Applying the census indicators “class of worker” and “occupation” to nationwide workforce data, we see that large shares of non-BA holders are outside the most general definition of the working class—non-managerial wage employees—while the vast majority of BA holders remain within it. Educational attainment thus does not give a clear-cut picture of social classes; it instead yields vague descriptive strata that straddle class divides. All of Commonsense Solidarity’s “key takeaways,” however, are about the views or behavior of “working-class voters” – claims which don’t hold up when derived from a mixed-class sample, such as theirs (5-6).
What makes this even stranger is that Commonsense Solidarity deploys a different class identifier when constructing hypothetical candidates. These non-existent persons are defined by their occupation: “Teacher,” “Construction Worker,” “Small Business Owner,” “CEO of Fortune 500 Company,” and so on. Based on subjects’ responses to these hypothetical candidates, the authors draw the conclusion that “[w]orking-class voters prefer working-class candidates,” (5). This claim is meaningless, however, because different notions of class are used for subjects and candidates. Its general thrust seems plausible, but the conceptual confusion of their data does not support it.
The authors attempt to remedy the shortcomings of education-as-class with “different definitions” later on: “manual vs. mental work,” levels of “supervision at work,” “routine vs. creative work,” “class self-identification” and a “three-dimensional variable for class” derived from the first three and split into “middle” and “lower” class subgroups (51, 54, 59, 61, 57, 63). While it is certainly admirable to explore these dimensions, all are applied within the same sample defined by the authors’ original (and problematic) notion of class. These are not, therefore, different definitions of the working class but rather additional subcategories within the same educationally-defined stratum.
Political identity and political messaging are also problematic. The authors of Commonsense Solidarity “restrict [their] sample to individuals who do not identify as Republicans.” This is despite their stated interest in “working-class voters,” a sizeable share of whom identify as Republican and oddly discordant with noted vote switching across recent presidential elections (19). Concern with primary battles in the Democratic Party can’t fully account for this choice either as the authors also include independents—even those who “lean Republican”—in their sampling frame.
On the candidate side, Commonsense Solidarity constrains things differently. Republicans and “generic Republican messaging” are included but independents off the Democratic ballot are not (10). Given that a historically high share of US adults say a third party is needed, that many eligible voters don’t vote, and that independent or third-party candidates, including Bernie Sanders in all of his non-presidential contests, have either won or significantly impacted general elections, this is a questionable assumption. These choices make for an oddly synthetic landscape: no Republicans among subjects and no independents among candidates. This does not yield a “more realistic portrait of voter attitudes,” as the authors claim (5).
And political messaging––how the study’s candidates “speak” to issues––is defined by a specious dichotomy between “wokeness” and economic populism. Study subjects were shown pairs of hypothetical candidates that varied on seven dimensions, one of which was “messaging style.” “Each communication style is delivered in the form of a hypothetical soundbite,” the authors tell us, which “draw[s] on language and themes from actual political messaging…and reflect the most important general strands of contemporary campaign rhetoric” (10). They condense these into a five-point scale ranging from “progressive populist” through “‘woke’ progressive,” “‘woke’ moderate” and “mainstream moderate” to “Republican” (10).
“Wokeness” is a key differentiator here, but it is not a politically precise term. Broadly meaning “alert[ness] to racial prejudice and discrimination” and popularized by Black Lives Matter struggles, it is wholly compatible with economic, social, and political egalitarianism. At the hands of Commonsense Solidarity authors, however, it is sectioned off as some contrary viewpoint. The alleged trade-off between economic “populism” and antiracist “wokeness” then becomes a central theme—indeed a core finding—of their inquiry:
Progressives do not need to surrender questions of social justice to win working-class voters, but “woke,” activist-inspired rhetoric is a liability. (5)
Blue-collar workers are especially sensitive to candidate messaging — and respond even more acutely to the differences between populist and “woke” language. (6)
We should be skeptical. Jacobin editors and affiliated authors have clearly staked out a position, developed long before this study, against the centrality of antiracism to left coalition-building. That Jacobin now sponsors a study in which “wokeness” is manipulated into a narrow alternative to economic populism raises multiple red flags—and indeed “take[s] the bait” of conservatives that one Jacobin author once warned against. Why should awareness of racial injustice be counterposed to redistributive economics when a very large share of those voicing the first also voice the second, and when the study’s truncated “woke” soundbites belie the actual statements of their real-life inspirations?4“[T]he woke progressive” archetype of Commonsense Solidarity, according to its authors, “borrows rhetoric from Alexandria Ocasio-Cortez and Ayanna Pressley” (10).
It shouldn’t. Such misrepresentation casts a shadow over the study’s design and the authors’ interpretation of what subjects’ choices mean – redistribution over racial justice, as if the two were exclusive.
But let us be charitable. Let us pretend for a moment that Commonsense Solidarity’s muddled notions of class, party, and messaging are unimportant. Would its conclusions hold up? Does the study have what researchers call “internal validity” – plausible support for its claims in its findings? Our deeper dive reveals that it does not.
The report makes use of a quasi-experimental technique known as conjoint analysis. Without getting too technical, what this effectively does is allow for simultaneous and differentiated assessment of the factors driving individual choice. These could be almost anything: brands of cereal, immigration policies, or political candidates, just to name a few. Overall, Commonsense Solidarity makes appropriate use of conjoint analysis and its authors even heed warnings to use “marginal means” rather than “average marginal component effects” when reporting their results.
Yet they fail to heed their own advice and a basic rule of statistical reasoning. In walking readers through that labyrinth, the authors state:
The bars around each dot indicate how confident we are that a given value is correct. Each bar to the left and right of the dot indicates the margin of error for a given estimate…It is important to note that if the bars overlap with the vertical line at .5, we cannot conclude that respondents had a net positive or a net negative opinion of that characteristic (since there is no significant difference between the value of the dot and .5).
Similarly, if the bars around one dot overlap with the bars around another dot, we cannot conclude that there is any statistical difference between the two characteristics. (23, emphasis added)
More straightforwardly, overlapping bars for two or more elements means there is no difference between them. It cannot be inferred that one exists, especially not among all “working-class voters” as the authors mistakenly do.