How We Examined Racial Discrimination in Auto Insurance Prices

Introduction

It has long been observed that the same safe driver in a predominantly minority zip code is often charged higher auto insurance premiums than if he or she lived in a non-minority neighborhood. A nationwide study by the Consumer Federation of America in 2015 found that predominantly African-American neighborhoods pay 70 percent more, on average, for premiums than other areas do. But the insurance industry has long argued that higher premiums in minority neighborhoods are justified by a higher risk of accidents. “Insurance rates are color-blind and solely based on risk,” the Property Casualty Insurers Association of America declared in response to the 2015 study.

We examined that assertion by studying public and private data on auto insurance risk and rates as well as the algorithms used to set prices in different states. We analyzed aggregate risk data by zip code collected by the insurance commissioners of California, Illinois, Missouri and Texas from insurers in their states. We compared that information with liability insurance premiums — the sum of bodily injury and property damage quotes — charged by the largest companies by market share in each of the four states.

We found that some insurers were charging statistically significantly higher premiums in predominantly minority zip codes, on average, than in similarly risky non-minority zip codes. The difference in premiums was especially stark in Illinois, where nearly every insurer showed a disparity at every risk level. In California, Texas and Missouri, we found disparities in the riskiest zip codes.

In Illinois, 33 of 34 companies that we analyzed were charging more than 10 percent higher on average for liability premiums in minority zip codes than in white zip codes on average. We also found that companies from the Allstate, Travelers, Metropolitan, Pekin, and Auto Owners insurance groups were charging 30 percent higher on average in risky minority zip codes than in non-minority zip codes with similar risk. Within the city of Chicago, we found that Berkshire Hathaway (Geico), Progressive, Metropolitan and Farmers showed disparities higher than 10 percent.

In Missouri, we found 21 of 25 companies with pricing disparities that didn’t appear to be justified by risk. Berkshire Hathaway, Liberty Mutual and Auto Club Enterprises all had subsidiaries that were charging 30 percent more on average in risky minority neighborhoods compared with similarly risky majority white neighborhoods.

In Texas, we found 9 of 18 companies charging residents of risky minority zip codes more than 10 percent higher on average than residents of similar non-minority zip codes.

And in California, 8 of 21 companies were charging more than 10 percent higher on average in risky minority zip codes than in similar non-minority zip codes. Liberty Mutual was charging residents of these minority zip codes 32 percent more on average than residents of non-minority zip codes.

How We Acquired the Data

We obtained data from two commercial data providers, Quadrant Information Services and S&P Global Inc. Quadrant provided us with 30 million premium quotes by zip code from the leading insurance companies across the nation. Those quotes reflected rates for forty-four separate profiles of drivers, varying by such factors as gender, age, and family makeup. We used data from the Census’s American Community Survey to make sure that these profiles were a representative slice of the United States.

S&P provided us with rate filing manuals, which describe in detail how each insurer sets rates. Where possible, we used public rate filings to check that the Quadrant premiums were correct.

We also filed public records requests for zip code level risk data with insurance commissioners from the 50 states and Washington D.C. Only California, Illinois, Missouri and Texas provided us with data. The other states all responded that they did not collect such data.

In California, Texas, Missouri and Illinois, we received zip-code level data covering the number of cars insured and payouts by the state’s insurers over the most recent five-year period for which data is available. In Texas and Missouri, this loss data covers 2011 to 2015. California’s most recent data covers 2007 through 2011. In 2011, Illinois changed its data collection requirements, so we only used three years of data, from 2012 through 2014.

In California and Texas, the insurance commissioner provided us with incurred losses, which are claims that were filed in those time periods that have been paid in addition to an estimate of claims that have not been paid. The Illinois and Missouri insurance commissioners collect paid losses – claims that were paid in the calendar year.

The California loss data is adjusted for “credibility” in areas with low population. In that state, we used these weighted numbers.

California’s losses are also capped: they only represent payouts of minimum coverage policies. People buying insurance can choose to buy more or less coverage, and that means that their claim payouts can be more or less depending on their policy. California’s data is “capped” to a single level of coverage in order to even out these differences. In the other three states, we used the raw loss because we did not have access to capped losses.

In Texas, the risk data only include claims and payouts by top insurers, which the state said accounts for approximately 70 percent of the private passenger automobile market. California says that its data covers 99 percent of the market. In Missouri, the risk data covers more than 98 percent of the market. Illinois did not respond to questions on how much of the market their data covers, but their regulations say that reporting requirements are “are applicable to all direct property and liability business written by insurers licensed by the State of Illinois.”

Quadrant’s data covered 1,816 zip codes in California, 2,023 zip codes in Texas, 1,413 zip codes in Illinois and 1,009 zip codes in Missouri.

Previous Work

The question of whether and to what extent disparities in auto insurance premiums between minority neighborhoods and non-minority neighborhoods are due to risk has been widely discussed. Research has been impeded by a lack of public data about risk at a zip code level.

A 1997 study of auto insurance rates in Texas found that “drivers in poor and minority communities were disproportionately rejected by standard insurers and forced into the higher cost non-standard” insurance plans that are designed as a last resort for people who can’t otherwise buy insurance.

In 1998, the National Association of Insurance Commissioners studied a sample of auto and homeowner insurance rates nationwide and found that higher minority population of a zip code was correlated with higher premiums, even after controlling for factors related to risk. The report noted that zip code level risk data was not available.

Also in 1998, researchers at the University of South Carolina examined auto insurance rates in Missouri and did not find evidence of racial discrimination. They used zip code level data, including aggregated insurance company losses per zip code, but did not examine individual companies.

