In fact, however, the difference between their interpretation of the evidence and ours is mostly due to misunderstandings of our analysis and to a slip in their re-analysis of the same data, which resulted in their double-counting “very liberal” Sanders supporters.
We argued that support for Sanders hinges on social identity more than ideology
The main point of our essay was that support for Sanders hinged less on ideology and issues, and more on social identities and group attachments, than common wisdom has suggested. In support of that point, we noted that exit polls of Democratic primary voters reveal much wider gaps in Sanders’s support between women and men, non-whites and whites, and Democrats and independents than between ideological liberals and moderates.
We also noted that data from a pilot survey conducted in January as part of the 2016 American National Election Study suggest that Sanders supporters were actually less likely than Clinton supporters to favor key policies Sanders has advocated on the campaign trail, including a higher minimum wage, increasing government spending on health care, and an expansion of government services financed by higher taxes.
“It is quite a stretch,” we suggested, “to view these people as the vanguard of a new, social-democratic-trending Democratic Party.”
Hare and Lupton’s constructive critique misses these three points
Hare and Lupton’s critique of that portion of our argument is constructive and highly substantive. However, it seems to us to miss the mark on three counts.
First, Hare and Lupton’s results are misleading because they are based on unweighted data from the ANES survey. (We thank Hare for elucidating this point via email.) The respondents in this pilot study were drawn from YouGov’s opt-in Internet panel. Opt-in panels are notoriously unrepresentative of the U.S. population, and tabulations based on unweighted data from such samples can sometimes be quite skewed. YouGov applies a combination of matching and weighting to the original panel to produce a representative sample, and the weights often make a substantial difference.
In the ANES pilot study, twice as many respondents classified themselves as “very liberal” or “very conservative” in the unweighted sample (22 percent) as in the weighted sample (11 percent). And among Democrats, Democratic leaners and pure independents — the focus of Hare and Lupton’s tabulations — twice as many Sanders supporters described themselves as “very liberal” in the unweighted sample (28 percent) as in the weighted sample (14 percent).
Thus, it is hardly surprising that Hare and Lupton’s unweighted sample looks more liberal than our weighted sample; they essentially double-counted Sanders’s most liberal supporters. Moreover, because liberal Clinton supporters were somewhat less overrepresented in the YouGov panel, Hare and Lupton’s failure to weight the data exaggerates not only the liberalism of Sanders supporters but also the difference in views between Sanders supporters and Clinton supporters.
Second, Hare and Lupton’s claim that “Sanders supporters are in fact more liberal than Clinton supporters across a host of issues” misconstrues the point of our analysis of the ANES data.
As we noted in the first paragraph of our essay, we were interested in whether “Sanders’s surprising success in the primary race is because of his liberal policy positions.” Thus, we focused on the most salient concrete policy issues stressed by Sanders on the campaign trail.
Hare and Lupton, on the other hand, compared Sanders and Clinton supporters on a rather miscellaneous set of attitudes (“Police Treat Whites Much Better”), issue priorities (“Environment a Top Four Issue”) and policy preferences tangential to the current campaign (“No Birth Control Exemption”).
Meanwhile, they ignored what seems to us to be the single most important issue in contemporary U.S. politics (and the key to Sanders’s aim of defining a more social democratic Democratic Party): support for expanding government services and spending.
Finally, Hare and Lupton assumed that we “failed to account for” the fact that some of the respondents coded as supporting Sanders or Clinton in the January survey were Republicans. This may be because their analysis of the unweighted ANES data produced results that seemed to contradict ours.
They suggest, not unreasonably, that some of these Republicans may not have been genuine Sanders supporters but party loyalists who merely took the opportunity to express a preference for Anyone But Clinton on the Democratic side — and that “analyzing only the beliefs of the Democrats and independents who support Sanders will give us a more accurate picture” of the bases of his support.
