Last week, NBC News ran a piece linking changes in the racial diversity of student bodies in U.S. schools to recent pushback against school districts’ diversity, equity, and inclusion measures. Per its authors, Tyler Kingkade and Nigel Chiwaya,
Student enrollment data suggests that [conflicts over these diversity and inclusion measures] tend to occur in communities that experienced significant demographic shifts in recent decades.
As someone whose dissertation centered on competition theories or race and ethnic relations, Kingkade and Chiwaya’s hypothesis rings true to me. As a social scientist, though, I found their research design wanting. Instead of looking at changes in racial diversity and the occurrence of recent pushback in all school districts nationwide, or even in a random sample of districts, the authors looked only at “33 cities and counties where school districts have faced rancor over equity initiatives this year in at least three recent school board meetings,” according to NBC News’ tracking of media reports. And they found that, in 22 of those 33 districts, “The exposure of white students to students of color increased more than that national average.”
This kind of design is called selection on the dependent variable. Researchers identify cases that experienced an event or change of interest and then look for commonalities among them to see what might be driving that change.
The problem with this design is that it doesn’t tell you whether those common elements differentiate the cases studied from cases that didn’t experience the event or change of interest, which is what you really want to know when you’re trying to draw inferences about cause and effect. Kingkade and Chiwaya nod in this direction by comparing changes in school diversity in their set of 33 districts to the national average, but that’s not as robust a test as we’d ideally like to see.
Like I said, I’m a social scientist, and I happen to work on a project that collects data on protest activity in the United States, including against critical race theory and other concepts connected to the school equity, diversity, and inclusion measures discussed by Kingkade and Chiwaya. So, I thought I’d take a crack at their idea, but with a statistical model applied ot a larger sample that escapes this problem of selection on the dependent variable, and with some other features that also affect, or at least co-vary, with counties’ propensity to protest.
To be clear, I’m not exactly replicating what Kingkade and Chiwaya did. Their analysis used school districts as its units of observation; their event of interest was people showing up repeatedly at school board meetings in 2021 to complain about diversity, equity, and inclusion measures; and they looked at change in the demographic make-up of student bodies in those districts as a potential driver. By contrast, my analysis uses counties as the unit of observation; my event of interest is the occurrence of at least one protest in 2021 at which participants complained about critical race theory; and I’m looking at demographic changes in the county as a whole as potential drivers, rather than in the student body in particular. Also notable, while some of the protests considered in my analysis occurred at or around school board meetings, not all of them did.
Okay, so what did I find?
First, it’s important to note that only a small fraction of counties have actually seen any protests against critical race theory in 2021. Of the 3,223 counties in the United States, CCC has only recorded anti-CRT events in 72 of them, or about 2.2 percent. Of those 72 counties, 60 had just one anti-CRT protest; eight had two protests; three had three protests; and one had six. By contrast, in the peak part of the George Floyd Uprising in June and July 2020, CCC recorded protests against racism in 1,400 counties (43 percent). So, while protests against CRT have garnered a fair bit of media attention this year, they haven’t exactly mushroomed, at least not so far.
Because these events are so rare, I decided to treat their occurrence as a binary outcome (any vs. none) rather than trying to model counts of events or participants, as I’d prefer to do if there were more variation in those things. So, instead of something like a Poisson or negative binomial model, I’m just running good old logistic regression.
For demographic change, I tried differences in two different measures, both estimates from the U.S. Census Bureau’s 2009 and 2019 five-year American Community Surveys: 1) change in percent white alone (i.e., not white and some other race(s)), and 2) change in percent foreign born. Per competition theory and consistent with Kingkade and Chiwaya, I expected to find that, other things being equal, counties which experienced larger declines in the white share of the population or larger increases in the share of the population born outside the U.S. from 2009 to 2019 would be more likely to have seen protests against CRT in 2021. Because these two measures are interrelated, I fit a separate model for each.
As for what other things we should hold equal, I included several in my model: total population (logged), median household income (logged), whites as a share of the county population in 2009 (to get a “pre-treatment” take on local demographics), and the share of the presidential vote won by GOP candidate John McCain in 2008 (to get a “pre-treatment” take on local partisan loyalties). Because prior analysis leads me to expect a non-linear relationship between GOP vote share and these events, I used a generalized additive model (GAM) with smoothing splines for that one covariate.
The first chart below plots the predicted probability of any anti-CRT events in 2021 across the observed range of changes in percent white, holding all the other socio-demographic covariates constant at their median values and holding the 2008 GOP vote share constant at 50 percent. The second does the same with changes in percent foreign born. In both plots, the vertical dashes along the x-axis show the distributions of those quantities (change in percent white or change in percent foreign born, respectively) in the observed data.
As I read those charts (and the tables summarizing the relevant models, which I won’t dump here), I see some support for the hypothesis that demographic changes are helping to drive recent right-wing mobilization against anti-racist education, which often gets shorthanded by its opponents this year to “critical race theory.” With both measures of demographic change, the effects work in the expected direction, and they are statistically significant (at the 0.05 level for change in percent white and the 0.01 level for change in percent foreign born). They are not substantively large, however. For percent white, the difference in the probability of any anti-CRT events from one end of the fat part of the distribution of observed values to the other (-0.2 to +0.2) is only about 0.005, or about 25 percent. For percent foreign born, it’s more like 0.01, or roughly double. That’s hardly trivial, but remember that most observed values are closer together than that.
So, on the whole, I’d say my analysis supports the view that different experiences with demographic change are contributing to local variation in recent backlash against purportedly leftist anti-racist education efforts in schools. Best I can tell, the effects aren’t large, but they do seem to be there at the margins, at least in the data I can assemble now. Where whites’ share of the population is shrinking faster or the foreign-born population is growing faster, people are more likely to gather in protest against anti-racist activism and ideas that some (many?) whites might construe as insulting or threatening. As others have noted, white grievance has effectively become the organizing principle of the Republican Party, and many whites apparently feel more aggrieved when they find that they’re encountering more and more people in their daily lives who don’t don’t share their racial identity.