In 2007, researchers at the University of California-Los Angeles examined auto insurance rates in Los Angeles by zip code, and by company. They found that “redlining factors are associated with variations in insurance costs” even after risk factors were taken into account.

Analysis

Insurance companies calculate premiums based on their prediction of how risky a driver will be. In general, an insurer will start with a base rate and increase or decrease that rate based on an insured’s driving record, demographics, location or other factors.

The types of factors used in these risk algorithms vary widely. Some states allow insurers to use factors aside from driving records such as occupation and credit scores as part of their risk calculations.

In order to control for factors outside of geography, we limited our analysis to a single profile – a 30-year-old female safe driver who is a teacher with a bachelor’s degree and excellent credit, with no accidents or moving violations, and who is purchasing standard coverage with the company for the first time. She drives a 2016 Toyota Camry, and has a fifteen-mile commute, and drives around 13,000 miles a year. She is purchasing a policy for $100,000 of property damage coverage and $100,000 to cover medical bills per person up to $300,000 per accident.

We focused on bodily injury and property damage insurance premiums because most states require drivers to purchase them. Within the insurance industry, the combination of bodily injury and property damage coverage is known as liability insurance.

We compared the premiums charged by each insurer in each zip code with the average risk of liability payouts by all insurers in that zip code.

We calculated a ratio for each zip code of payouts divided by the number of insured cars garaged in that zip code. In California, this is called “average loss” and is a measure of how expensive it is on average for an insurer to write a policy in that zip code:

\(Average\ Loss\ =\ \frac{Dollars\ Paid\ Out for\ Liability\ Claims\ by\ Insurers} {Number\ of\ Cars\ Insured\ by\ Insurers}\)

To be sure, average loss will not exactly match each individual insurer’s payouts in a given zip code. But since insurers are not required to publicly divulge information about their own losses, average loss is the best approximation of loss available on a geographic basis.

We then sought to determine whether zip codes with predominantly minority populations were priced differently than zip codes with the same risk and different demographics.

In California, we used the Department of Insurance’s definition of “underserved” zip codes. The department identifies 145 “underserved” zip codes that have a minority population greater than 66 percent, a household income below the median, and higher than average proportion of uninsured motorists.

In Texas, Illinois and Missouri, we used data from the Census Department’s American Community Survey 2014 five-year estimates to identify minority zip codes. In Texas, we followed California’s definition of defined minority zip codes as having greater than 66 percent minority population. In Missouri and Illinois, we defined minority zip codes as having greater than 50 percent minority population in order to have a sufficiently large sample size.

We then created separate statistical models for minority and non-minority zip codes to examine differences in pricing. We fit smoothing splines to each predictor: in this case, Average Loss. Because insurance companies do not publish their underlying modeling procedures, we used these splines for their flexibility. This is how we defined each model:

\(Bodily\ Injury\ Premium = Spline(Average\ Loss) + e\)

We limited our analysis to zip codes with average losses ranging from $25 to $400, except in Illinois where we limited the window from $50 to $300 in an effort to remove outliers. Here are the model predictions:

In Illinois, we found that premiums were much higher on average in minority zip codes compared to non-minority zip codes with similar risk. In fact, many insurers’ rates showed little association to risk.

In the other three states, we saw a similar disparity between minority and non-minority zip codes but only in those with high average losses. We calculated average premiums for minority and non-minority zip codes using predictions from the spline models. This allowed us to calculate a ratio for every company in a state:

\(Ratio = \frac{Average\ Minority\ Prediction}{Average\ NonMinority\ Prediction}\)

We only calculated these ratios for areas with loss measures greater than $250, except in Illinois where we lowered this threshold to $200. We chose these thresholds to ensure that the average loss of the minority zip codes was similar to the average loss of non-minority zip codes. We found that our conclusions were similar when we used linear models instead of splines.

When we repeated the analysis across all 44 profiles, the results were similar.

The results of this analysis are provided in the table below. Ratios reflecting a disparity of greater than 10 percent are in red. Ratios higher than 5 percent are in yellow.

