Book
Affluence and Influence: Economic Inequality and Political Power in America
Gilens collects 1,779 national survey questions (1981-2002) that ask whether Americans favor a specific change in federal policy, breaks each result out by respondent income, and checks whether the change was adopted within four years. He reports that outcomes track the preferences of the 90th income percentile, and that where the 10th or 50th percentile differs from the 90th by more than 10 points, their link to outcomes is statistically indistinguishable from zero. Later chapters repeat the test by policy domain, interest-group alignment, election cycle, party control, 1964-2006 trends and gridlock, and the last chapter ties the pattern to campaign money. He states that the data cannot reach the very rich and that the book measures responsiveness to preferences, not whether policies are wise.
Who made it
- Author Martin Gilens is a political scientist. The acknowledgments place him at Yale in the 1990s, UCLA from 2000 and Princeton from 2003 (Acknowledgments).
- Publisher is Princeton University Press, co-published with the Russell Sage Foundation; the copyright page names both.
- Funding named in the acknowledgments: the Institute for Social Science Research and Academic Senate at UCLA, the Bobst Center and the Committee for Research in the Humanities and Social Sciences at Princeton, and the Russell Sage Foundation, which with the Institute for Advanced Study also provided sabbatical funding (Acknowledgments).
- The acknowledgments say Benjamin Page, Larry Bartels and Kay Schlozman read the entire manuscript, and thank Bartels as director of Princeton's Center for the Study of Democratic Politics. Page and Bartels are authors of the pilot survey of the wealthy that the book cites as pending.
- Hardcover 2012; paperback 6 April 2014, 352 pages (Princeton University Press page). Publisher-listed awards: 2013 Woodrow Wilson Foundation Award (American Political Science Association), 2016 AAPOR Book Award, Choice Outstanding Academic Title 2013.
- Gilens is co-author of the 2014 Gilens and Page article (exhibit Exhibit 693) and wrote replies to Enns (2015) and Bashir (2016).
Its claims checked against the record
15 claims checked:4 supported · 11 partly supported · 0 contradicted · 0 not in the record
A verdict rests on the documents named in its row.
| The work says | Verdict | What the record shows |
|---|---|---|
| Government responds to the public, but responsiveness is strongly tilted toward the affluent, and under most circumstances the preferences of the vast majority of Americans have essentially no impact on which policies are adopted. | Partly supported | The book's logit coefficients rise from.31 (10th percentile) to.37 (50th) to.49 (90th) (Table 3.1, p. 76). On proposals where the groups differ by more than 10 points, the coefficient is.02 (p =.85) for the 10th against.46 for the 90th, and -.01 (p =.93) for the 50th against.47 for the 90th (Table 3.2, p. 79). A descriptive count on the same data, 185 cases with opposing majorities, gives the middle-income side its way 47 percent and the affluent side 53 percent (Bashir; Branham, Soroka and Wlezien as reported by Elkjaer and Klitgaard; accepted by Gilens and Page 2016). The 185-case set and the book's more-than-10-point sets (723 and 322 questions) are different subsets, and no document in the register applies the book's coefficient test to the 185. |
| When less well-off Americans' preferences diverge from the affluent's, responsiveness to the well-off stays strong and responsiveness to lower-income groups all but disappears; most of the association between policy and middle-class or poor preferences comes from preferences they share with the affluent. | Partly supported | The book says low- and high-income respondents fall within 5 points on about one-third of proposals and middle and high within 5 points on about half, and that the poor and middle class are as likely as the affluent to get what they want when they agree with them. Its appendix reaches the same answer by a second method, a deflated-covariance OLS in which the 90th percentile is.51 and the 10th and 50th are not distinguishable from zero (Table A3.3). The divergence half is the disputed part: 322 of 1,779 questions is 18 percent, the subset where Enns finds r =.68 and policy 'about where we would expect if policymakers represented the middle class'; Bashir's simulation returns a near-zero middle-income estimate in over 20 percent of trials at a.78 correlation. |
| The impact of preferences on policy is clearest at high levels of support or opposition; a 20-point difference matters little when opinion is evenly divided and a great deal when opinion leans heavily one way. | Supported by the record | The book's observed adoption rate rises from 11 percent for proposals favored by fewer than one in ten respondents to 60 percent for those favored by nine in ten, and is flat from about 25 to 55 percent support. Of 240 proposals favored by 80 percent or more, 48 percent were adopted; of the evenly split, 28 percent; three-in-four support brings change 39 percent of the time and three-in-four opposition blocks it 78 percent. The 2014 article's 'about 43 percent' for 80 percent majorities is the book's predicted probability at 80 percent support (Table 3.1), not an observed rate, and Bashir's 47 and 52 percent at 80-plus support cut the same dataset differently. |