Contrary to Hare and Lupton’s assumption, the same thought had occurred to us. We examined various subsets of the ANES data (using the proper sample weights) in order to ensure that our conclusions did not hinge on the preferences expressed by Republican “spoilers.” When we simply excluded Republicans, as Hare and Lupton advocate, we still found Sanders supporters slightly to the right of Clinton supporters on each of the three key issues we considered: government services and spending, health-care spending, and raising the minimum wage.
When we excluded potential “spoilers” more directly based on their ratings of the Democratic candidates — specifically, excluding respondents who also expressed a Republican candidate preference and who gave both Sanders and Clinton neutral or negative “feeling thermometer” ratings — we again found Sanders supporters slightly to the right of Clinton supporters on all three issues.
Focusing on Democrats, Democratic-leaning independents and pure independents, as Hare and Lupton do, provides a more mixed picture, with Sanders supporters slightly to the right of Clinton supporters on the issue of government services and slightly to the left on health-care spending and the minimum wage. We see no strong reason to prefer one of these sample definitions to the others.
But regardless of which we employ, the differences in policy preferences between Sanders and Clinton supporters are quite small, averaging just two percentage points. All of these results are quite consistent with Hare and Lupton’s own summary of our findings: “respondents who prefer Sanders held roughly the same positions as — or were more conservative than — Clinton voters, overall.”
There’s no reason to doubt that many of Sanders’s followers are left-wing Democrats
Nothing in our analysis provides any reason to doubt that Sanders has drawn considerable support from committed Democrats who share his left-wing views.
However, political observers (many of whom are committed Democrats who share his left-wing views) have mostly overlooked the fact that there are not enough of them to have made him a competitive candidate for the Democratic nomination.
In the ANES pilot study, Sanders trailed Clinton by just 8 percentage points overall, but by 30 percentage points among Democratic identifiers. Exit polls of Democratic primary voters tell a roughly similar story: Sanders trailed Clinton by an average of 16 percentage points overall, but by 30 percentage points among Democratic identifiers.
The real and perceived success of Sanders’s campaign, by comparison with typical progressive insurgencies in Democratic primaries, seems to rest substantially on his overwhelming support from independents — and they are mostly not in it for the socialism.
Why do American voters support a particular candidate? It’s not necessarily ideology and policy.
Hare and Lupton worry that if support for Sanders is not driven by support for his issue positions, “we would have to question whether American voters know how to pick candidates whose ideologies and policies match their own.”
But researchers have vigorously questioned the coherence and electoral impact of American voters’ ideologies and policy preferences for more than half a century. We summarize the extensive evidence on this point in our recent book. On that score, we see no reason to suppose that Sanders’s supporters are any different from left-voting citizens in previous elections — or, for that matter, from everyone else.
Of course, the ANES pilot study is just one survey conducted in the very early stages of the 2016 campaign. We suspect that the strength of the relationship between issue positions and candidate preferences has increased over the course of the spring, as it has in other primary seasons.
However, careful analysis suggests that increases of that sort are mostly due to attentive citizens learning how to rationalize their candidate preferences by adopting the positions of candidates they support for other reasons.
Thus, even if Sanders supporters turn out in the end to be much more likely than Clinton supporters to favor his policy positions, that would not be persuasive evidence that they support him because they prefer those positions.
The critical response to our essay, of which Hare and Lupton’s critique is just one (unusually substantive) example, underlines the powerful hold in U.S. political culture of the idea that citizens vote for politicians primarily because they agree with them on the issues. Despite its familiar appeal, that idea has long been proven false. That is why we have written a book attacking what we call the “folk theory” of democracy and sketching an alternate view of democracy better grounded in the scientific evidence.
Christopher H. Achen is Roger Williams Straus Professor of Social Sciences and professor of politics at Princeton University.
Larry M. Bartels is May Werthan Shayne Chair of Public Policy and Social Science at Vanderbilt University.
Together they are the authors of “Democracy for Realists: Why Elections Do Not Produce Responsive Government.”