Group Company Name CA Chicago, IL IL MO TX
ALLSTATE INS GRP Allstate Fire & Cas Ins Co - 1.042 1.226 1.139 1.124
ALLSTATE INS GRP Allstate Ind Co 0.981 1.015 1.42 1.171 1.104
ALLSTATE INS GRP Northbrook Ind Co 0.981 - - - -
AMERICAN FAMILY INS GRP American Family Mut Ins Co - 1 1.195 1.271 -
AMERICAN FAMILY INS GRP American Standard Ins Co of WI - 1 1.184 1.27 -
Auto Club Enterprises Ins Grp Automobile Club Interins Exch - - - 1.305 -
AUTO OWNERS GRP Owners Ins Co - 1.046 1.55 - -
BERKSHIRE HATHAWAY GRP Geico Cas Co 1.034 1.096 1.16 1.248 -
BERKSHIRE HATHAWAY GRP Geico Gen Ins Co 1.031 1.189 1.21 - 1.102
BERKSHIRE HATHAWAY GRP Geico Ind Co 1.031 1.18 1.219 1.274 1.1
BERKSHIRE HATHAWAY GRP Government Employees Ins Co - 1.189 1.21 1.406 1.102
Country Ins & Fin Serv Grp Country Mut Ins Co - 1 1.149 - -
Country Ins & Fin Serv Grp Country Pref Ins Co - 1 1.152 - -
ERIE INS GRP Erie Ins Co - 1.006 1.244 - -
ERIE INS GRP Erie Ins Exch - 1.006 1.246 - -
FARMERS INS GRP 21st Century Centennial Ins Co - - - 1.216 1.079
FARMERS INS GRP Farmers Ins Co Inc - - - 1.197 -
FARMERS INS GRP Farmers Ins Exch 1.073 - - - -
FARMERS INS GRP Farmers TX Cnty Mut Ins Co - - - - 1.054
FARMERS INS GRP Illinois Farmers Ins Co - 1.107 1.275 - -
FARMERS INS GRP Mid Century Ins Co 1.073 - - - -
HARTFORD FIRE & CAS GRP Trumbull Ins Co - 1.007 1.13 - -
HARTFORD FIRE & CAS GRP Twin City Fire Ins Co Co - - - 1.133 -
LIBERTY MUT GRP First Liberty Ins Corp - 1.003 1.235 - -
LIBERTY MUT GRP Liberty Cnty Mut Ins Co - - - - 1.053
LIBERTY MUT GRP Liberty Mut Fire Ins Co 1.327 1.003 1.234 - -
LIBERTY MUT GRP Safeco Ins Co Of Amer 1.115 - - - -
LIBERTY MUT GRP Safeco Ins Co Of IL - 1.044 1.174 1.336 -
MERCURY GEN GRP California Automobile Ins Co 1.019 - - - -
MERCURY GEN GRP Mercury Ins Co 1.02 - - - -
METROPOLITAN GRP Economy Preferred Ins Co - 1.002 1.406 - -
METROPOLITAN GRP Metropolitan Cas Ins Co - 1.005 1.397 - -
METROPOLITAN GRP Metropolitan Grp Prop & Cas Ins Co - 1.152 1.327 - -
METROPOLITAN GRP Metropolitan Prop & Cas Ins Co - 1.044 1.55 - -
NATIONWIDE CORP GRP Allied Prop & Cas Ins Co 1.14 - - 1.246 -
NATIONWIDE CORP GRP Amco Ins Co 1.14 - - - -
NATIONWIDE CORP GRP Colonial Cnty Mut Ins Co - - - - 1.152
NATIONWIDE CORP GRP Nationwide Affinity Co of Amer - - - 1.242 -
NATIONWIDE CORP GRP Nationwide Ins Co Of Amer 1.099 - - 1.22 -
NATIONWIDE CORP GRP Nationwide Mut Ins Co - - - - 1.081
PEKIN INS GRP Farmers Automobile Ins Assoc - 1.051 2.097 - -
PROGRESSIVE GRP Progressive Advanced Ins Co - - - 1.177 -
PROGRESSIVE GRP Progressive Cas Ins Co - - - 1.265 -
PROGRESSIVE GRP Progressive Cnty Mut Ins Co - - - - 1.014
PROGRESSIVE GRP Progressive Direct Ins Co - 1.111 1.156 - -
PROGRESSIVE GRP Progressive Northern Ins Co - 1.153 1.21 - -
PROGRESSIVE GRP Progressive Northwestern Ins Co - - - 1.291 -
PROGRESSIVE GRP Progressive Universal Ins Co - 1.11 1.149 - -
PROGRESSIVE GRP Progressive West Ins Co 1.073 - - - -
PROGRESSIVE GRP United Financial Cas Co 1.05 - - - -
SHELTER INS GRP Shelter Mut Ins Co - - - 1.286 -
STATE FARM GRP State Farm Cnty Mut Ins Co Of TX - - - - 1.059
STATE FARM GRP State Farm Fire & Cas Co - 0.994 1.19 1.283 1.064
STATE FARM GRP State Farm Mut Auto Ins Co 1.063 0.995 1.19 1.283 1.064
Travelers Grp Travelers Commercial Ins Co - 1.06 1.391 - -
Travelers Grp Travelers Home & Marine Ins Co - 1.06 1.393 - -
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co 1.189 1.034 1.092 0.994 1.112
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn 1.189 1.029 1.134 1.042 1.105
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co 1.189 1.035 1.113 0.992 1.108
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co 1.178 1.053 1.142 1 1.088

Higher ratios indicate larger difference between prices in minority neighborhoods to those in similarly risky non-minority neighborhoods, on average.

The ratios are the starkest in Illinois, which is one of the least regulated states for auto insurance. But some insurers also have high ratios in California.

An Examination of Affordability

We also used the Quadrant premium data to examine auto insurance affordability by zip code.

We used our 30 million auto insurance quotes to calculate a ratio of annual premiums as a share of median household income. These annual premiums include liability insurance as well as more optional coverage like collision insurance, comprehensive insurance and other premiums that are designed to protect the car owner.

We found that auto insurance quotes were more than twice the share of median household income in minority-majority zip codes (11 percent) on average when compared with households in majority white neighborhoods (5 percent).

The U.S. Treasury Department has defined auto insurance as affordable if it costs 2 percent or less of household income.

A Look at California

In California’s highly-regulated insurance market, insurers also must give less weight to location than to driving record. Insurers are allowed two different factors for location-based risk: one for frequency of claims — number of claims divided by policies written in a year — and another for severity of claims — insurance payouts divided by the number of claims.

California regulations require that company specific data for frequency and severity be “credible”, where “credibility” is defined in terms of statistical accuracy and precision as mandated by state regulations. These rules dictate that a company has received claims from a minimum of 6000 accidents in a zip code over a certain time period, and that the statistical accuracy of this company-chosen time period meets the state’s requirements. If the data does not meet those thresholds, companies must combine their data with the state’s.

