How important are labor unions to the Democratic Party?
The question isn’t easy to answer. Perhaps the biggest way that unions help the Democrats is by donating to them: unions are among the top contributors to Democratic campaigns and left-of-center causes, having given tens of millions of dollars to them in 2010. But, although campaign contributions undoubtedly have a fairly profound effect on elections and voting in the Congress, the literature studying this question is somewhat out of date and has to contend with numerous tricky methodological issues.
One thing that’s a little easier to examine is the effect that being a member of a labor union, or being in a union household, has on someone’s likelihood of voting Democratic.
In 2008, for instance, 59 percent of people in union households voted for Barack Obama, as compared to 51 percent of people in non-union households — a difference of 8 percentage points, according to the national exit poll. An extremely simple analysis might conclude, then, that the presence of the labor union vote boosted Mr. Obama’s share of the vote by slightly under 2 points overall: the 8 percentage point “bonus” that he received among union voters, multiplied by the 21 percent of the sample that was in labor union households, which is 1.68 percent.
The potential problem with this is that labor union voters are not distributed randomly throughout the population. Instead, virtually every other demographic variable — age, income, geography, occupation, gender, race, and so forth — is correlated in some with the likelihood of being in a union.
It could be, for instance, that because labor unions are concentrated in blue states, especially those in the Northeast and the industrial Midwest, the apparent influence of union membership on voting is really just a matter of geography. Alternatively, it could be that union members tend to vote Democratic despite having certain other characteristics that are ordinarily harmful to Democrats: for instance, union members tend to skew a bit older than the rest of the population and older voters normally tend to vote Republican. If so, the quick-and-dirty estimate from the exit poll might understate the effect of union membership on voting behavior.
Fortunately, there is very good data that we can use that can help to isolate the effect of different types of demographic variables. This data set is the National Annenberg Election Survey, which conducted interviews with tens of thousands of adults over the course of 2008.
I ran a logistic regression analysis on the Annenberg data, looking at respondents who said they planned to vote for either Barack Obama or John McCain (I eliminated those who said they were undecided or planned to vote for a minor-party candidate). My study covered interviews from July 2008 onward, during which time Annenberg was asking respondents about a large number of demographic characteristics in addition to their voting preferences.
After eliminating those respondents who refused to answer one or more demographic questions, I was left with a data set of 14,927 respondents, of which about 51 percent said they planned to vote for Barack Obama (fairly close to his actual vote of about 53 percent). Because of the extremely large sample size, we are able to isolate the effects of different types of demographic variables quite effectively (a smaller data set would create problems because a lot of these explanatory variables are highly correlated with one another).
We can never be completely certain that there aren’t other variables that we have failed to account for. But the analysis ought to be reasonably robust — much more so than just looking at the topline numbers from exit polls.
The chart below illustrates the effect of 23 different demographic variables on one’s intended presidential vote. The size of the bars corresponds to the impact that the variable had on the respondent’s likelihood of voting for Barack Obama. Blue, left-leaning bars indicate that the demographic variables made the respondent more likely to vote for Mr. Obama; red, right-leaning bars meant it predicted higher levels of voting for Mr. McCain. This is a pretty interesting chart, and so I’d encourage you to take a few moments with it.
As you can see, by far the most important demographic variable was being African-American. It overwhelms everything else: for instance, even if every other demographic characteristic of the respondent tended to predict a Republican vote, the voter was nevertheless 81 percent likely to vote for Mr. Obama provided that they were black, according to the model.
The next-most important demographic characteristics, after being African-American, were two others related to race, two others related to religion — and two related to union membership.
Specifically, 64 percent of union members in the Annenberg data set voted for Barack Obama. By contrast, if these same voters were not members of unions but every other demographic characteristic were held constant, the analysis predicts that 52 percent of them would have voted for Mr. Obama anyway. Thus, the marginal impact of being a union member on the likelihood on voting for Mr. Obama was 12 percentage points (64 percent less 52 percent) on average. Because about 10 percent of voters in the study were union members, this boosted Mr. Obama’s overall vote share by 1.2 percentage points.
There was also an effect from voters who were not members of a union themselves, but had someone else in their household who was (these voters are designated as “union member in household” in the chart). These respondents, which represent about 5 percent of the Annenberg data set, were between 8 and 9 percentage points more likely to vote for Mr. Obama than they otherwise would be. This improved Mr. Obama’s overall vote share by a further 0.5 points.
Combined, the union voters and the union-household voters improved Mr. Obama’s share of the vote by 1.7 percentage points, according to the model.
This is almost exactly identical, you might notice, to the quick-and-dirty estimate provided by the exit poll. But note that the Annenberg sample contained somewhat fewer union voters — about 15 percent, counting both union members and voters in union households — than did the exit poll (21 percent), possibly because it did not use a likely voter model. If we assume that the our analysis of the Annenberg data was correct as to the magnitude of the effect of union affiliation, but that the exit poll was correct as to the share of union voters in the actual electorate, then the net effect on Mr. Obama’s vote was 2.4 percentage points.
Also, our study is measured in terms of the marginal effect on Mr. Obama’s vote. But the way that we have designed the analysis, any votes that did not go to Mr. Obama instead went to Senator John McCain. Therefore, the impact on the margin between the two candidates was twice as large: not 2.4 points, but 4.8 points.
This is fairly meaningful. Of the last 10 elections in which the Democratic candidate won the popular vote (counting 2000, when Al Gore lost in the Electoral College), he did so by 4.8 points or fewer on 4 occasions (2000, 1976, 1960, 1948). So, while the impact of union voting is not gigantic in the abstract, it has the potential to sway quite a few presidential elections, since presidential elections are usually fairly close.
With that said, we need to be careful here. In the counter-factual world where unions did not exist, a number of other things would undoubtedly have evolved differently as well, which would have produced different coalitions and constituencies.
Likewise, unions are not just going to disappear (even if a few Republicans might wish they would). If a Supreme Court decision came down on Monday that declared unions illegal, some of these voters might continue to vote Democratic because they were formerly members of unions and continued to identify with them. Or unions might be replaced with some other type of political organizations.
More tangibly, Republican efforts to decrease the influence of unions — while potentially worthwhile to their electoral prospects in the long-term — could contribute to a backlash in the near-term, making union members even more likely to vote Democratic and even more likely to turn out. If, for instance, the share of union households voting for Democrats was not 60 percent but closer to 70 percent, Republicans would have difficulty winning presidential elections for a couple of cycles until the number of union voters diminished further.
*-*
As a very brief postscript, here is the same analysis when applied to voting for the House of Representatives rather than for president.
The magnitude of the effect is essentially identical to that observed in presidential voting. The presence of union voters increased the Democrats’ share of the congressional vote by 1.7 points using the Annenberg data set, or by 2.4 points if one adjusts the share of union voters to match that observed in the 2008 exit poll.