Reading Legacy Census Redistricting Data

The US Census Bureau releases redistricting data (P.L. 94-171) following the decennial census. This article walks through the basics of reading that data and getting it into working form.

The P.L. 94-171 Legacy File

The redistricting data mandated by P.L. 94-171 is composed of six tables, five of population characteristics, and one of housing characteristics. In the legacy data format, these six tables (and dozens of levels of geography) are split into four files: File01, File02, File03, and the Geographic Header Record. Files 01–03 have the actual decennial census data for each table, while the geographic header has geographic identifiers (FIPS code, precinct IDs, etc.) and information (area, water, etc.).

P.L. 94-171 file layout

These four files have the same rows, each of which is identified by a LOGRECNO number. Combining the columns shared across the four files yields the full P.L. 94-171 file

The basic unit of Census geography is the block; all other geographies are constructed from them. But the P.L. 94-171 file is not composed of blocks alone. The Census has already tabulated the six table statistics across every possible level of geographies. To get information for a single geography level—blocks, or counties, or school districts—one needs only to subset to the rows which correspond to this geographic level (coded in the SUMLEV column).

So the basic process for working with legacy P.L. 94-171 data is:

  1. Read in the four P.L. files (pl_read(), pl_url())
  2. Combine the four files into one, and subset to the desired geography level (pl_subset(); geography level codes listed in pl_geog_levels)
  3. Select the desired variables from the six included tables (pl_select_standard())
  4. Optionally, combine the processed data with the corresponding tigris shapefile.

The utility function pl_tidy_shp() combines all of these steps into one function for most common use case of tabulating basic redistricting information at the block level. This is demonstrated in the README

Using the PL94171 package

library(PL94171)

The four components of the P.L. 94-171 file should be downloaded into their own directory. Here, we’ll use the example data included in the package, from the 2018 end-to-end Census test in Providence County, Rhode Island, and read it into our R session. In general, you can provide a URL in place of a file path, and the package will read the data from the URL. The pl_url() function will automatically construct the URL to the data for a given state and year.

# `extdata/ri2018_2020Style.pl` is a directory with the four P.L. 94-171 files
path <- system.file("extdata/ri2018_2020Style.pl", package = "PL94171")
pl_raw <- pl_read(path)
# try `pl_read(pl_url("RI", 2010))`

This creates a large list where each individual P.L. 94-171 file component is a separate entry in the list. If we look at the top of one of these entries, we see the same structure as in the schematic above: each redistricting variable is a column, the rows are indexed by LOGRECNO, and various levels of aggregation are all included as different sets of rows in the same table (notice the countywide population counts in the first two rows).

head(pl_raw$`00003`)
#> # A tibble: 6 x 15
#>   FILEID STUSAB CHARITER CIFSN LOGRECNO P0050001 P0050002 P0050003 P0050004
#>   <chr>  <chr>  <chr>    <chr>    <int>    <int>    <int>    <int>    <int>
#> 1 PLST   RI     000      03           1    15766     7311     3266      261
#> 2 PLST   RI     000      03           2    15766     7311     3266      261
#> 3 PLST   RI     000      03          19      732        0        0        0
#> 4 PLST   RI     000      03          20        0        0        0        0
#> 5 PLST   RI     000      03          21        1        0        0        0
#> 6 PLST   RI     000      03          22        1        1        0        0
#> # ... with 6 more variables: P0050005 <int>, P0050006 <int>, P0050007 <int>,
#> #   P0050008 <int>, P0050009 <int>, P0050010 <int>

To subset to a desired geography level, we must first identify the corresponding SUMLEV code.

print(pl_geog_levels)
#> # A tibble: 85 x 2
#>    SUMLEV SUMLEV_description                                     
#>    <chr>  <chr>                                                  
#>  1 040    State                                                  
#>  2 050    State-County                                           
#>  3 060    State-County-County Subdivision                        
#>  4 067    State-County-County Subdivision-Subminor Civil Division
#>  5 140    State-County-Census Tract                              
#>  6 150    State-County-Census Tract-Block Group                  
#>  7 155    State-Place-County                                     
#>  8 160    State-Place                                            
#>  9 170    State-Consolidated City                                
#> 10 172    State-Consolidated City-Place within Consolidated City 
#> # ... with 75 more rows

Here, we’ll look at Census tracts, which are SUMLEV=140.

pl <- pl_subset(pl_raw, sumlev="140")
print(dim(pl))
#> [1]   7 397

We see that all four components have been combined into one large table, with data for each of the seven Census tracts in the example file recorded in a single row. To extract commonly-used variables from the 397 columns, we can run the following:

pl <- pl_select_standard(pl, clean_names = TRUE)
print(pl)
#> # A tibble: 7 x 24
#>   GEOID       state county row_id summary_level vtd     pop pop_hisp pop_white
#>   <chr>       <chr> <chr>   <int> <chr>         <chr> <int>    <int>     <int>
#> 1 44007000101 RI    007        19 140           <NA>   3970     1442      1445
#> 2 44007000102 RI    007        20 140           <NA>   4735     2555       815
#> 3 44007000200 RI    007        21 140           <NA>   5703     3766       570
#> 4 44007000300 RI    007        22 140           <NA>   6647     3827       754
#> 5 44007000400 RI    007        23 140           <NA>   3433     2130       142
#> 6 44007000500 RI    007        24 140           <NA>   2940     2249        39
#> 7 44007000600 RI    007        25 140           <NA>   1797      778       170
#> # ... with 15 more variables: pop_black <int>, pop_aian <int>, pop_asian <int>,
#> #   pop_nhpi <int>, pop_other <int>, pop_two <int>, vap <int>, vap_hisp <int>,
#> #   vap_white <int>, vap_black <int>, vap_aian <int>, vap_asian <int>,
#> #   vap_nhpi <int>, vap_other <int>, vap_two <int>

Above, we set clean_names = TRUE, which is the default. This creates a set of variables familiar to the redist family of packages.

To combine these data with a shapefile, we must use the tigris package. The GEOID column is shared between the P.L. 94-171 data and the TIGER shapefiles from tigris.

library(tigris)
library(sf)
library(dplyr)
library(ggplot2)

ri_tracts = tracts("RI", county="Providence", year=2020, progress_bar=FALSE)

full_join(pl, ri_tracts, by="GEOID") %>%
ggplot(aes(fill=pop, geometry=geometry)) +
    geom_sf(size=0) +
    theme_void()