However in areas where the combined data still isn’t credible, companies are able to use a provision of the law that allows insurers to group non-credible zip codes together in order to meet that requirement.

The companies with the highest ratios in our analysis disclosed in their public filings that they use this method. Some companies disclose that they use clustering algorithms such as K-Means to draw these areas of contiguous zip codes. Others don’t disclose their algorithms.

Often, the companies use these algorithms to group these zip codes with nearby areas that also have few policy-holders, according to insurers’ rate filings. This potentially causes higher risk zip codes in these areas to be assigned the average risk for the region, and in many cases, lower risk than what the state estimates.

A Look at Chicago

In an effort to control disparate pricing, Illinois passed a law in 1972 requiring each car insurance company to charge the same price for bodily injury insurance throughout Chicago. However, each company’s liability insurance premiums can still vary within the city because drivers are required to buy policies that include both bodily injury and property damage coverage, and there is no similar restriction on the variation of prices for property damage.

Because of Chicago’s unique situation, we analyzed Chicago zip codes separately from the rest of Illinois. We found eight companies that were charging higher prices on average for risky zip codes in predominantly minority neighborhoods compared to similarly risky non-minority zip codes within Chicago (see table above).

Limitations

A limitation of this study is that in some cases the spline models we used had relatively low R-squared values (see model fits below). Because these values were so low, overall averages may not be precise estimates.

In order to be confident in our results, we also calculated the medians of the fitted values, the means of the actual values and the medians of the actual values. We compared this extra information to the means we calculated in the study and found that they were similar.

However, these low R-squared values also point to another conclusion. The fact that the two distributions don’t line up suggests that insurance companies’ premiums are weakly related to risk.

Another limitation is that companies may have a different distribution of risk than the state’s aggregate risk numbers. But it is unlikely that those differences would result in a consistent pattern of higher prices for minority neighborhoods.

While companies do not generally publicly disclose their underlying risk numbers, we found a 2015 rate filing in California from Nationwide that did include its internal loss numbers. This filing showed that Nationwide’s internal numbers were different than the state’s, but it also showed that Nationwide’s internal risk numbers were statistically significantly correlated with the state’s (Pearson: 0.43, 95% CI:0.39, 0.47).

We ran the same analysis as above on premiums from Nationwide’s Allied subsidiary and Nationwide’s internal loss and found that the disparity between premiums in risky underserved areas was 21 percent higher on average than in similarly risky non-underserved areas.That disparity is slightly greater than the 14 percent we found when comparing Allied premiums to overall state risk data.