| Differential responsiveness is not caused by income-group differences in the reliability of preference measures, breadth of opinionation or strength of opinion. | Partly supported | The book's own numbers: estimated reliabilities of.77,.80 and.77 for the 10th, 50th and 90th percentiles; poor respondents about twice as likely to say 'Don't know' but no difference between middle and upper incomes; no difference in 'strongly' versus 'somewhat' across 160 questions; similar shares of lopsided questions (two-thirds or more on one side) at every income level. The reliability estimate comes from the 387 questions with alternate versions, and the appendix says that error structure may not carry over to subsets of the data. The critics' rival explanation is correlated error and better information among high-income respondents, which the book addresses only for the full dataset by deflating covariances by 14 to 19 percent; its own first multivariate model gave negative coefficients for the 10th and 50th percentiles and a 90th-percentile coefficient that doubled from.51 to 1.01. |
| The association between preferences and policy reflects the public's influence over policy more than policy makers' influence on the public's preferences. | Partly supported | The chapter's evidence is indirect. Education, a closer proxy for attention to elite cues than income, produces a similar gradient (.37 to.50 against.31 to.49), and income matters more than education when the two are separated (Figure 3.9;). The rest is cited work, not tests on the book's data: only 29 percent of Americans can name their House member, and presidential budget proposals track opinion most when the president needs public support. The body says elite persuasion is 'not a likely explanation'; the Introduction's 'likely accounts for the bulk' is stronger than any estimate in the chapter, and no test of which side moves first appears in it. |
| Unequal responsiveness is reproduced in each of four issue domains (foreign policy, religious or moral, economic, social welfare), always strongest for the highest incomes. | Partly supported | The book reports that responsiveness is strongest for the 90th percentile in every domain, with inequality highest in foreign policy, lowest in social welfare and the widest affluent-to-middle gap in economic policy. The domain samples are 389 economic and tax, 399 social welfare, 428 foreign policy and 161 moral or religious questions. In the economic domain, the one outside test, Elkjaer and Iversen's fiscal and redistributive analysis, finds policy aligned with the middle class; it uses preferred spending levels and 21 democracies, so it differs in measure and scope from the book's estimates. |
| Social welfare is a partial exception: policy tracks poor and middle-income preferences on Social Security, Medicare, school vouchers and public works because powerful interest groups (AARP, the teachers' unions, pharmaceutical and construction interests) align with them. | Partly supported | The book names the four issues after reviewing the outcomes and then tests the same 399 social welfare questions: for those issues the decline in responsiveness as preferences diverge is -.08 (10th) and.08 (50th) with p =.34 and.35, against -.53 and -.39 (p =.02 and.05) for the remaining issues. It names AARP, the AFT and NEA, pharmaceutical companies and the construction industry as the allies, calls most of the alignment 'a happy coincidence' and not the poor's influence, and notes that interest-group positions are not more aligned with lower-income preferences in the domain overall. |
| Interest groups are a parallel channel of influence: alignment is strongly related to outcomes, but it does not explain representational inequality, and when interest groups and the affluent disagree, policy typically does not change. | Partly supported | In the book the Net Interest Group Alignment Index has a logistic coefficient of.36 (s.e..05) and correlates.02 to.04 with public preferences at every income level; adding it leaves the income coefficients at.29 to.49 (Table 5.2), and engagement and interaction terms are not significant (Tables 5.3-5.4). Of 72 proposals favored by interest groups but opposed by two-thirds of the affluent, 16 were adopted. The index counts group positions, not dollars: 43 groups from Fortune's Power 25 lists plus ten industries, coded by the author from public statements, with three added 'phantom' abortion-rights groups, and the book concedes that groups on winning sides are perceived as more powerful. |
| Responsiveness to all income groups is highest in presidential election years, the boost is greatest for the least well-off and inequality is lowest then; responsiveness to the well-off rises in the first years after control of the presidency changes party, much less for the middle class and not at all for the poor. | Supported by the record | The overall coefficient is.35 in nonelection and congressional election years against.65 in presidential election years. On proposals where preferences diverge, responsiveness to the poor and middle class is absent in nonelection years and significant only in presidential election years, while the affluent are best represented in every year. In a new presidential regime's first Congress, on divergent proposals, the 10th percentile shows essentially no responsiveness and the middle class some; by the third Congress responsiveness is about half the first. These are the book's own estimates; nothing in the register tests or contradicts them. |
