ZCTA to State Crosswalk

[alamstein]

I am unsure whether to post this here or the tidycensus package.

I am doing a ZIP level analysis in a business context and we are interested in getting ZCTA level demographic data from Census because it is the best approximation for ZIP codes. A problem we are running into is that none of the ZCTA data has a state associated with it. E.g.

library(tidycensus)
library(mapview)

zcta_income <- get_acs(
  geography = "zcta",
  variables = "B19013_001",
  year = 2021,
  geometry = TRUE
)

# No State info
head(zcta_income) 

# Creates a map of the entire US, whereas I'd like to just be of a single state or county
mapview(zcta_income, zcol = "estimate")

When I google this I see comments like "Well, this is because Census does not return State data". But when I created the choroplethrZip package I had this exact same problem and was able to solve it. I stored it in an object called zip.regions. E.g.

library(choroplethrZip)
data(zip.regions)
head(zip.regions)
  region state.name county.name county.fips.numeric  cbsa cbsa.title
1  70560  louisiana      iberia               22045 10020       <NA>
2  70510  louisiana   vermilion               22113 10020       <NA>
3  70592  louisiana   lafayette               22055 10020       <NA>
4  70548  louisiana   vermilion               22113 10020       <NA>
5  70560  louisiana   vermilion               22113 10020       <NA>
6  70578  louisiana   lafayette               22055 10020       <NA>
  metropolitan.micropolitan.statistical.area
1                                       <NA>
2                                       <NA>
3                                       <NA>
4                                       <NA>
5                                       <NA>
6                                       <NA>

To see where I got the data from, type ?zip.regions:

Data comes from the US Census Bureau's 2010 ZCTA to County Relationship File (https://www.census.gov/geo/maps-data/data/zcta_rel_download.html), 2010 ZCTA to Metropolitan and Micropolitan Statistical Areas Relationship File and the Core Based Statistical Areas (CBSAs) and Combined Statistical Areas (CSAs) Metropolitan and Micropolitan Delineation files (http://www.census.gov/population/metro/data/def.html).

Note that this data is from 2010 and the first link no longer works. But if I Google "ZCTA Relationship File" you wind up here and if you click through to the "2020 ZCTA to County Relationship File" you wind up here. It appears that perhaps they only update this file in years ending in 0. If you search that file for "ZCTA5 + < your zip code >" You'll probably find both the state and county for your ZCTA. For example, here is row for "ZCTA5 90210":

221704258470394|90210|ZCTA5 90210|27823432|153478|G6350|B5|S|275901063468976|06037|Los Angeles County|10513491099|1787501506|G4020|H1|A|27823432|153478

The key point is not just the text "Los Angeles County". It is also the entry before that (06037): 06 is the FIPS code for California and 037 is the FIPS code for Los Angeles County. So this file will let you filter a national dataset to "All ZCTAs in a state" as well as "All ZCTAs in a county."

I am not sure if you were aware of this file or not. But I think it would help a lot of people if this data wound up in either tigris or tidycensus, since so many demographic questions in business / marketing try to use ZCTAs.

@walkerke Is this data / functionality that you are interested in adding to the package?

[walkerke]

Thanks Ari! I'll give this some thought as users often do ask for ZCTAs by state, and Census has discontinued the state option for ZCTAs on the API since the 2016-2020 ACS.

[alamstein]

@walkerke , it occurred to me that my team might need this before you are able to implement it. Also, it seems not terribly hard for me to implement. While I could be wrong, I think that I will need to:

1. Download the relationship file, do some minor processing of it, and call it something like zcta_crosswalk
2. Write a function get_zctas_in_state
3. Write a function get_zctas_in_county
4. Probably implement it as a package so that people can run the code without any conflicts with their local environment, so that I can update the code later to do things like get the latest data, possibly store historical data, and so on. (I am still not 100% clear on how often Census updates this dataset).

