In the last post here, I described the data from the November 2018 Current Population Survey. Now it’s time to discuss the strategy for linking registration and voting.
In the United States, residents must register to vote before participating in an election. People usually need to register only once after moving to a new address or jurisdiction. So in a given year, the process will be different for those who need to register and those who don’t.
As a result, later I may want to separate people who have registered in the same year as the election from those who registered earlier. But for now, I’m just going to compare registration rates and voting rates in different communities, regardless of when registration took place.
My aim is to find out whether there are wide differences in registration and voting between Black and White people. I would also like to see whether the relationship between registration and voting differs across these communities. In essence, I want to figure out (1) where registration and voting may be unusually low in the Black community and (2) which part of the process is more likely to be a roadblock to their participation in elections: registration, or voting.
The first step in this strategy is to set a benchmark. My assumption is that White people face no barriers to voting that are specific to their racial background. So I compare Black participation to White participation as a way of suggesting the presence of barriers in each geographical area.
The best way to make this comparison is to look at the behavior of groups who are very similar except for their racial backgrounds. I try to do this by adjusting the participation rates in both groups for age, gender, education, and income level. The overall universe for the study consists of adults who are U.S. citizens aged 18 or older. The CPS did not ask the voting questions of people who did not fit these criteria, even though some might have been eligible to vote.
I categorize voters as male or female, college graduates or not, and under or over age 65, according to the CPS. For income level, I use the log income per adult member of each household, where adults are over age 21. The CPS only reports household income by stratum, ranging from 0-$5,000 per year to more than $125,000 per year, so these figures are approximate. (Because incomes are topcoded, I’m assuming households above $125,000 per year interviewed by the CPS have an average income of $200,000 per year.)
I perform the adjustment for registration via logistic regression on the demographic variables above, using the CPS weights – but only for White respondents. I exclude Black respondents, because I’m engaging in a thought experiment; I want to see how the regression might predict the registration of Black respondents if they were White. So I use the results of the regression among White respondents to generate the predicted likelihood of registering for each respondent in the full sample, based on his or her demographics. Then I subtract that predicted value (which ranges from 0 to 1) from the respondent’s actual status (0 = not registered, 1 = registered). The difference can be thought of as a respondent’s “excess registration” – essentially, how much their registration appeared to depend on other factors, including racial background, after controlling for their demographics. I use the same process to generate an “excess voting” variable among the respondents who were registered.
At this point, it’s worth noting that “excess registration” averaged 0.04 among Black respondents, using CPS weights. So Black respondents were, on average, more likely to register after controlling for demographic factors. The average for “excess voting” was 0.03 among Black respondents. On the whole, the registration difference may not have turned into voting on a one-for-one basis.
By itself, the registration difference doesn’t necessarily say much about a jurisdiction. The difference was smallest (or negative) around the Washington (in Maryland), Virginia Beach, Miami, and Houston areas. It was biggest around Jackson, Philadelphia (in Pennsylvania), Dallas, Chicago, Richmond, Atlanta, and New York (in New York state). A smaller difference in registrations by Black people may signify discrimination (if they are intrinsically more inclined to register, wherever they live) or, on the contrary, a lack of discrimination (if they feel voting is less important where they live).
A better signal of discrimination might be, once again, when differences in registration rates don’t fully translate into differences in voting rates. The areas where this occurred were around Dallas, Richmond, Philadelphia (in Pennsylvania), Birmingham, Washington (in the District of Columbia), and Atlanta. There was also a small gap around Baltimore (in Maryland). So in some of the areas where Black people were more likely to register than White people, the follow-through to voting looked muted.
One way to visualize these data is to show how the excess registration and voting numbers differed across the communities for each geographical area using a sort of vector. I’ve been experimenting with this approach – here’s an example for the big cities listed above:
You can see that around Dallas, Richmond, and Birmingham, adjusted registration rates were much higher in the Black community than among White people, but adjusted voting rates barely differed at all. Around Atlanta, Philadelphia (in Pennsylvania), and Baltimore, there was a closer relationship between registration and voting, but not a one-to-one correspondence. And in Washington (the District of Columbia), adjusted registration was about the same for both communities, but Black people’s adjusted voting was lower.
What does this tell us? Well, one caveat with this analysis is that registration can be based on many years of behavior, while voting only relates to the current year. As a result, people who may have registered for one reason years ago may not vote this year for another reason; the original reason may have dissipated. So Black people in Washington, for example, may have registered to vote when it seemed more urgent to do so, but they’re no longer as inclined to visit the polls. But if the desire to register and the desire to vote are more concurrent, then it might suggest that voting is more difficult for Black people in the areas listed above.
I think it’s also useful to look at the unadjusted registration and voting rates for the same geographical areas, which I’ve graphed here:
The startling thing here is that registration for Black people is lower than for White people almost everywhere. Our adjustment process suggests that this difference can be almost entirely explained by age, gender, education, and income. Since the ages and genders of the communities may not be too different when taken as a whole, education and income could be the driving factors here. This leads to the unsurprising conclusion that a lack of educational and economic opportunity can correspond to lower access to political power.
These are preliminary results, and I’m still considering the pluses and minuses of the statistical approach. I’ll publish similar results for medium-sized urban areas and more rural areas soon. And then I’ll take a look at the figures for November 2020 to see if there were any changes; those might help me to unpick the question of concurrency.