Surveys of the proportion of Americans who are atheists show an incidence of between 3% and 11%, but of course those are often phone surveys, and people may be reluctant to divulge their nonbelief. That probably means that there are more atheists than those who admit it. And that’s the conclusion of two psychologists from the University of Kentucky, Will M. Gervais and Maxine B. Najle, who have a new paper up on “psyarxiv” that used a questionnaire to answer the question (link and reference below; I’m not sure if the paper has yet been accepted anywhere).
Their conclusion came, which came from two surveys of 2000 American adults is this: “[A]theist prevalence exceeds 11% with greater than .99 probability and exceeds 20% with roughly .8 probability”. . . “our most credible indirect estimate is 26% (albeit with considerable estimate and method uncertainty).”
They used a survey method I was unaware of: the “unmatched count technique”. This method involves giving people a list of personality and behavior traits, one of which was either “I believe in God” or “I do not believe in God”, depending on whether the survey asked respondents to identify the traits that “are NOT true for me” (“I believe in God”) or “are true of me” (“I do not believe in God”). There were two lists, one including the God statement and the other omitting it; these were given to two independent groups of people. A third and independent group was asked to self report whether they believed in God. Here’s an example of the “negative” survey from the paper:
Note that there are nine items in column 2 and ten in column 3, which adds the “I believe in God” item that you’re suppose to consider whether it’s among those statements not true of you.
The incidence of atheism can then be gauged by simply looking at the difference in the number of statements given at the bottom of the two columns. Bayesian analysis of the data can then give you an estimate of the proportion of atheists in the sample. They also did a “positive” survey with six versus seven items that are supposed to be true of you. (There’s a control for credibility based on a math question, but you can read about that in the paper.)

The results (authors’ wording, my emphasis); the numbers in brackets are the 95% confidence limits from the Bayesian analysis:
1). Sample I’s unmatched count data revealed atheism rates much higher than existing self-reports suggest: the most credible indirect measure estimate from Sample I is that 32% [11%, 54%] of Americans do not believe in God, Figure 1.
2). Sample II included a conceptual replication effort of Sample I’s indirect estimate by comparing the baseline and critical conditions. Sample II also included an additional condition assessing validity of the indirect count technique by comparing the baseline and mathematical impossibility conditions.
Sample II yielded an indirect atheism rate estimate of 20% [6%, 35%], Figure 1. This atheism estimate is lower than that in Sample I. Speculatively, this difference may reflect (among other things) a difference in how participants respond to positive versus negative framing of the unmatched count tasks. That is, Sample II primarily differed from Sample I in that it included a positive affirmation of atheism (agreeing with the statement “I do not believe in God”) rather than a more passive denial of theism as in Sample I.
3). Our aggregate analysis, pooling across samples, provided an indirect atheism prevalence rate of 26% [13%, 39%]. Unsurprisingly, this estimate is intermediate between both samples’ individual point estimates, but with a tighter range of plausible values than either alone.
The estimate of self-reported atheism in the survey is 17% (error limits 14% and 20%), which is, as expected, lower than the indirect reports, but still higher than previous estimates. This may reflect either a sampling issue or the fact that Americans are more likely to say they’re atheists on paper than in a telephone survey. But the upshot is that as many as one in four Americans may be atheists.
Now the paper is a fair one, and does highlight its problems; read it for yourself. It also shows that the estimate of atheism is, as most of us know, higher among men than among women, among Democrats and Independents than among Republicans, and increases with level of education. The authors’ conclusions:
Existing nationally representative polls indicate that atheist prevalence is relatively low in the United States, perhaps only 3% (Pew, 2015) to 11% (Gallup, 2015). Given the heavy stigmatization of atheism (Edgell et al., 2006), we hypothesized that many atheists might be reluctant to disclose their disbelief to pollsters. We therefore deployed two nationally representative samples in an attempt to indirectly measure atheist prevalence using the unmatched count technique (Raghavarao & Federer, 1979). These indirect measures suggest that roughly one in four (26%) American adults may be atheists—2.4 to 8.7 times as many as telephone polls (Gallup, 2015; Pew, 2015) suggest. This implies the existence of potentially more than 80 million American atheists. The disparity between self-report and indirectly measured atheism rates underscores the potent stigma faced by atheists (Edgell et al., 2006; Gervais, 2013), as even in an anonymous online survey, about a third of American atheists may be effectively “closeted,” even in anonymous telephone polls.
The lesson for us: Atheists, while still heavily stigmatized in America, are increasing in number. Most will not admit it for obvious reasons. But the more of us willing to declare our nonbelief, the more likely it is that those in the “closet” will come out. So declare your atheism—loudly and proudly.
__________
Gervais, Will M, and Maxine B Najle. 2017. “How Many Atheists Are There?”. PsyArXiv. March 3.