all_counties(state)}
  if geography == "county" {data = bind_rows(data, get_county_data(county))}
  for each county:
    if is.null(tract) {tract <- get_all_tracts(county)}
    if geography == "tract" {data = bind_rows(get_tract_data(tract))}

return(data)

Right now to get block group data, I have to specify county. That wasn't clear to me at first, and I was getting 400/404 messages. In the above approach, if I specify a state and "block group" for geography, I'd get all the block groups in that state. The main improvement is that I don't have to handle each special case or combination. The user simply narrows the scope of the request by adding more options to the function (state/county/etc).

Admittedly, I'm not familiar with the census API, so this may not be appropriate. Also, I'm not familiar with all the various census geographies - some may not fit nicely into this (e.g. county parts).

[walkerke]

I've just pushed this functionality in 9eaff0a. For example:

az <- get_acs(geography = "block group", state = "AZ", variables = c(totalpop = "B01003_001"), geometry = TRUE)

head(az)

Simple feature collection with 6 features and 5 fields
geometry type:  MULTIPOLYGON
dimension:      XY
bbox:           xmin: -109.8778 ymin: 34.2605 xmax: -109.0454 ymax: 36.99921
epsg (SRID):    4269
proj4string:    +proj=longlat +datum=NAD83 +no_defs
# A tibble: 6 x 6
         GEOID                                                        NAME variable estimate
         <chr>                                                       <chr>    <chr>    <dbl>
1 040019702001    Block Group 1, Census Tract 9702, Apache County, Arizona totalpop     2484
2 040019441005    Block Group 5, Census Tract 9441, Apache County, Arizona totalpop      987
3 040019442022 Block Group 2, Census Tract 9442.02, Apache County, Arizona totalpop     1704
4 040019443002    Block Group 2, Census Tract 9443, Apache County, Arizona totalpop     1512
5 040019427002    Block Group 2, Census Tract 9427, Apache County, Arizona totalpop     1640
6 040019702002    Block Group 2, Census Tract 9702, Apache County, Arizona totalpop     1275
# ... with 2 more variables: moe <dbl>, geometry <simple_feature>

It is a bit slow when getting feature geometry for states with large number of counties (e.g. Texas) due to the way tidycensus handles this internally but it works.

[dkyleward]

Awesome. I'll take functionality over speed any day! (especially for first attempt.) Thanks for doing this.