Model Diagnostics

Illinois Model

Name Company Name Minority Model P Value of the State Risk Term R Squared
ALLSTATE INS GRP Allstate Fire & Cas Ins Co TRUE 0.023 0.129
ALLSTATE INS GRP Allstate Fire & Cas Ins Co FALSE < 0.001 0.227
ALLSTATE INS GRP Allstate Ind Co TRUE 0.11 0.060
ALLSTATE INS GRP Allstate Ind Co FALSE < 0.001 0.215
AMERICAN FAMILY INS GRP American Family Mut Ins Co TRUE 0.031 0.090
AMERICAN FAMILY INS GRP American Family Mut Ins Co FALSE < 0.001 0.273
AMERICAN FAMILY INS GRP American Standard Ins Co of WI TRUE 0.03 0.091
AMERICAN FAMILY INS GRP American Standard Ins Co of WI FALSE < 0.001 0.207
AUTO OWNERS GRP Owners Ins Co TRUE 0.067 0.075
AUTO OWNERS GRP Owners Ins Co FALSE < 0.001 0.213
BERKSHIRE HATHAWAY GRP Geico Cas Co TRUE 0.03 0.108
BERKSHIRE HATHAWAY GRP Geico Cas Co FALSE < 0.001 0.316
BERKSHIRE HATHAWAY GRP Geico Gen Ins Co TRUE 0.076 0.067
BERKSHIRE HATHAWAY GRP Geico Gen Ins Co FALSE < 0.001 0.177
BERKSHIRE HATHAWAY GRP Geico Ind Co TRUE 0.065 0.077
BERKSHIRE HATHAWAY GRP Geico Ind Co FALSE < 0.001 0.210
BERKSHIRE HATHAWAY GRP Government Employees Ins Co TRUE 0.076 0.067
BERKSHIRE HATHAWAY GRP Government Employees Ins Co FALSE < 0.001 0.177
Country Ins & Fin Serv Grp Country Mut Ins Co TRUE 0.051 0.091
Country Ins & Fin Serv Grp Country Mut Ins Co FALSE < 0.001 0.272
Country Ins & Fin Serv Grp Country Pref Ins Co TRUE 0.056 0.089
Country Ins & Fin Serv Grp Country Pref Ins Co FALSE < 0.001 0.276
ERIE INS GRP Erie Ins Co TRUE 0.02 0.125
ERIE INS GRP Erie Ins Co FALSE < 0.001 0.308
ERIE INS GRP Erie Ins Exch TRUE 0.018 0.128
ERIE INS GRP Erie Ins Exch FALSE < 0.001 0.307
FARMERS INS GRP Illinois Farmers Ins Co TRUE 0.017 0.114
FARMERS INS GRP Illinois Farmers Ins Co FALSE < 0.001 0.131
HARTFORD FIRE & CAS GRP Trumbull Ins Co TRUE 0.004 0.135
HARTFORD FIRE & CAS GRP Trumbull Ins Co FALSE < 0.001 0.288
LIBERTY MUT GRP First Liberty Ins Corp TRUE 0.077 0.079
LIBERTY MUT GRP First Liberty Ins Corp FALSE < 0.001 0.246
LIBERTY MUT GRP Liberty Mut Fire Ins Co TRUE 0.074 0.080
LIBERTY MUT GRP Liberty Mut Fire Ins Co FALSE < 0.001 0.246
LIBERTY MUT GRP Safeco Ins Co Of IL TRUE 0.01 0.138
LIBERTY MUT GRP Safeco Ins Co Of IL FALSE < 0.001 0.262
METROPOLITAN GRP Economy Preferred Ins Co TRUE 0.16 0.061
METROPOLITAN GRP Economy Preferred Ins Co FALSE < 0.001 0.169
METROPOLITAN GRP Metropolitan Cas Ins Co TRUE 0.14 0.065
METROPOLITAN GRP Metropolitan Cas Ins Co FALSE < 0.001 0.179
METROPOLITAN GRP Metropolitan Grp Prop & Cas Ins Co TRUE 0.51 0.017
METROPOLITAN GRP Metropolitan Grp Prop & Cas Ins Co FALSE < 0.001 0.254
METROPOLITAN GRP Metropolitan Prop & Cas Ins Co TRUE 0.1 0.076
METROPOLITAN GRP Metropolitan Prop & Cas Ins Co FALSE < 0.001 0.155
PEKIN INS GRP Farmers Automobile Ins Assoc TRUE 0.12 0.014
PEKIN INS GRP Farmers Automobile Ins Assoc FALSE < 0.001 0.030
PROGRESSIVE GRP Progressive Direct Ins Co TRUE 0.01 0.090
PROGRESSIVE GRP Progressive Direct Ins Co FALSE < 0.001 0.319
PROGRESSIVE GRP Progressive Northern Ins Co TRUE 0.009 0.092
PROGRESSIVE GRP Progressive Northern Ins Co FALSE < 0.001 0.325
PROGRESSIVE GRP Progressive Universal Ins Co TRUE 0.009 0.091
PROGRESSIVE GRP Progressive Universal Ins Co FALSE < 0.001 0.323
STATE FARM GRP State Farm Fire & Cas Co TRUE 0.01 0.137
STATE FARM GRP State Farm Fire & Cas Co FALSE < 0.001 0.362
STATE FARM GRP State Farm Mut Auto Ins Co TRUE 0.009 0.138
STATE FARM GRP State Farm Mut Auto Ins Co FALSE < 0.001 0.362
Travelers Grp Travelers Commercial Ins Co TRUE 0.076 0.081
Travelers Grp Travelers Commercial Ins Co FALSE < 0.001 0.126
Travelers Grp Travelers Home & Marine Ins Co TRUE 0.077 0.080
Travelers Grp Travelers Home & Marine Ins Co FALSE < 0.001 0.126
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co TRUE 0.4 0.009
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co FALSE < 0.001 0.166
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn TRUE 0.12 0.036
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn FALSE < 0.001 0.132
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co TRUE 0.34 0.028
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co FALSE < 0.001 0.128
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co TRUE 0.1 0.054
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co FALSE < 0.001 0.229

