(Occupational) Class and Party Identification
For a while I have been frustrated by my inability to find a concrete answer to what seemed to me to be a very simple question: how has the class composition of the Democratic and Republican party electorates changed over recent decades? The Trump era has obviously brought renewed urgency to this question. A great deal of political commentary and strategic debate now revolves around the question of whether, and to what extent, Trump’s nationalist courting of downwardly mobile white Americans and contempt for all norms of presidential conduct has provoked a class realignment of party support.
Much of the difficulty had to do with how you measure these sorts of things, and what data is available. Essentially, commentators draw inferences about the preferences of different groups based on how the data allow them to divide people up. In the U.S., most surveys — for example the major Pew survey of American political attitudes — ask participants to identify themselves by race, age, gender, level of education completed, religion, and family income (but not wealth) broken out into ranges of a few tens of thousands of dollars, but generally stopping at $250,000/year. The big Democracy Fund “Voter Study Group” is similar.
Such surveys certainly reveal a good deal about the electorate. But when it comes to class, one is forced to rely on what are really in effect proxies: income, which when self-reported I always consider suspect; and education, which is certainly correlated with class but not reducible to it. Using data of this sort, researchers are able to say, for example, that Americans with a post-graduate education are trending increasingly Democratic. But it has real limits. The same levels of income and education can include people doing quite drastically different kinds of work. And although it may come as a surprise to many Americans, the kind of socialization that happens at work is actually widely understood to have effects on an individual’s political outlook.
So from a sociological perspective, I find most American survey data impoverished. This is especially so in comparison to, for example, the French, whose semi-official polling organizations track occupation and family wealth in both pre- and post-election surveys.
Recently, though, I finally found the work that I knew must be out there. Sociologists (and it figures that it would be sociologists) Stephen M. Morgan and Jiwon Lee of Johns Hopkins have a series of papers, which they have admirably published in open access journals, that get right to the heart of the matter. Their key contributions are empirical. What they have done is taken a widely used sociological framework for sorting all occupations into broad, class-like clusters, and then applied this framework to a set of U.S. survey data sources. Together, the result is a plausible picture of the evolution of political party alignment by social class over time.
The occupation-to-class framework, I learned, is usually referred to as “EGP,” for Erikson, Goldthorpe, and Portocarero, who developed the technique in the 1970s and 80s. Morgan and Lee’s modified version of the original framework looks like this:
Morgan explains some of the benefits of a schema like this in a companion piece (Morgan 2017). In particular, it is not unidimensional —aside from categories I and II, it’s not immediate obvious which categories rank above the others, and within each category there is probably a wide range of wages earned. It seems to me that he skips over the key virtue of such a schema, though, which is its relationship to classical understandings of class. That is, it is designed to incorporate not just what work people do (with clear distinctions between “mental” and “manual” labor), but also the type of social relation that their work places them in. Specifically, the categories are designed to reflect differences in the types of employment contracts and the types of supervision arrangements that different kinds of occupations entail. The basic intuition is that these types of social relations have an important experiential quality to them, and that the experiences one acquires through employment-based social relations have an effect on how one thinks about the world. Though political preferences sorted by class in this way may correlate with preferences sorted by income, race, education and so forth, starting from occupation at least allows us to ask which of those is really the causal driver, a question that usually gets skirted over whenever polls report political differences among Americans across income level, education, and race.
To get this schema to yield empirical results, Morgan and Lee first work with data from the General Social Survey (GSS). This is a long-running survey that asks tens of thousands of interviewees dozens (hundreds?) of questions on a whole range of subjects of broad sociological interest. My understanding is that it is much more widely used in sociology than in political science, since only a small number of the questions speak directly to Politics, capital P. The GSS asks people what their occupation is, or, if they’re not working, what their last occupation was. Before 2012 it seems that there was some inconsistency in how this was coded, but since 2012 the GSS has (if I understand correctly) made respondents place themselves into one of several hundred employment categories tracked by the American Community Survey (ACS), the Census’s detailed sample survey. Morgan and Lee assign each of the ACS’s occupational categories to one of their 11 “EGP class descriptions,” allowing them to place practically every GSS respondent into one of their class categories. Some work by the GSS to “recode” their pre-2012 surveys means that this same sorting function can be performed on GSS data back to 1994.
