A lot of people are worried about the impact that changing racial demographics will have on US elections. These fears are largely based on the fact that Hispanics tend to overwhelmingly vote democrat, and the US Hispanic population is accounting for a larger proportion of the general population every decade.
Consider this data from NYT exit polling:
This leads some people to fear that the entire US will someday look like California. Some people have pointed out that it could also look like Texas.
A lot could be said about this, but what I want to do here is see how well race/ethnicity predicts state level voting. Specifically, how well does the proportion of a state which is White/Black/Hispanic correlate with the net republican vote in that state in US presidential elections, and how does this compare to other demographic variables?
Here is a cross section of US presidential elections held between 2004 and 2016 using data pulled from the Census (2005 data was used for the 2004 election):
A few things are worth noting here:
- Education is the best predictor of state voting, and its significance is growing with time.
- The proportion of a state which is Black has essentially no relationship with how that state votes even though Blacks heavily favor democrats. This should caution us agiant assuming how a group will impact elections based on how they vote.
- % Hispanic and % White are not great predictors of state election outcomes, but they are growing in importance with time. With Whites, this reflects a consistent trend. With Hispanics, this is pure “Trump effect”.
Using the same data, we can look at how changes in demographics predict changes in state-level election results over time:
As can be seen, this paints a very different picture than did the cross-sectional data. Changes in the Hispanic proportional population size is the largest driver of voting change in this set of predictors.
In terms of the absolute magnitude of the effect, a 1% increase in the proportion of a state that is Hispanic predicts a .041 point decrease in the net republican vote (SE = .013, P=.003).
When looking at the cross-sectional data, we saw that percent Hispanic was only really predictive of state-level voting in 2016. Given this, it might be that increasing Hispanic populations between the years 2004 and 2016 only predicted less republican voting because of the exceptionally strong power of percent Hispanic as a predictor in the 2016 election.
To test this, I ran the same analysis looking at changes between 2004 and 2012. Changes in % Hispanic was still the best predictor, but it was a bad predictor. A 1% increase in % Hispanic predicted a .03 point decrease in the net republican vote with a standard error of 0.3, a p-value of .329, and an r squared value of .02.
It is hard to say what is to be made of this data. Here is how I imagine two people, one who favors immigration restriction and one who does not, might interpret this analysis:
Anti-restrictionist: “This data shows that race is not a particularly strong predictor of election outcomes, and that demographic changes are only going to hurt the republican party if it continues to move towards nationalism. Given this, the correct response to demographic shifts is to move away from nationalism as fast as possible.”
Restrictionist: “Even if the effect of demographic change on voting isn’t particularly strong, it is consistently in the same direction. Longitudinally, it is stronger that the other variables looked at in this analysis, and the effect was negative even before Trump so your point about nationalism does not stand. Furthermore, even if the net republican vote doesn’t change a lot when the % Hispanic increases by 1 point, it’s going to change a ton when % Hispanic increases by ten or more points over the coming decades. Given that US elections often come down to just a few points in key swing states, even a small effect could decide an election.”
Anti-restrictionist: “You are assuming that we can use this data to predict decades into the future. Since Hispanic voting is highly variable, as seen in the first table, and since the relationship between changes in the population that is Hispanic and net republican vote is so dependent upon the year we look at, this probably isn’t a prediction we can be very confident in. Moreover, there are other variables, like education and marriage rates, that are also changing in a way that could hurt republicans, and these changes are less geographically clustered than Hispanic immigration, who tend to not be concentrated in swing states. The major exception to this is Florida, but Cubans vote republican.”
Restrictionist: “We probably can’t be highly confident in an exact prediction, but we can say that it likely won’t help republicans. We just can’t say if this effect will be big or small. The other variables you named are stronger cross-sectional predictors than ethnicity but weaker longitudinal predictors. Besides, we can oppose the expansion of leftist in the university and the collapse of marriage too.”
At some point, someone would probably mention that what is and is not a swing state isn’t very consistent across decades. On the one hand, this makes the point that Hispanics are not clustered in current swing states less powerful. On the other hand, it strengthens the point that predicting electoral changes decades in advance is really hard to do.
This back and forth could go on, but I think it would converge on a fairly moderate interpretation of the data. People who think the rising Hispanic population ensures democrat victory are probably wrong, as are people who think that it will have no effect. It will probably have a slight to moderate effect. But ethnicity isn’t the only important variable, and its effects could be totally swamped by other changes.
This makes coming to a policy position on the basis of electoral impact difficult. To some people, the preceding paragraph will imply that we should allow immigration, and to others, it will imply that we should restrict it.