Chicago Model

Name Company Name Minority Model P Value of the State Risk Term R Squared
ALLSTATE INS GRP Allstate Fire & Cas Ins Co TRUE 0.34 -0.002
ALLSTATE INS GRP Allstate Fire & Cas Ins Co FALSE 0.5 0.086
ALLSTATE INS GRP Allstate Ind Co TRUE 0.57 -0.024
ALLSTATE INS GRP Allstate Ind Co FALSE 0.9 -0.058
AMERICAN FAMILY INS GRP American Family Mut Ins Co TRUE 0.6 -0.025
AMERICAN FAMILY INS GRP American Family Mut Ins Co FALSE - -Inf
AMERICAN FAMILY INS GRP American Standard Ins Co of WI TRUE 0.6 -0.025
AMERICAN FAMILY INS GRP American Standard Ins Co of WI FALSE - -Inf
AUTO OWNERS GRP Owners Ins Co TRUE 0.34 -0.002
AUTO OWNERS GRP Owners Ins Co FALSE 0.43 -0.020
BERKSHIRE HATHAWAY GRP Geico Cas Co TRUE 0.21 0.021
BERKSHIRE HATHAWAY GRP Geico Cas Co FALSE 0.21 0.038
BERKSHIRE HATHAWAY GRP Geico Gen Ins Co TRUE 0.84 -0.034
BERKSHIRE HATHAWAY GRP Geico Gen Ins Co FALSE 0.22 0.033
BERKSHIRE HATHAWAY GRP Geico Ind Co TRUE 0.92 -0.035
BERKSHIRE HATHAWAY GRP Geico Ind Co FALSE 0.31 0.006
BERKSHIRE HATHAWAY GRP Government Employees Ins Co TRUE 0.84 -0.034
BERKSHIRE HATHAWAY GRP Government Employees Ins Co FALSE 0.22 0.033
Country Ins & Fin Serv Grp Country Mut Ins Co TRUE - -
Country Ins & Fin Serv Grp Country Mut Ins Co FALSE - -Inf
Country Ins & Fin Serv Grp Country Pref Ins Co TRUE > 0.99 -Inf
Country Ins & Fin Serv Grp Country Pref Ins Co FALSE - -Inf
ERIE INS GRP Erie Ins Co TRUE 0.34 -0.002
ERIE INS GRP Erie Ins Co FALSE 0.48 -0.027
ERIE INS GRP Erie Ins Exch TRUE 0.34 -0.002
ERIE INS GRP Erie Ins Exch FALSE 0.48 -0.027
FARMERS INS GRP Illinois Farmers Ins Co TRUE 0.23 0.017
FARMERS INS GRP Illinois Farmers Ins Co FALSE 0.83 -0.056
HARTFORD FIRE & CAS GRP Trumbull Ins Co TRUE 0.68 -0.029
HARTFORD FIRE & CAS GRP Trumbull Ins Co FALSE 0.09 0.110
LIBERTY MUT GRP First Liberty Ins Corp TRUE 0.6 -0.025
LIBERTY MUT GRP First Liberty Ins Corp FALSE 0.48 -0.027
LIBERTY MUT GRP Liberty Mut Fire Ins Co TRUE 0.6 -0.025
LIBERTY MUT GRP Liberty Mut Fire Ins Co FALSE 0.48 -0.027
LIBERTY MUT GRP Safeco Ins Co Of IL TRUE 0.15 0.040
LIBERTY MUT GRP Safeco Ins Co Of IL FALSE 0.095 0.329
METROPOLITAN GRP Economy Preferred Ins Co TRUE 0.021 0.145
METROPOLITAN GRP Economy Preferred Ins Co FALSE 0.54 0.079
METROPOLITAN GRP Metropolitan Cas Ins Co TRUE 0.017 0.156
METROPOLITAN GRP Metropolitan Cas Ins Co FALSE 0.54 0.081
METROPOLITAN GRP Metropolitan Grp Prop & Cas Ins Co TRUE 0.26 0.011
METROPOLITAN GRP Metropolitan Grp Prop & Cas Ins Co FALSE 0.28 0.014
METROPOLITAN GRP Metropolitan Prop & Cas Ins Co TRUE 0.51 0.022
METROPOLITAN GRP Metropolitan Prop & Cas Ins Co FALSE 0.1 0.327
PEKIN INS GRP Farmers Automobile Ins Assoc TRUE 0.34 -0.002
PEKIN INS GRP Farmers Automobile Ins Assoc FALSE 0.48 -0.027
PROGRESSIVE GRP Progressive Direct Ins Co TRUE 0.2 0.078
PROGRESSIVE GRP Progressive Direct Ins Co FALSE 0.3 0.007
PROGRESSIVE GRP Progressive Northern Ins Co TRUE 0.2 0.079
PROGRESSIVE GRP Progressive Northern Ins Co FALSE 0.31 0.006
PROGRESSIVE GRP Progressive Universal Ins Co TRUE 0.19 0.081
PROGRESSIVE GRP Progressive Universal Ins Co FALSE 0.31 0.005
STATE FARM GRP State Farm Fire & Cas Co TRUE > 0.99 -Inf
STATE FARM GRP State Farm Fire & Cas Co FALSE 0.1 0.278
STATE FARM GRP State Farm Mut Auto Ins Co TRUE - -
STATE FARM GRP State Farm Mut Auto Ins Co FALSE 0.089 0.296
Travelers Grp Travelers Commercial Ins Co TRUE 0.41 0.075
Travelers Grp Travelers Commercial Ins Co FALSE 0.008 0.307
Travelers Grp Travelers Home & Marine Ins Co TRUE 0.41 0.073
Travelers Grp Travelers Home & Marine Ins Co FALSE 0.008 0.306
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co TRUE 0.44 -0.013
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co FALSE 0.83 -0.056
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn TRUE 0.31 0.002
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn FALSE 0.29 0.250
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co TRUE 0.36 -0.005
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co FALSE 0.98 -0.059
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co TRUE 0.35 0.106
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co FALSE 0.47 0.122

