2. Roeber’s Target

While Roeber never presents his position in exactly this form, I think it’s a fair encapsulation of his argumentative strategy in the book:

Just what is “political knowledge,” according to Roeber? He’s quite explicit:

On a very flatfooted reading—one that admittedly does not match the illustrative examples he uses—one might think “political knowledge,” then, would be knowledge of things like how to effectively gerrymander this or that district, or where to target super PAC funds for maximum leverage in flipping the house or senate. That’s the sort of knowledge Democrats or Republicans need to capture or maintain political power.Footnote ² Roeber clarifies that this wasn’t his intent; by “useful” he means “argumentatively useful,” that is, the sort of proposition that one or the other side could effectively use in arguments for their own candidates and positions.Footnote ³ Neither side of the political aisle would be wise to put propositions about how to gerrymander in their campaign ads.

But I worry that, on this interpretation, it’s not so plausible that testimony concerning political matters is systematically unreliable, or at least a lot will turn on just how much wiggle room there is in “systematically,” (admittedly my word rather than his). When Roeber is (helpfully) qualifying his claims, he acknowledges there are some examples where we can acquire political knowledge by testimony—cases where we can know some claim by testimony even though the claim is argumentatively useful to one or the other side of American politics. His example is “that Hilary Clinton used a private email server for sensitive government communications in the lead up to the 2016 election.” (p. 97) Even though this claim was argumentatively useful to Republicans, the sorts of reasons he gives for doubting the reliability of political testimony in general do not apply here; most notably, Democrats did not deny it.

But while Roeber presents this example as a relatively isolated one in which political testimony can be trusted, it seems to me that there are lots of such cases, most of which are too boring or uncontroversial to attract much attention. Consider “the United States went to war with Iran during the second Trump administration.” I’m sure Roeber does not intend his skepticism to extend to claims like this. But it’s also hard to deny that this claim is argumentatively useful to Democrats; the Iran war was comparatively unpopular from its inception, and opposition has grown over time.Footnote ⁴ Examples can be multiplied. If Gavin Newsom is the Democratic presidential nominee in 2028, I’m sure “Gavin Newsom was governor of California,” will be a recurring theme in Republican ads, and that it will provide Republicans with some argumentative leverage. But again, no skeptical argument should threaten it. The point is that lots of propositions that are argumentatively useful to one or the other side of the American political spectrum can nevertheless be known by testimony.

Neil Levy (Reference Levy2024) characterizes Roeber’s target directly in terms of controversy:

This strikes me as more promising, but still not quite right, now primarily because it threatens to be an under, rather than overinclusive characterization of Roeber’s proper target. Plenty of claims where Roeber’s skeptical arguments do get traction, in my view, aren’t typically directly contested by political parties. One notable feature of contemporary partisan politics, not just in the United States, is that different political factions emphasize different issues. What this means is that you can find claims that one side makes that the other side does not contest, not because they concede that it’s true—as in the case of Clinton’s email, or the Iran war—but because they just discuss the whole topic much less.

For example, one of the main reasons Roeber offers for skepticism about political testimony is the existence of “testimonial feedback loops.” The basic idea is that some claim whose non-testimonial evidential basis is actually quite thin can come to be conventional wisdom in the right kind of echo chamber, as a typical member of that echo chamber will have heard it repeated enough times from enough different people that it will seem like there must be lots of solid evidence for it.Footnote ⁵ For present purposes, what matters is that nothing in this process requires that anyone outside the feedback loop contests or is even aware of the claim. And my sense is that testimonial feedback loops where only one side is discussing the topic are in fact not uncommon.

Any examples I pick will necessarily be contentious, but I’ll nevertheless offer a couple, with the caveat that in neither case am I certain the claim in question is false. Rather, I offer both as claims that, as far as I can tell, circulate within subsets of one side of the political spectrum without being much discussed by the other, and in both cases I suspect that broadly Roeberian reasons for skepticism are very much in force. On the left, consider the claim that losses from wage theft—when employers fail to pay their employees the full wages they legally earned—exceed losses from all other forms of property crime.Footnote ⁶ Personally, I’ve heard this in various left-leaning spaces, and I’m not sure I’ve ever seen wage theft discussed on the right. But does the ultimate evidential basis for this claim depend on a tendentious definition of wage theft, or a methodologically shaky strategy for estimating its prevalence? I have no idea, and I think I’m rational to be uncertain; for the sorts of reasons Roeber nicely brings out in his discussion of testimonial feedback loops, I think the claim could easily be often repeated even without strong ground-level support.

Similarly, consider the claim that “federal regulation’s total compliance costs and economic effects are at least $2.153 trillion annually, and certainly vastly higher.”Footnote ⁷ The economic cost of complying with federal regulation is much more a right-wing preoccupation than a left-wing one; I’m not sure I’ve ever seen people on the left dispute that estimates like the one above are too high. And yet, for essentially the same reasons, I think Roeber’s skeptical arguments ought to imply that I cannot know the claim above on the basis of testimony. In these cases, Roeber’s official characterization of “political” knowledge in terms of argumentative usefulness to one or the other side seems closer to the mark in characterizing why left-wing claims about wage theft or right-wing claims about regulatory compliance costs warrant skepticism.

