New Connecticut counties, er... regions

This regards those new Connecticut "county equivalents" or "planning regions."
My apologies if this issue has already been covered and I overlooked it.

Some users have helpfully pointed out how data for the new areas can be aggregated from lower-level geographies, such as county subdivisions.
Which is fine, if one has the data at those more detailed geo levels.

But what about the wealth of useful data published at the county levels (such as health indicators)?  How might we convert new data for the "county equivalents" back to the previous CT county structure so it can be used with pre-existing data for analysis, comparison, reporting, trends, etc.  We don't want to leave a big hole for the state of Connecticut until all agencies and companies switch over to the new counties in a year or two.
Or when we switch over to the new CT regions to include in our list of all US counties, how should we convert data for the earlier CT counties to its new CT equivalent geographies?

The only workable solution I can imagine to convert new CT counties to previous counties (and vice versa) is to create an allocation table which shows the percent of each existing CT county in the new regions/equivalents.  The value for allocation would be either area or population.

Anyone have a better idea?  Would this approach give "good enough" estimates of the new counties?

Bert Sperling

  • Sorry to be the bearer of bad news, but in short, there is no general solution that produces reliably "good enough" estimates for all use cases.

    The first thing to try is what you noted: if you can get the needed pre-2022 data for county subdivisions, then aggregate that to the 2022 county equivalents. (Another caveat here: this works smoothly only if your data are counts. It's not so easy to aggregate medians or quotients, etc.)

    A second good option is to allocate in the other direction, from post-2022 county subdivisions to pre-2022 counties.

    Note that for either of the above strategies, I think the most useful resource is the county-to-county-subdivsion crosswalk available through the Bureau's Substantial Changes to Counties and County Equivalent Entites page (which is, unfortunately, not linked to anywhere on the Bureau's 2022 Relationship Files page... not yet anyway).

    If all else fails, then yes, you could create a crosswalk that goes directly from pre-2022 counties to the new county equivalents, or vice versa. But I'd do this only as a last resort. It will always be better if you can allocate from lower-level units: block groups or census tracts would also work well, or even ZIP Code Tabulation Areas, which don't nest within counties, but could still be used to model distributions of population characteristics within counties. If you _do_ use a direct county-to-county-equivalent crosswalk, avoid weighting by areas of intersection if possible. Population weights are significantly better than area weights in most cases, but even then, population weights assume that population characteristics are uniform within each county, i.e., perfectly integrated, demographically and socioeconomically, with the same % minority, % children, % in poverty, etc., in every sub-part of the county. As you can imagine, this assumption can be very wrong! Population weights also don't work well with housing data; if household sizes vary a lot and/or there are many vacant housing units or large group quarters populations, then housing and population distributions can differ substantially, so for housing data, I'd recommend using housing weights if possible.

    Despite these limitations, a county-to-county-equivalent allocation could still be "good enough" in some settings, but it'll always be up to you to decide; it can't generally be guaranteed.

  • This is the general approach I use in Geocorr. You can allocate by area or population, but in practice, I almost never use area.

    Here are a couple of tables I just made. One is a list showing (2020) county and (2022) planning region for each block in CT. The smaller file is a list of allocation factors (afacts), based on 2020 block populations, between counties and planning regions.

    Eventually I'll add the new CT regions to Geocorr, but there are a number of unanswered questions. For example, will CT tracts be renumbered? We're used to the assumption that tract numbers are unique within a county, but not between counties. I haven't checked, but there could now be duplicate tract numbers within a planning region. Conversely, if the tract numbers and geoids will remain the same, we won't be able to derive a tract's planning region by looking at the first five digits of the geoid. 

    [hmm, can't seem to attach a file... I'll post to MCDC site]

    Link to block file:

    Link to allocation factor file:

    From our "georef" collection: 

    EDIT: is correct, allocation is far from perfect -- especially because of the assumption of uniformity. Unfortunately it's sometimes the best we've got to work with.

  • Jonathan and Glenn,
    Thanks very much for your comments on this issue.
    I agree that using lower-level data and aggregating it to the new 2022 county equivalents is the best solution.  When using Census and ACS data, that lower-level detail data is almost always available with a little digging.
    But there's a wealth of other data sets which are only available by county, such as the CDC's BRFSS health statistics and BLS economic data.  In those cases, we'll need to convert data between the old and new Connecticut counties/equivalents, at least until all agencies get on the same page.  

    Glenn, thank you for the allocation tables!  I was in the process of creating my own, but I'll use yours instead.  
    And you bring up some good points about the fallout which will come creating these new counties and their fips codes.
    What a mess.  I wonder if these changes were actually needed or just a whim of the Connecticut legislature.


  • I wonder if these changes were actually needed or just a whim of the Connecticut legislature.

    The FR notice you linked says CT counties "ceased to function as governmental and administrative entities in 1960". I'm not familiar with the history of these planning regions, but it seems they have had substantial administrative function since 2015 or so, though not taxation. They are about the same size as counties. It makes sense.

    I wonder if we're going to see anything similar happening with Rhode Island, which also doesn't have county government -- or is RI too small to warrant second level administration?

  • Bumping up this old thread- does anyone know if the 2018-2022 5 year estimates scheduled to come out next month will use the new 9 "regions/counties" or the old 8 counties?   Sorry if I missed this elsewhere, just trying to plan if I have to do these extra painful steps to get complete US or not with the most current data.

  • The 2022 5-year estimates will use the new regions/counties. 

    FWIW, it took a little digging to find documentation that confirmed this. Ordinarily, I'd check this page ("Geography Boundaries by Year"), which is designed to provide answers to almost all questions about which geographic vintage is used in each ACS data release. But it tells me only that the counties used in 2022 1-year & 2018-2022 5-year ACS data are those in effect as of January 1, 2022, and it doesn't say which version of Connecticut counties were in effect on that date. Ordinarily, I'd check this page ("Substantial Changes to Counties and County Equivalent Entities: 1970-Present") to determine when county definition changes occurred, but for the Connecticut changes, the effective date is not listed on the page. There is, however, a link there to this document, which explains:

    The Census Bureau completed the geographic updates in the spring of 2022.  The Census Bureau will release geospatial products in late summer of 2022. Data users can expect to see Population Estimates and American Community Survey Data in late 2023.

    I also confirmed that uses the new county equivalents in 2022 1-year data.

  • Thanks!  I did a little searching and found this bit here: "ACS 5-year Comparison Profiles for Connecticut county-equivalent geographies will not be available until the 2027 ACS release at which time both non-overlapping periods (2018-2022 v 2023-2027) will finally be based on the new county-equivalent geographies".  So to me this sounds like they will keep the old counties until 2027, but I am still a little unclear.

  • Thanks for the information, Jonathan. Looks like we have some challenges to deal with until the dust settles and everyone gets on the same page, in a couple of years or so.  Slight frown
    And thanks for the question, Ali15.