Missouri Model

Name Company Name Minority Model P Value of the State Risk Term R Squared
ALLSTATE INS GRP Allstate Fire & Cas Ins Co TRUE 0.006 0.151
ALLSTATE INS GRP Allstate Fire & Cas Ins Co FALSE < 0.001 0.216
ALLSTATE INS GRP Allstate Ind Co TRUE 0.009 0.185
ALLSTATE INS GRP Allstate Ind Co FALSE < 0.001 0.211
AMERICAN FAMILY INS GRP American Family Mut Ins Co TRUE < 0.001 0.315
AMERICAN FAMILY INS GRP American Family Mut Ins Co FALSE < 0.001 0.320
AMERICAN FAMILY INS GRP American Standard Ins Co of WI TRUE < 0.001 0.315
AMERICAN FAMILY INS GRP American Standard Ins Co of WI FALSE < 0.001 0.320
Auto Club Enterprises Ins Grp Automobile Club Interins Exch TRUE < 0.001 0.374
Auto Club Enterprises Ins Grp Automobile Club Interins Exch FALSE < 0.001 0.342
BERKSHIRE HATHAWAY GRP Geico Cas Co TRUE < 0.001 0.374
BERKSHIRE HATHAWAY GRP Geico Cas Co FALSE < 0.001 0.283
BERKSHIRE HATHAWAY GRP Geico Ind Co TRUE < 0.001 0.340
BERKSHIRE HATHAWAY GRP Geico Ind Co FALSE < 0.001 0.295
BERKSHIRE HATHAWAY GRP Government Employees Ins Co TRUE < 0.001 0.287
BERKSHIRE HATHAWAY GRP Government Employees Ins Co FALSE < 0.001 0.194
FARMERS INS GRP 21st Century Centennial Ins Co TRUE < 0.001 0.369
FARMERS INS GRP 21st Century Centennial Ins Co FALSE < 0.001 0.329
FARMERS INS GRP Farmers Ins Co Inc TRUE < 0.001 0.258
FARMERS INS GRP Farmers Ins Co Inc FALSE < 0.001 0.220
HARTFORD FIRE & CAS GRP Twin City Fire Ins Co Co TRUE < 0.001 0.379
HARTFORD FIRE & CAS GRP Twin City Fire Ins Co Co FALSE < 0.001 0.294
LIBERTY MUT GRP Safeco Ins Co Of IL TRUE < 0.001 0.357
LIBERTY MUT GRP Safeco Ins Co Of IL FALSE < 0.001 0.265
NATIONWIDE CORP GRP Allied Prop & Cas Ins Co TRUE < 0.001 0.514
NATIONWIDE CORP GRP Allied Prop & Cas Ins Co FALSE < 0.001 0.332
NATIONWIDE CORP GRP Nationwide Affinity Co of Amer TRUE < 0.001 0.487
NATIONWIDE CORP GRP Nationwide Affinity Co of Amer FALSE < 0.001 0.344
NATIONWIDE CORP GRP Nationwide Ins Co Of Amer TRUE < 0.001 0.449
NATIONWIDE CORP GRP Nationwide Ins Co Of Amer FALSE < 0.001 0.321
PROGRESSIVE GRP Progressive Advanced Ins Co TRUE < 0.001 0.357
PROGRESSIVE GRP Progressive Advanced Ins Co FALSE < 0.001 0.340
PROGRESSIVE GRP Progressive Cas Ins Co TRUE < 0.001 0.358
PROGRESSIVE GRP Progressive Cas Ins Co FALSE < 0.001 0.341
PROGRESSIVE GRP Progressive Northwestern Ins Co TRUE < 0.001 0.351
PROGRESSIVE GRP Progressive Northwestern Ins Co FALSE < 0.001 0.337
SHELTER INS GRP Shelter Mut Ins Co TRUE < 0.001 0.342
SHELTER INS GRP Shelter Mut Ins Co FALSE < 0.001 0.365
STATE FARM GRP State Farm Fire & Cas Co TRUE < 0.001 0.312
STATE FARM GRP State Farm Fire & Cas Co FALSE < 0.001 0.277
STATE FARM GRP State Farm Mut Auto Ins Co TRUE < 0.001 0.313
STATE FARM GRP State Farm Mut Auto Ins Co FALSE < 0.001 0.277
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co TRUE 0.33 0.000
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co FALSE < 0.001 0.027
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn TRUE 0.86 -0.018
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn FALSE 0.98 -0.001
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co TRUE 0.31 0.002
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co FALSE < 0.001 0.026
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co TRUE 0.51 -0.014
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co FALSE 0.001 0.016

Texas Model

Name Company Name Minority Model P Value of the State Risk Term R Squared
ALLSTATE INS GRP Allstate Fire & Cas Ins Co TRUE < 0.001 0.526
ALLSTATE INS GRP Allstate Fire & Cas Ins Co FALSE < 0.001 0.315
ALLSTATE INS GRP Allstate Ind Co TRUE < 0.001 0.510
ALLSTATE INS GRP Allstate Ind Co FALSE < 0.001 0.495
BERKSHIRE HATHAWAY GRP Geico Gen Ins Co TRUE < 0.001 0.435
BERKSHIRE HATHAWAY GRP Geico Gen Ins Co FALSE < 0.001 0.373
BERKSHIRE HATHAWAY GRP Geico Ind Co TRUE < 0.001 0.412
BERKSHIRE HATHAWAY GRP Geico Ind Co FALSE < 0.001 0.348
BERKSHIRE HATHAWAY GRP Government Employees Ins Co TRUE < 0.001 0.435
BERKSHIRE HATHAWAY GRP Government Employees Ins Co FALSE < 0.001 0.373
FARMERS INS GRP 21st Century Centennial Ins Co TRUE < 0.001 0.441
FARMERS INS GRP 21st Century Centennial Ins Co FALSE < 0.001 0.367
FARMERS INS GRP Farmers TX Cnty Mut Ins Co TRUE < 0.001 0.630
FARMERS INS GRP Farmers TX Cnty Mut Ins Co FALSE < 0.001 0.543
LIBERTY MUT GRP Liberty Cnty Mut Ins Co TRUE < 0.001 0.380
LIBERTY MUT GRP Liberty Cnty Mut Ins Co FALSE < 0.001 0.331
NATIONWIDE CORP GRP Colonial Cnty Mut Ins Co TRUE 0.002 0.037
NATIONWIDE CORP GRP Colonial Cnty Mut Ins Co FALSE < 0.001 0.020
NATIONWIDE CORP GRP Nationwide Mut Ins Co TRUE < 0.001 0.359
NATIONWIDE CORP GRP Nationwide Mut Ins Co FALSE < 0.001 0.320
PROGRESSIVE GRP Progressive Cnty Mut Ins Co TRUE < 0.001 0.069
PROGRESSIVE GRP Progressive Cnty Mut Ins Co FALSE 0.019 0.009
STATE FARM GRP State Farm Cnty Mut Ins Co Of TX TRUE < 0.001 0.587
STATE FARM GRP State Farm Cnty Mut Ins Co Of TX FALSE < 0.001 0.496
STATE FARM GRP State Farm Fire & Cas Co TRUE < 0.001 0.601
STATE FARM GRP State Farm Fire & Cas Co FALSE < 0.001 0.518
STATE FARM GRP State Farm Mut Auto Ins Co TRUE < 0.001 0.601
STATE FARM GRP State Farm Mut Auto Ins Co FALSE < 0.001 0.518
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co TRUE < 0.001 0.434
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co FALSE < 0.001 0.431
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn TRUE < 0.001 0.438
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn FALSE < 0.001 0.452
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co TRUE < 0.001 0.465
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co FALSE < 0.001 0.427
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co TRUE < 0.001 0.527
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co FALSE < 0.001 0.485

