Among the many millions of words that Barack Obama and Hillary Clinton uttered over the past decade are these two statements. Obama: “They get bitter, they cling to guns or religion …” Clinton: “you could put half of Trump’s supporters into what I call the basket of deplorables.” These remarks are widely interpreted as evidence of the politicians’ authentic, private opinions about millions of Americans who are rural, white, lower-income, and tend to vote conservatively.

This method of analysis is practiced across the spectrum. In the midst of a long forum in Iowa, Joe Biden recently said that “poor kids” are “just as talented as white kids,” apparently revealing a hidden worldview in which poor = people of color and success = whiteness.

Michael Kinsley famously defined a “gaffe” as “when a politician tells the truth – some obvious truth he isn’t supposed to say.” Kinsley later clarified that the gaffe reveals “the truth about what he or she is really thinking” (not the truth about the world).

If you are trained to think in terms of representative samples, this method seems invalid. Take a large random sample of, say, Barack Obama’s public comments, and you will not find any pejorative comments about rural Americans. The “cling-to-guns” remark is constantly quoted because it is a statistical outlier. As Kinsley asked (criticizing his own concept): “why should something a politician says by accident — and soon wishes he hadn’t, whether true or not — automatically be taken as a better sign of his or her real thinking than something he or she says on purpose?”

In fact, there is a plausible theoretical reason to interpret gaffes as evidence of sincere beliefs. Let us assume that many individuals hold stable private beliefs about important topics, such as white rural voters or children of color. They realize that some of these beliefs are best kept to themselves. What they believe is unpopular and likely to be condemned. So they exercise mental discipline to block themselves from saying what they believe–most of the time. The problem is that we also have a tendency to state what we do believe. That tendency sometimes defeats the individual’s self-censorship, and out pops a gaffe.

You would not expect a sincere but impolitic belief to be common in the speaker’s discourse. It would not appear with statistical frequency, because self-censorship is pretty effective. But an anomaly is revealing. Why did Biden utter his remark about poor kids and white kids unless, in his private thoughts, poor = minority?

I summarize this theory because I think it can be valid in some cases, and I would not rule out the practice of pouncing on gaffes. But it is worth considering some alternative theoretical frameworks:

Perhaps in addition to some stable private beliefs, we also hold many unstable beliefs–ideas that come and go, that we half-believe or only occasionally believe, that we believe even though we also believe their opposite, that we adopted habitually early on but have sincerely rejected since then, or that we believe until we consider their logical implications, at which point we drop them.

Perhaps there are other common speech acts besides stating a sincere belief or not stating that belief. For example: trying out an idea that you’re not sure is true, saying something that you disbelieve by pure accident, saying something purely for its rhetorical affect, or saying something that you half-believe because you’re trying to make some other point that is salient for you at the time.

Perhaps what we believe is rarely stable because we are strongly influenced by the immediate context, by what we happen to notice at the moment from amid the Jamesian “blooming, buzzing confusion” of the world.

Perhaps conversation is highly relational, so that often what we’re doing when we talk is responding to a discussion partner. Responsiveness can turn into hypocrisy when we say one thing to one audience and a different thing to a different audience, just to win their favor. But responsiveness is also a virtue. Particularly if you consider a topic that isn’t politically or ethically loaded, it can be praiseworthy to be able to say different things to different people, just because you care about them.

To the extent that these theories obtain, deriving information from a gaffe is invalid.

See also character understood in network terms; stability of character; responsiveness as a virtue; marginalizing odious views: a strategy.