Martin Gilens and I agree that because of their similar policy preferences, both median-income and affluent respondents “fairly often get the policies they favor.” ^{Footnote 1} We also agree that “democracy by coincidence” can hold normatively concerning implications (refer to both of our conclusions). We disagree, however, on the extent and implications of this coincidental representation.

In his reply, Gilens states, “There simply is not enough coincidence of policy outcomes and middle-class preferences to justify the conclusion that middle-income Americans are likely to be satisfied with the policies their government adopts.” First, I should clarify that I am not arguing that median-income Americans are consistently satisfied with government policies. My partisanship analysis suggested that between 2000 and 2004, median-income (and perhaps affluent) respondents who identified as strong Democrats should generally not have been satisfied with the policies adopted. My claim is that those in the economic middle should generally be about as satisfied with policy outcomes as the affluent. Interestingly, Gilens’ response supports this claim. Although Gilens concludes that “the affluent are twice as likely to see the policies they strongly favored adopted,” we must remember that this conclusion comes from an analysis of a narrow subset (5.1 percent) of his data. If the rate of policy congruence for middle-income respondents equaled that of the affluent for these proposed policies, we would only expect 20 different policy outcomes between 1981 and 2004, which represents just 1.1 percent of the proposed policies in the data.

Omar Bashir’s analysis of Gilens’ data further illustrates this point. ^{Footnote 2} Bashir shows when a majority of median-income respondents support (oppose) a policy change and a majority of affluent respondents oppose (support) the change, median-income respondents see their desired outcome 47 percent of the time (compared to 53 percent for the affluent). Of course, some of these cases may be important to the average citizen (Gilens cites increasing the minimum wage and trade policy with Japan as examples), but in Gilens’ data, even when the median and affluent disagree, policy change aligns with the preferences of ordinary Americans almost as often as it aligns with the preferences of the affluent. Ironically, the failure of media to recognize this point could depress political participation. Instead of emphasizing these similarities between the median and the affluent, we have seen headlines like “The Politics of Always Ignoring What Average Americans Want.” ^{Footnote 3} It is possible that these exaggerated headlines reduce feelings of political efficacy, which in turn reduce political involvement, and thus the actual influence of those in the middle. ^{Footnote 4}

Gilens’ critique of my analysis also illustrates the prevalence of coincidental representation. Gilens suggests that my figure 6 does not account for the fact that “the set of policies that would fall at the high and the low ends of the preference distribution would differ if we were considering a scenario in which the middle-class rather than the affluent shape policy outcomes.” This is not the case. If those in the middle strongly favored or opposed different policies than the affluent, the solid black line in figure 6 would follow a different trajectory—as it does in the partisanship analysis depicted in my figure 7. Instead, the policy preferences of median-income and affluent respondents tend to align at the both ends of the preference distribution. ^{Footnote 5} Gilens also critiques the functional form I impose and the confidence intervals I estimate. The functional form comes from Gilens’ analysis, so I believe it was an appropriate choice. In his work, Gilens has not reported uncertainty estimates around predicted values. If we follow Gilens and ignore this uncertainty (i.e., only consider the absolute difference between predicted probabilities and the quadratic fit in my figure 6), my conclusion still holds. ^{Footnote 6}

My analyses built on the idea of “relative policy support.” I regret that this discussion came across as “vague” and “half-articulated.” As my article notes, it has become ubiquitous to conclude that a preference gap (i.e., the percentage favoring a policy differs across two groups) implies unequal representation. My discussion of relative policy support was intended to show that even when preference gaps exist, representation as defined by Gilens can still be equal because the two groups’ support for the policies can—and often does—follow the same relative preference ordering. If both groups’ relative preferences are identical (i.e., the preference ordering is the same), regardless of which group policymakers follow, the probability of policy change will be highest (lowest) for each group’s most (least) preferred policies. Scholars cannot treat preference gaps as synonymous with unequal representation.

Although my article focused on income groups, the importance of relative policy support applies to any analysis of groups. Jan Leighley and Jonathan Nagler, for example, present data on support for five redistributive policies among voters and nonvoters. ^{Footnote 7} Because nonvoters are 8 to 14 percent more supportive of these policies, they suggest that redistributive policies would be more generous if turnout were higher. Despite this preference gap, relative policy support is identical for voters and nonvoters. Both groups are most supportive of increasing the minimum wage, followed by government health insurance for children, government health insurance for workers, more federal assistance to schools, and finally, making union organizing easier. ^{Footnote 8} Should we expect that union organizing would be made easier if turnout out were higher because nonvoters are 12.4 percent more supportive of this policy? Or should we expect union organizing policy to remain the same because out of five prominent redistributive issues, voters and nonvoters both express the least amount of support for changing this policy? If minimum wage were increased, would this better represent nonvoters who express more support for this change, or would both groups be equally represented because this was the most preferred policy change (for all 22 policies considered) for voters and nonvoters? I propose that we must consider support for union organizing and minimum wage relative to support for other policies.

In addition to moving beyond preference gaps as an indicator of unequal representation, understanding why policy preferences do not differ more by income is a crucial avenue for future research. Equally important, although there are many reasons to expect the affluent to have disproportionate political influence, ^{Footnote 9} we must not overstate inequality in representation. Whether we consider how often they get their preferred policy or the probability of policy change under various counterfactual scenarios, according to Gilens’ data, the economic middle does almost as well as the affluent. This is not to say we should be normatively satisfied with representation in the United States, but the average citizen should feel much more political efficacy than media coverage of Gilens’ work would suggest. Indeed, a variety of research shows that, depending on the context, the economic middle (and sometimes even those at the lower end of the economic distribution) can receive as much representation as the wealthy. ^{Footnote 10} To the extent those in the middle are not satisfied with policy outcomes, policy responsiveness to the ninetieth income percentile is unlikely to be the primary cause. Perhaps, as Gilens notes, we need to look beyond the ninetieth income percentile to the influence of the truly rich (i.e., the richest 1 percent or .1 percent). However, given the similar policy preferences of the fiftieth and ninetieth income percentiles, it may be that the truly rich influence policy at the expense of the middle and those whom Gilens’ considers the affluent. While this scenario differs substantially from that proposed by Gilens, the implications for unequal representation would be even more extreme.