2018-23 U.S. Homeowners Insurance Non-renewal Data

Version 2: I added a CSV with the state rows stripped out and added to each county so counties can be disambiguated. The code for this operation is included.

"This new report is the culmination of an investigation the Senate Budget Committee launched last fall.  Over the course of the probe, the Committee obtained national county-level non-renewal data from nearly two dozen companies that collectively account for about 65 percent of the national homeowners’ insurance market.  The data cover 249 million total insurance policies over the six-year period beginning in 2018 and ending in 2023.  While sky-rocketing insurance premiums have been well documented, comprehensive non-renewal data of this scope and magnitude has never been publicly available—until now."

From the report: "The Committee ultimately obtained national, county-level non-renewal data from 23 of the 41 companies from which it requested this data. The data cover the years 2018 through 2023, and the companies responding collectively account for approximately 65 percent of the homeowners’ insurance market nationwide. The data released with this Report demonstrate climate change beginning to upend insurance markets around the country."

I am uploading this data here in February 2025 out of an abundance of caution given that climate change related data is currently being scrubbed from U.S. federal websites. In addition, I created a CSV of the data that is in a more usable format, in my opinion. The CSV adds a column for the year instead of separating data on different worksheet tabs, making longitudinal analysis more difficult.

Note that the formatting of the data is still terrible. County data is duplicated in state totals. "AK" is not a U.S. county, it is the state of Alaska, and it is the sum of the next several rows, which are for its counties. If you sum a column, you receive double the "grand total" listed at the bottom due to this duplication. Without states to disambiguoate, counties/districts are not uniquely identified and there are many duplicate county names. One should be able to move states/territories out and use them to disambiguate counties.

To the best of my knowledge, this dataset is in the public domain, as are most U.S. federal government publications.