Polls, Adam Berinsky warns, are flawed. The source of this flaw is what he labels “exclusion bias,” defined as “the exclusion of the preferences of the sometimes sizable portion of the public who say they ‘don't know’ where they stand on the issues of the day” (p. 2).

What would happen, Berinsky asks, if those people who said “don't know,” whether for lack of an attitude or because they declined to express one, chose a side? He is prepared to answer the question. “The key to this empirical enterprise,” Berinsky writes, “is to ascribe interests to people who do not provide answers to survey questions” (p. 36). Using statistical inference, essentially regressing various demographic variables on the preferences of those who did express an opinion, then applying the coefficients to the “don't knows,” he estimates the underlying opinion of every don't-know respondent in several polls. The implications, he asserts, are great. By ignoring don't-know respondents, “under certain circumstances opinion polls may subvert the communication of the public's preferences” (p. 51).

Berinsky casts this effort in language of inclusion: “I ask what would happen if we did not draw the firm line between question-answerers and question-abstainers—what would happen if we assumed that everyone, no matter how muddled their political voice, should be heard as part of the democratic process?” (p. 49). The short answer, regardless of the statistical sophistication of the effort, is that we would put words in their mouths, words that reflect thoughts they do not have and may never have. Indeed, in the author's scheme, there is no such thing as genuinely not having a preference for any survey question ever asked.

Berinsky first explores responses to racial issues using a series of National Election Studies questions on school integration—centered on whether “Washington” should create equal racial balance in local schools. He suggests, as have many previous studies, that some respondents will feel uncomfortable expressing negative responses and will therefore offer what amount to false don't knows.

Limiting his attention to white respondents in the 1990s, the author finds that a majority who answered were opposed to the racial balance notion, but that many respondents offered a don't-know response. The don't knows were not a random assemblage of respondents; thus, he posits that their nonresponse likely changed the overall tilt of the survey result. Indeed, estimating the opinions of those who did not respond ultimately produces a 4% increase in the size of the anti-integration majority.

Berinsky attempts to validate his model by testing it against a poll relating to a known outcome. Using a New York City mayoral race that pitted a white candidate against an African American candidate, he applies his don't-know correction to a preelection poll showing the African American candidate 13% up in a race he would win days later by less than 2%. His model reduces the African American candidate's margin by 1.4%. In other words, the model narrowed the mistake of the poll but was still 10% off reality. He takes this improvement as vindication for his model, but one must be cautious given the sizable remaining disparity.

One of the curiosities of Silent Voices is that for all of his attention to false don't knows, Berinsky has little interest in false directional responses. Surely false answers, not just false nonanswers, occur when social desirability is a factor. Similarly, he has little interest in people who completely refuse to participate in polls or cannot be reached by pollsters.

The author next applies his model to questions on social welfare policies in the 1990s and on the Vietnam War in the 1960s. Here, he theorizes not a wariness to express one's opinion but a difficulty in forming an opinion as the source of nonresponse. Again he finds that nonresponders were not a random group. Although they leave Berinsky unfazed, the variables he finds most associated with nonresponse (lack of ideology for social welfare questions, lack of information about Vietnam for war questions) serve to solidify the impression that there is such a thing as legitimate nonresponse. That is, it would seem that there are people who are not on the cusp of expressing a preference that is lurking within them but who are truly without a preference. Nevertheless, after applying his model to correct for all don't knows, he finds that the social welfare and Vietnam War poll results exaggerated the size of conservative majorities by between .5% and 2%.

Berinsky notes that larger effects are visible beneath the national level. For example, with the use of 1996 NES data, opinion on social welfare questions shifts 6% among Arkansas residents when don't knows are assigned positions. Although he does not directly mention this fact, the 1996 NES had 24 respondents in Arkansas—including 11 who took no position on social welfare. This is not to make too much of what is by any measure a quite minor observation in the book—but it goes to the very point of what he's arguing. The reason Berinsky says we need to pay attention to don't knows is that individuals are being left out, skewing the results of polls away from what the people really believe. The overstatement of a majority position by half a percent not being terribly compelling, he is left touting such things as the unexpressed beliefs of 11 Arkansans.

Nevertheless, Berinsky argues that the problem of uncorrected nonresponse “serves inadvertently to disenfranchise individuals of a particular ideological or political type” (p. 140), creating “the potential to fracture the ongoing communication between representatives and their constituency [that], in the process, may subvert effective political representation” (p. 83).

One is left to wonder when that could ever happen. Surely it has not in the context of the three areas Berinsky examines—where by his estimation, if don't knows had expressed opinions, the size of the majority would have been only marginally changed.

Berinsky has a legitimate concern that something is lost when many respondents choose “don't know,” but neither his notion of the scope of the problem nor his response to it seem particularly realistic. Public opinion scholars will benefit from considering his arguments under the heading of a theory of survey nonresponse, a small corollary to John R. Zaller's (1992, The Nature and Origins of Mass Opinion) theory of survey response.