Lately I’ve often heard that family is the key divide between Democrats and Republicans. Democrats are supposed to be single and childless, Republicans married with kids. So I decided to check this out for myself in the General Social Survey.
To be precise: What is the effect of being married (married=1, other=0) and number of kids on party identification, controlling for: log income, education, race, gender, age, year/1000, church attendance, and Biblical literalism? (See my class notes (here, here, and here) for details on the variable definitions and coding).
Over the entire sample (1972-2010), family variables have quite a small effect. Being married made you 2.3 percentage-points less likely to be a Democrat, and 1.8 percentage-points more likely to be a Republican. The effect of children was statistically insignificant – .3 percentage-points less Democratic per child, .1 percentage-points more Republican per child.
Since most of the people pushing this thesis emphasize recent politics, I decided to limit the sample to 2000-2010. Here’s what I found.
Conditional probability of being a Democrat:
More than in the past, for sure. But still marginal, right? The Republican results are similar:
P.S. Controlling for ideology, the family variables make no difference at all. Given the stability of ideology over people’s lives, the simplest explanation is that political ideology sways marriage and child-bearing as well as voting.
READER COMMENTS
Anthony
Jan 8 2013 at 11:16pm
It’s appropriate to control for those covariates if you think that causality doesn’t run in the other direction – that having kids doesn’t cause you to become more religious (or to change your ideology.) I would guess it runs in both directions.
Jonathan
Jan 9 2013 at 12:14am
Poor word choice (“effect of being married… on party identification”, “being married made you”) since you’re not doing causal inference. May confuse naive readers.
Brandon Berg
Jan 9 2013 at 12:31am
I’m not a social scientist or statistician, so maybe there’s an obvious explanation for this that I overlooked, but why did you divide year by 1000? Seems like it would make more sense to subtract 2000. As it is, it barely moves at all, so the coefficient is huge just because of random noise.
Anthony
Jan 9 2013 at 1:25am
Actually, even controlling for income may not be appropriate. If marriage makes people more likely to vote Republican by raising their income, that’s still marriage making people more likely to vote Republican. I guess what to control for depends not just on the assumptions you’re making about what causes what, but also what questions you’re trying to answer. If the question you want to answer is whether married people/parents who look otherwise the same as non-married people/people without kids vote the same, controlling for everything is certainly the way to go. But if the question you’re trying to answer is, say, “how is the shift away from marriage, as well as the shift away from certain values that cause people to get married, affecting voting”, you wouldn’t want to. And if all you want to do is evaluate the validity of “Democrats are supposed to be single and childless, Republicans married with kids,” you just want some descriptive statistics, not a regression at all. It would also be interesting to look at actual voting behavior in addition to party identification, since GSS has that as well.
@Brandon: Since there’s no statistical significance to year, the interpretation is that year has no explanatory value. This t-statistic/probability is not affected by linear transformations of the year variable: you could add/subtract/multiply/divide it by whatever you wanted, and those numbers would not change. (The coefficient would change and the constant would change.) If that variable did have explanatory power, it would be good to put it into a form that made interpreting it easier. But it doesn’t, so it really doesn’t matter what form it’s in.
ajb
Jan 9 2013 at 9:01am
Anthony is exactly right. Saying marriage and family is shorthand for “What kind of people are you?” And those who are on the fence about bourgeois values may get drawn into the Republican or Democrat camps depending on how strongly they adapt to the social concomitants of conventional marriage and children. Even if 90% of the effect is due to social and genetic predisposition, the reverse effects of marriage and kids on voting at the margin, may be extremely important for the large groups near the boundaries.
Think of the way that easy divorce law coupled to fewer social sanctions led to an explosion in divorces in the last few decades. Laws favoring conventional marriages, conventional social relationships, and penalizing or stigmatizing out of wedlock births, alternative living arrangements, and helping middle class families have safe, bourgeois schools in the major cities would potentially have a huge effect on a country like the US if even 3-5% of the population shifted from blue to red. Not to mention worrying about the effects of the low end immigration that Caplan is always championing.
lighthouse
Jan 9 2013 at 9:01am
[Comment removed for supplying false email address. Email the webmaster@econlib.org to request restoring your comment. We’d be happy to publish your comment. A valid email address is nevertheless required to post comments on EconLog and EconTalk.–Econlib Ed.]
Ken B
Jan 9 2013 at 10:43am
Hmmm. I spot a possibility. Families cause religiosity. This is actually pretty common I think: a less or unreligious spouse adopts the religion of the more religious spouse; absent this family formation he wouldn’t.
John Fembup
Jan 9 2013 at 10:51am
Bryan, you found that “Over the entire sample (1972-2010), family variables have quite a small effect.”
This entire period of 38 years is more than a whole generation and it’s known that party affiliations varied during that interval. Did you happen to look at whether there was some trend in your basic finding over the period? Do the data even permit looking at changes over time?
DarwinCatholic
Jan 9 2013 at 11:44am
Right now your most predictive factors involve race, which we know has a very strong effect on political alignment. It seems like it might be interesting to re-run looking at Whites only.
The next up question would be whether it’s appropriate to control for religious practice.
Steve Sailer
Jan 9 2013 at 7:28pm
http://www.theamericanconservative.com/articles/value-voters/
John Fast
Jan 10 2013 at 1:47pm
It makes sense to me that political ideology — and traditionalism — are correlated with both family size and political affiliation/voting.
However, unless I’m overlooking something, it’s also correct to say that family size is correlated with political affiliation/voting.
Sure, if you control for ideology, family size has no effect — but that’s not necessarily the issue.
Possibility #1: If people said “Having kids *changes* people from leftists/Democrats to conservatives/Republicans,” it looks like you can point to the data to refute that.
Possibility #2: If someone simply says “The average conservative/Republican has more kids than the average leftistDemocrat,” they’re spot on.
Are you interpreting “Democrats are single and childless, Republicans are married with kids” to mean Possibility #1 rather than Possibility #2?
Finch
Jan 10 2013 at 2:38pm
@John Fast
#1 means that if Republicans adopt and establish policies that make it easier to form and maintain families, they’ll get more Republicans.
#2 means that if Republicans adopt and establish policies that make it easier to form and maintain families, they’ll make existing Republicans happier, but won’t grow the party.
So it’s a good question to ask.
Finch
Jan 10 2013 at 2:46pm
> #1 means that if Republicans adopt and establish
> policies that make it easier to form and maintain
> families, they’ll get more Republicans.
I meant this in the short-term sense in which people switch political affiliations, but there’s a long term sense in which it’s true, and that’s that Republicans beget Republicans. That feeds into the oft-discussed framing of American politics as a race between Republican reproduction and immigration. That aside, I still think the original question is interesting.
Comments are closed.