Dubious CJCJ Report Claims that Republican Counties Suffer from More Violent Crime
California, like the rest of the country, suffered a major increase in homicide in 2020. The Center on Juvenile and Criminal Justice (CJCJ) has released a report by Mike Males that presents differences between Republican- and Democratic-voting counties (identified by majority voting Democrat in 2020) in terms of homicide rates. He argues that Republican counties tend to have higher homicide rates, stating that: “the clearest difference between areas that have cut crime substantially and those suffering the worst crime trends and rates is not geographic nor demographic, but how they vote – Republican versus Democratic.” However, this report ignores a number of important variables that could be obscuring this finding. Namely, Republican-voting counties tend to be more rural, suburban, and overall have a lower median income when compared with Democratic-voting counterparts, all of which could also impact homicide rates. The classic phrase to remember here is: “correlation does not equal causation.”
It is perplexing why a researcher would compare two groups that are vastly different from each other without attempting to control for outside factors that might differ between groups, such as geography or population size. As a result, the groups are not entirely comparable because they are not similar enough; in this situation researchers would need to apply adequate statistical controls to account for these differences, something that is missing from this analysis. Ideally, a well-conducted study would attempt to control for all factors that differ between counties, except for the political affiliation, i.e., the main variable of interest.
Males examined 23 of California’s Republican-voting counties and compared aggregate rates with 35 Democratic-voting counties. The report displays aggregate homicide rates (per 100,000) for Republican counties vs. Democratic counties from 1990-2020, and percent changes are compared between 1990 and 2020. The author reports that in 2020, Republican-voting counties had a homicide rate of 6.9 per 100,000 residents compared with 5.4 per 100,000 for Democratic-voting counties. This is compared to 1990 statistics, which show rates of 9.3 for Republican-voting counties and 13.5 for Democratic-voting counties. He uses this to argue that homicide trends in Republican counties have increased while trends in Democratic counties have decreased since reforms have been implemented. However, both trend lines actually look somewhat similar, with large declines in the mid-1990s followed by a recent spike beginning around 2015 and worsening in 2020.
Most of the author’s analysis compares homicide and death rates between the years 1990, 2010 (the year before major statewide criminal justice reforms began), and 2020. The analysis presents rates, which are generally preferable to relying on crime counts. However, vast differences in population size can also have a large impact on the rates for certain types of crime – especially those that already occur at a low rate, such as homicide. Republican counties are generally smaller, so one or two homicides tends to increase the rate fairly high. For precisely this reason, the California Department of Justice (DOJ) declines to calculate a rate for very small (measured by population) counties. In fact, for 11 of the Republican counties, California’s DOJ reports “–” (too small to calculate). This is not noted in the CJCJ report despite being a major limitation of the study.
For the Republican counties for which a rate can be calculated, California DOJ reports 2020 homicide rates for Republican counties that are large enough for a rate to be calculated. These rates (per 100,000) are highest in Kern County (12.7), followed by Shasta (6.8), Tulare (6.0), Madera (5.7), Sutter (4.9), Yuba (2.7), and Placer (1.8) counties. Breaking down these numbers by county shows that Kern County appears to be an outlier, with rates that are nearly double that of the adjacent category. Kern County is also the largest Republican county, including Bakersfield and surrounding rural areas. When calculating the overall rate for the above seven counties, the homicide rate is 5.8 per 100,000. When excluding Kern County from the calculations, the rate drops to 4.7. One question that remains is whether Kern County is an anomaly, and if so, why? To answer this question, future research would need to examine county-level rates and attempt to control for important outside factors such as population size, average income, demographics, and rural vs. urban environment. A potential reason for this might be that Kern County has a decent amount of drug gang activity as well as a major highway that is often used for drug trafficking.
The Democratic average homicide rate appears to be driven down by certain counties, including: Marin (0.4), Napa (0.7), Orange (1.8), Santa Barbara (1.8), Sonoma (2.0), San Mateo (2.1), Ventura (2.4), Yolo (2.4), and Santa Clara (2.7) counties. Conversely, the Democratic counties with the highest rates are San Joaquin (10.8), Fresno (9.2), Alameda (8.6), Merced (8.4), San Bernardino (8.3), Los Angeles (6.7), Sacramento (6.4), and Riverside (6.3) counties. When averaging the homicide rate for the above 17 counties, the rate is about 4.7, which is equal to the rate for Republican counties after excluding the outlier (i.e., Kern County). The numbers among Democratic counties appear to be more evenly spread out, absent of major outliers.
Presenting findings in the aggregate (rather than county-by-county) makes it easier to obscure rates and can make outliers undetectable. Further, it is not possible to detect differences between counties. Even within Republican-voting counties, there is likely still a fair degree of variation in crime rates (which we can see from looking at the limited homicide data above), which often paints a much different picture than aggregate-level trends. There is a similar problem with Democratic-voting counties, i.e., there is likely some variation at a county-level driven by outside factors that can obscure aggregate-level trends. The author also states that Republican counties tend to be more rural and dominated by White residents than Democratic ones, however, he fails to discuss how geographic and/or demographic variables may affect the relationship between conservative policies and crime rates. Democratic-voting counties tend to include large cities and inner suburbs, racially diverse populations, and more liberal policies. They also have larger populations and tend to have micro-level “hot spots” of crime. Thus, looking at these rates in the aggregate detracts from the high violent crime levels occurring in crime hot spots of various cities.