Here’s what I think is going on. The heart of Roeber’s book, in my view, is his extraction of epistemological lessons from psychological research on “politically motivated reasoning.”Footnote ⁸ The work is hard to summarize, and Roeber does so excellently with a mix of well-chosen illustrative examples and higher-level explanations. In a very small nutshell, a large body of research suggests that once a question becomes entangled with partisan identity, evidence stops being weighed on its merits and starts being weighed for its fit with the position one’s group already holds; we largely uncritically accept information congenial to our political identities, while carefully scrutinizing information that would push us to uncomfortable conclusions. Crucially for Roeber’s purposes, this is not an obstacle to knowledge that individuals can overcome by improving their personal epistemic hygiene, so long as they are embedded in testimonial networks where lots of other people engage in politically motivated cognition (which we all are). Even if I do not engage in politically motivated reasoning, if my only potential routes to knowledge on some matter ineliminably go via testimony from lots of people who do, I’m epistemically in the lurch.

The reason this is hard to crystallize into a premise/conclusion style argument about the limits of our knowledge as I tried above is that there’s no crisp independent characterization of the sorts of questions where we’ll typically be led astray by politically motivated cognition. The best we can do, I’d suggest, is describe the sorts of unreliable cognitive processes Roeber is discussing, and then let the target of the skeptical argument be, roughly, “whichever beliefs we’re not in a position to know, due to the operation (or potential operation among our testimonial sources) of such processes.” When we try to independently characterize the target—in terms of argumentative usefulness, political controversy, or something else—there will inevitably be some mismatch.

Is it a problem for Roeber’s argumentative strategy if we cannot independently identify the beliefs that, if he’s right, do not amount to knowledge? I do not think so. To motivate this view, I’ll start from where Roeber ends—with the idea that the best we can aspire to in thinking about contentious political questions is educated guesses, rather than knowledge.

3 Political humility for Bayesians

There’s a rich tradition in epistemology of focusing on degrees of belief or confidence, rather than outright belief or knowledge. To an applied epistemologist working in this tradition, questions about political epistemology will tend to be better posed not as “do we know this political claim or not?” but instead as “how confident should we be in this or that political claim?”Footnote ⁹ Whether any viable such approach can completely eschew talk of knowledge is, unsurprisingly, contentious,Footnote ¹⁰ but there are modest—and I think, quite plausible—versions of the view on which, once you are asking about how well supported P is by some body of putative evidence, you are already treating P as something that is at best probable, and not known.Footnote ¹¹ Call this broad orientation “Bayesian.” While Roeber’s main aims in the book are negative, he also presents a positive picture of how we should think about political questions once we acknowledge we have very little knowledge in the domain, and what he says sounds, to my ears, pretty congenial to Bayesians; Roeber’s “educated guesses” sound a lot like the Bayesian’s non-extreme credences, updated in light of evidence.

While the Bayesian did not need to be convinced that we do not have much political knowledge, Roeber’s arguments still have important lessons for them. His case for the unreliability of political testimony amounts to reason for the Bayesian to treat political testimony as weaker evidence than she would otherwise have done. A Bayesian who takes on board Roeber’s lessons will not just confine herself to educated guesses about politics—which she was already doing—but will make more tentative educated guesses; her credences concerning political matters will be less extreme.

But here’s something the Bayesian would not have expected from the outset: a clean rule specifying some precise threshold of confidence beyond which credences about political matters may not go. Considerations bearing on how much to trust a type of evidence do not, in general, come with sharp boundaries; they bear on different questions with different force, and how far they push depends on the messy particulars of each case. So the Bayesian is unsurprised to find that Roeber’s considerations resist any tidy characterization of just which beliefs they undermine, and by how much. If we go looking for a general characterization anyway, we can certainly find one, but it will be trivial. Political testimony that P is weak evidence, the Bayesian will say, exactly when you’d have been about as likely to hear it whether P were true or false, and strong exactly when you’d have been likely to hear it only if P were true. Fair enough; but this is no special fact about politics. It’s an immediate consequence of how the Bayesian thinks about evidence of any kind, and it would be just as true of testimony about sports, or the weather, or the contents of my neighbor’s refrigerator.

The value of Roeber’s discussion lies elsewhere. It’s that he shows, case by case and mechanism by mechanism, how easy it is, across a wide range of political questions, to end up hearing that P even when P is false. That this range of cases admits no precise boundary is not a defect in the demonstration. The lesson—be more tentative the more your evidence resembles the sort of evidence politically motivated testimony can serve up independently of truth—is one that applies in degree, claim by claim, with no bright lines on offer. From this perspective, the “counterexamples” considered earlier—e.g., concerning Clinton’s emails—should not seem particularly embarrassing to Roeber’s view. Sometimes—often, in fact!—testimony on political matters is of a sort we’d be highly unlikely to get if the target claim were false.

I’ll close with an analogy. A common lesson people draw from Wittgenstein’s (Reference Wittgenstein1953) discussion of games is that most of our concepts aren’t best characterized by necessary and sufficient conditions, but instead are something more like prototypes.Footnote ¹² I suggest that Roeber’s arguments are most forceful when understood not as giving necessary and sufficient conditions for the impossibility of political knowledge, but instead as identifying a set of prototypes of political ignorance—cases where we might have taken ourselves to be in a position to have political knowledge, but in fact are not. The prototypes are useful because lots of cases where we might take ourselves to know—or, if we are Bayesians, might be highly confident—will resemble the prototypes to a great enough degree that we might reasonably back off, either to a mere educated guess (if we thought we knew) or to a more tentative educated guess (if we started out Bayesian). This is a useful and important contribution even if, as with prototypes generally, just how closely a given case resembles them is a vague and contestable matter that resists characterization in terms of tidy criteria.Footnote ¹³