California Model

Name Company Name Minority Model P Value of the State Risk Term R Squared
ALLSTATE INS GRP Allstate Ind Co TRUE < 0.001 0.850
ALLSTATE INS GRP Allstate Ind Co FALSE < 0.001 0.736
ALLSTATE INS GRP Northbrook Ind Co TRUE < 0.001 0.851
ALLSTATE INS GRP Northbrook Ind Co FALSE < 0.001 0.739
BERKSHIRE HATHAWAY GRP Geico Cas Co TRUE < 0.001 0.824
BERKSHIRE HATHAWAY GRP Geico Cas Co FALSE < 0.001 0.740
BERKSHIRE HATHAWAY GRP Geico Gen Ins Co TRUE < 0.001 0.831
BERKSHIRE HATHAWAY GRP Geico Gen Ins Co FALSE < 0.001 0.750
BERKSHIRE HATHAWAY GRP Geico Ind Co TRUE < 0.001 0.831
BERKSHIRE HATHAWAY GRP Geico Ind Co FALSE < 0.001 0.750
FARMERS INS GRP 21st Century Ins Co TRUE < 0.001 0.671
FARMERS INS GRP 21st Century Ins Co FALSE < 0.001 0.580
FARMERS INS GRP Farmers Ins Exch TRUE < 0.001 0.833
FARMERS INS GRP Farmers Ins Exch FALSE < 0.001 0.752
FARMERS INS GRP Mid Century Ins Co TRUE < 0.001 0.833
FARMERS INS GRP Mid Century Ins Co FALSE < 0.001 0.752
LIBERTY MUT GRP Liberty Mut Fire Ins Co TRUE < 0.001 0.706
LIBERTY MUT GRP Liberty Mut Fire Ins Co FALSE < 0.001 0.695
LIBERTY MUT GRP Safeco Ins Co Of Amer TRUE < 0.001 0.864
LIBERTY MUT GRP Safeco Ins Co Of Amer FALSE < 0.001 0.713
MERCURY GEN GRP California Automobile Ins Co TRUE < 0.001 0.856
MERCURY GEN GRP California Automobile Ins Co FALSE < 0.001 0.733
MERCURY GEN GRP Mercury Ins Co TRUE < 0.001 0.857
MERCURY GEN GRP Mercury Ins Co FALSE < 0.001 0.733
NATIONWIDE CORP GRP Allied Prop & Cas Ins Co TRUE < 0.001 0.855
NATIONWIDE CORP GRP Allied Prop & Cas Ins Co FALSE < 0.001 0.757
NATIONWIDE CORP GRP Amco Ins Co TRUE < 0.001 0.855
NATIONWIDE CORP GRP Amco Ins Co FALSE < 0.001 0.757
NATIONWIDE CORP GRP Nationwide Ins Co Of Amer TRUE < 0.001 0.788
NATIONWIDE CORP GRP Nationwide Ins Co Of Amer FALSE < 0.001 0.754
PROGRESSIVE GRP Progressive West Ins Co TRUE < 0.001 0.876
PROGRESSIVE GRP Progressive West Ins Co FALSE < 0.001 0.772
PROGRESSIVE GRP United Financial Cas Co TRUE < 0.001 0.752
PROGRESSIVE GRP United Financial Cas Co FALSE < 0.001 0.752
STATE FARM GRP State Farm Mut Auto Ins Co TRUE < 0.001 0.838
STATE FARM GRP State Farm Mut Auto Ins Co FALSE < 0.001 0.781
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co TRUE < 0.001 0.544
UNITED SERV AUTOMOBILE ASSN GRP Garrison Prop & Cas Ins Co FALSE < 0.001 0.328
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn TRUE < 0.001 0.543
UNITED SERV AUTOMOBILE ASSN GRP United Serv Automobile Assn FALSE < 0.001 0.328
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co TRUE < 0.001 0.544
UNITED SERV AUTOMOBILE ASSN GRP USAA Cas Ins Co FALSE < 0.001 0.328
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co TRUE < 0.001 0.548
UNITED SERV AUTOMOBILE ASSN GRP USAA Gen Ind Co FALSE < 0.001 0.366

Author photo

Julia Angwin is a senior reporter at ProPublica. From 2000 to 2013, she was a reporter at The Wall Street Journal, where she led a privacy investigative team that was a finalist for a Pulitzer Prize in Explanatory Reporting in 2011 and won a Gerald Loeb Award in 2010.

Author photo

Jeff Larson is a news applications developer at ProPublica. He is a winner of the Livingston Award for the 2011 series Redistricting: How Powerful Interests are Drawing You Out of a Vote.

Author photo

Lauren Kirchner is a senior reporting fellow at ProPublica. She has covered digital security and press freedom issues for the Columbia Journalism Review, and crime and criminal justice for Pacific Standard magazine.

Surya Mattu was a contributing researcher at ProPublica.

Production by Sisi Wei.