Secondly, Proposition 47, passed in 2014, reclassified several drug and property offenses from felonies to misdemeanors, which caused a large “decrease” in crime. However, because drug and property offenses are classified differently after Proposition 47, a large reduction in reported property offenses is likely related to a documentation issue rather than a prevalence issue. If some crimes were previously felonies and are now being classified as misdemeanors, obviously the numbers and rates for these felony crimes would decrease due to the changes in statute. In fact, reported property crimes in California actually did not decline between 2010-2013 and only started declining after the passage of Proposition 47. Thus, it seems more plausible that decreases in reported property crime are reflective of how crimes are being documented rather than actually reflecting “real” drops in the prevalence of crime.
Another key statement in the report is that “after mammoth declines in the state’s urban areas and increases in Republican areas during the 2010s, Republican counties have homicide rates that are 28 percent higher than Democratic counties.” This statement is similar to others made in a prior CJCJ report published in June 2021 that claimed that California’s “urban crime rates” were declining in 2020 despite increases in homicides. In the June 2021 report, Mike Males and co-author Maureen Washburn of CJCJ essentially argue that despite a substantial one-year increase in homicide, overall crime is lower than it was in 1990, even after accounting for the 2020 homicide increases. They say part of this is due to California’s major criminal justice reforms during the 2010s, e.g., Proposition 47. The conclusions from both of these reports argue that liberal crime policies tend to decrease crime while conservative policies increase crime. In reality though, evaluating the impact of any policy is difficult, and even the most rigorous studies have a hard time doing so. Part of the reason for this is because when numerous reforms take place, it is difficult to isolate the impact of each one individually. Second, there are many other factors impacting crime rates that are difficult to control for (e.g., rural vs. urban, population size, demographics, etc.), making it even more difficult to isolate the impacts resulting from policy and/or political affiliation alone. Thus, it seems a bit bold to conclude that correlations between a county’s political affiliation and their crime rates equates to a causal relationship between these factors. Correlations are relatively easy to prove, though causation is much harder to prove, requiring a sound research design and analytic strategy. Neither CJCJ study meets these criteria, and we cannot stress enough that “correlation does not equal causation.”
While it is not entirely clear why, the author states that homicide death rates among Whites were similar between Republican and Democratic areas in the 1990s, but that by 2020, White homicide rates were nearly twice as high in Republican counties than in Democratic ones. He also compares homicide death rates for White residents in Republican counties with residents of color in Democratic counties, finding the former to be higher than the latter. He claims that this comparison challenges the perceptions that communities of color and urban areas are epicenters of California’s recent homicide spike. This seems like an odd comparison though, given that there are multiple differences between the two groups (i.e., White residents in Republican counties and Nonwhite residents in Democratic counties are very different from each other) that could have impacted crime rates. Again, “correlation does not equal causation.” Further, there are many studies over time that strongly support the reality of crime hot spots, which are micro-level locations (typically within urban areas) that have higher concentrations of crime relative to adjacent areas.
One part of the report displays the convergence of arrest rates among people of different races over time, again comparing White residents of Republican counties to Nonwhite residents in Democratic areas. The report states that White residents of Republican counties had much lower arrest rates for all crime than Nonwhite residents of Democratic areas in 1990, a disparity that “largely evaporated” by 2010. It goes on to explain that by 2020, White residents of Republican counties had higher violent death rates than Nonwhite residents in Democratic counties. In addition to the odd groups selected for comparison, an issue with this portion of the report is that the analysis looks at “violent death rates” instead of homicide rates. The concern with this measure is that the “violent death rate” is calculated as a sum of homicides, suicides, and accidents, rather than just being a sum of homicides alone. It is unclear why the author would use this type of measure when it does not explicitly pertain to homicide. Similarly, the report also states that firearm deaths are also more prevalent among White people in Republican-voting counties. It uses this finding to negate the “widely held belief” that communities of color in urban areas experience the most gun deaths and/or violent deaths. However, gun deaths and violent deaths do not equate to homicides, and this is a key measurement problem. For example, the report also says that violent deaths occurring in Republican counties are largely comprised of suicides, which illustrates our point and further discredits its conclusions.
The CJCJ report also states that: “drug and alcohol deaths soar among White Californians – especially in Republican areas.” A key part of the CJCJ report’s argument here is that “mortality trends show that alcohol and illicit drug use, especially among older populations, is linked to increasing arrest and incarceration.” However, once again, the author does not account for outside factors (e.g., type of employment). This finding is actually not surprising given the vast differences between urban and rural areas; for example, one key difference is in the type of work available (white collar vs. blue collar). Blue collar work is more common in rural areas and Republican counties, and is more often linked to injuries that are often treated with opioids. Thus, opioid addictions and deaths are typically higher in rural areas irrespective of arrest and/or incarceration trends. As we know, “correlation does not equal causation,” and key variables that differ between Republican and Democratic counties should be accounted for (via methodology or statistical analysis) to establish a causal relationship.
Overall, the recent CJCJ report claims that Republican-voting counties are suffering from more crime than Democratic-voting counties. However, due to the various reasons stated above, the study is not of high enough quality to establish causation between these two variables. That is, there are possibly other differences between Republican and Democratic counties aside from their political affiliation that might be impacting crime rates.