Having done that work, they can then work with the one of the key political questions that the GSS does ask, namely,
“Generally speaking, do you usually think of yourself as a Republican, Democrat, Independent, or what?”
If the respondent selects independent, they receive the follow-up question:
“Do you think of yourself as closer to the Republican or Democratic Party?”
Excluding, for methodological reasons, respondents under 25, they are then able to show shifts in party preferences by class category over time. Grouping survey responses across presidencies, interrupted by “transitions,” you get a data point for 1994–1996–1998; 2000; 2002–2004–2006; 2008; 2010–2012–2014; and 2016. For each class, they provide three codings: one which includes only people who reported that they did vote, with people who said that they “lean Democrat” or “lean Republican” grouped with the party they lean towards; one which also is limited to voters but groups the “leaners” with independents; and one with all eligible voters, i.e. including those who said they didn’t vote in the last election, with “leaners” grouped with independents.
They begin with the three categories most comfortably characterized as working class: IIIb (essentially non-professional, non-supervisory service workers); VI (skilled manual workers employed by others); and VIIa (semi- or unskilled manual workers employed by others).
They then look at those who by most definitions are not working class, namely managers, professionals, and high-skilled service workers (though I strongly suspect that a large portion of the last-named group really should be considered working class under a broad definition).
And finally they look at the categories that most resemble “petit bourgeois,” that is, self-employed workers, including those who employ others (IVab), and then the heterogeneous Category V, which seems to include police officers as well as lower level managers and supervisors. (The two agricultural categories have too few respondents to generate meaningful results.)
So what do we see?
First, those in the lowest tiers of the labor market — Classes IIIb and VIIa, lean Democrat by very large margins, without much apparent change over time, except that the level of Republican identification for these groups is clearly declining. But, for the manual workers (VIIa), who also have the lowest levels of education of all the classes included, there does seem to be an upward trend in the share of all eligible voters who identify as independents, with their numbers appearing to come from both Democrats and Republicans. The implication of this is that for the lowest skilled workers, there is an increasing alienation from both parties. Most likely, since this trend is only visible when looking at all eligible voters, it means that people are describing themselves as independents when what they mean is that they have stopped voting altogether.
Class VI meanwhile shows a lot of volatility. I would guess that a good number of the Obama-Trump voters are in this group. These are people e.g. in Ohio and Michigan who voted for Obama in ’08 because he promised to do more about the foreclosure crisis and in ’12 because of the auto bailouts, but thought Clinton was too close to Wall Street and liked Trump’s focus on bringing blue collar jobs back to the U.S. This last group probably includes also a lot of those who were once known as “Reagan Democrats.” Many of them never became Republicans, but did become actual political independents, and stayed that way.
The pictures of Classes I and II confirm pretty starkly that what we could call the professional-managerial class is indeed trending strongly towards Democrats, while nevertheless remaining quite evenly divided overall. Consistent with most accounts about 2016, people at the very top of the occupational hierarchy went sharply for Romney in 2012, and veered even more sharply towards Clinton in 2016.
The pictures of Classes IVab and V would seem to show (especially for IVab) that the petit bourgeoisie is indeed the principal social base of the Republican Party. This is not a new fact.
Works cited
Morgan, Stephen M. 2017. GSS Methodological Report No. 125. (August). https://mfr.osf.io/render?url=https://osf.io/yr68a/?action=download%26mode=render
Morgan, Stephen M. and Jiwon Lee. 2017a. “Social Class and Party Identification During the Clinton, Bush, and Obama Presidencies.” Sociological Science 4 (August): 394–423. DOI: 10.15195/v4.a16.