I mention this because it seems like, once the tool is written internally, it wouldn't be too hard to put it on github and/or CRAN. Of course I would need corporate permission for that. But if I can open source the code, then you could incorporate it at your leisure (either as a new package dependency for tigris or just copying the code directly into tigris).

Is this something that interests you?

[alamstein]

FYI, I just wrote an initial version of this package. My company was kind enough to let me open source it: https://github.com/MarketBridge/zctaCrosswalk.

[walkerke]

Nice, thanks Ari! I think people will find your package useful.

[alamstein]

@walkerke Here is an example of using the development version of zctaCrosswalk with tidycensus to make a ZCTA level map of ACS data for a single county.

Note that this initial example pretty much comes directly from your recent U-M workshop (you may have had national data by county, not zcta, but the code is very similar):

library(tidycensus)
library(mapview)
zcta_income <- get_acs(
  geography = "zcta",
  variables = "B19013_001",
  year = 2021,
  geometry = TRUE
)
mapview(zcta_income, zcol = "estimate")

There are two problems here:

1. There are too many ZCTAs at the national level for the map to be useful.
2. Even when you zoom in, the scale doesn't change, so the map is still not very useful.

zctaCrosswalk is designed to make it easy for you to subset the above dataframe by one or more counties or states prior to rendering:

library(devtools)
library(dplyr)
install_github('https://github.com/MarketBridge/zctaCrosswalk')

# San Francisco has FIPS code "06075"
sf_zcta_income = filter(zcta_income, 
                        GEOID %in% get_zctas_by_county("06075"))
mapview(sf_zcta_income, zcol = "estimate")

The latest version of zctaCrosswalk is designed to be pretty user-friendly. For example:

• get_zctas_by_state accepts "06" (character with leading 0), 6 (numeric), california (case insensitive) and CA (official USPS abbreviation) for all 56 "states" in the dataset.
• get_zctas_by_county accepts both "06075" and 6075 but not San Francisco County because of the large number of duplicate counties (30 states have a "Washington County"!). Filtering by county name seems like something people say they want ... until they get bitten by this bug.
• Function get_zcta_crosswalk contains the initial code I used to scrape and process the dataset from the Census website. Although note that this function is really just the beginning of the processing, not the end - for example, the initial dataset does not contain state names or abbreviations (I had to get that data elsewhere and add it in manually).
• The package is designed to have the smallest possible footprint - it does not have any superfluous dependencies.

Now that I'm near the end of the project (although it's still not on CRAN, so famous last words), I do have two thoughts I wanted to share with someone who works in the field:

• I am not aware of a "FIPS" or "state names" package, but it might make sense to create one. Adding in the "missing state names", plus their postal codes, was tedious. It might be nice to have a minimal package somewhere that does this for the entire dataset (I stopped at the 56 "states" in the zcta dataset, but there are more!). Solving this problem in an authoritative manner, and putting it on CRAN (if it does not already exist), might be in the public interest.
• The ZCTA dataset changes, but I don't know how frequently. get_zcta_crosswalk was an attempt to facilitate the automation of getting that dataset into an R package, but there are still lots more steps that have to be done. For example, I believe that they will reject my package in its current form because the dataset contains Añasco Municipio, which contains a non-ASCII character (so they will insist that I rename it Anasco Municipo).
• In many ways I view this work as supplanting my choroplethrZip package. I say this for two reasons: (1) that package was always too large for CRAN to host, which limited its reach and (2) the underlying dataset has presumably changed a lot since it was initially published, but I never updated the package. The fact that tigris allows API access to the maps seems to have in many ways obviated choroplethr's approach of storing the maps as packages.

[alamstein]

The zctaCrosswalk package is now on CRAN, so I believe that this issue can be closed.

Here is a blog post announcing and describing the package.

Please note that the package's second vignette (Workflow with tidycensus) has an example of using the package to subset the national ZCTA data from tidycensus::get_acs by county and then map it.

Presumably there could be a performance improvement if tidycensus allowed you to specify a number of states or counties, and then only requested information on them from the API. But I'm not sure if the Census API supports that.