| Both parties favor the affluent, and the poor's responsiveness is not higher under Democrats; representational inequality is greater and responsiveness to all income levels lower under Democratic than Republican control, except in social welfare, where responsiveness to the poor is greater under Democrats and to the affluent under Republicans. | Partly supported | The book finds responsiveness strongest for the affluent under either party and the preference-policy link over twice as strong under maximum Republican as maximum Democratic control. Its own text splits the inequality claim: affluent-to-poor inequality is somewhat greater under Republican control and affluent-to-middle somewhat greater under Democratic, so the Introduction's 'inequality is greater under Democratic control' is not what chapter 6 reports. The result rests on seven years of unified Democratic and two of unified Republican control, the rest divided; dropping the Johnson years raises the Democratic estimate from.25 to.40 and leaves a nonsignificant Republican edge. Redistributive proposals show a party effect: adoption of downward redistribution.31 under Democratic against.15 under Republican control, upward.07 against.39. Elkjaer and Klitgaard's US studies and Lax et al. find Democrats overrepresent the poor relative to Republicans, on roll-call measures. |
| Responsiveness to the well-off rose steadily over four decades while responsiveness to other income levels did not; an evenly divided Congress and gridlock increase responsiveness to all groups. | Partly supported | The body says all income groups show stronger responsiveness at the end of the period, but only the affluent show a steady rise; for the poor and middle class there is little change from the 1960s to the early 1980s and an upward trend after, so the Introduction's 'did not' is overstated. Gridlock raises responsiveness (interactions significant for the 50th and 90th percentiles only) and a small Senate seat advantage raises it for the middle class and affluent; on divergent proposals the poor get little responsiveness under high gridlock and under any Senate margin. Elkjaer and Iversen's stable middle-class result is for fiscal policy with a different measure. |
| Responsiveness was very low under Lyndon Johnson and, with unified Republican control, again in 2005-06; it was highest to all income levels in the evenly divided early years of George W. Bush, when the Bush tax cuts and estate tax repeal were supported by majorities at every income level. | Supported by the record | The Bush-years coefficient is 1.03 for all questions and.84 without foreign policy; 2001-02 responsiveness is far above 2005-06 even excluding war and terror questions; the book says neither affluent nor other Americans had more influence in 2005-06 than at the height of Democratic control in the mid-1960s. The Johnson estimate rests on a limited number of 1964-68 questions and 'is subject to considerable uncertainty'. For the 2001 cuts the book reports 55 percent support at the 10th percentile and about 68 percent at the 90th, and 68 and 79 percent for estate tax repeal. The tests are the book's own regressions. |
| The affluent in the book are the 90th income percentile (a 2010 household income below $135,000 for 90 percent of households), not millionaires or billionaires; the book lacks data on the truly rich, whose preferences are somewhat more extreme but follow the same pattern. | Supported by the record | The book states it cannot reliably estimate the 'tiny sliver' at the top and that the influence it finds at the 90th percentile might reflect the preferences of the 99th or 99.9th. Its only evidence on the very rich is Page and Hennessy's 132 General Social Survey respondents in the top 4 percent, who match the top third on most issues and are more extreme on many. The 2013 Chicago pilot measures preferences and activity and not outcomes; Gilens and Page 2016 and Enns each say the data cannot separate the 90th percentile from the very rich. |
| Using actual policy outcomes in place of roll calls captures agenda-setting power and decisions made behind closed doors, and the dataset is neither too broad nor too narrow, including important proposals political actors kept off the agenda. | Partly supported | The book builds its set from survey questions found by keyword search ('oppose') in two poll databases: 1,923 questions, cut to 1,779 after dropping proposals needing a constitutional amendment or Supreme Court ruling, partial changes and questions without income breakdowns. It includes items Congress never took up, such as a five-year immigration ban, and says the survey agenda tilts away from obscure issues and may omit options the less well-off prefer, giving federal job creation as an example. Average support for the polled changes is 54.8, 56.0 and 56.3 percent at the 10th, 50th and 90th percentiles. The article by the same author states the data cannot separate agenda blocking from rejection, and Enns concedes off-agenda policies cannot be tested. |
| Donating money is the only form of political participation (voting, campaign work, donating) that mirrors the pattern of representational inequality, so affluent turnout and campaign involvement do not explain disproportionate responsiveness; money is the root of representational inequality. | Partly supported | Chapter 8 compares participation by income across different surveys (turnout from the 2000 Current Population Survey; campaign work and donations from the 1988 American Citizen Participation Study) and finds only donating tracks the pattern (Figure 8.1;). The chapter calls its evidence 'circumstantial' and the explanation 'highly plausible', calls Bartels's money evidence 'indirect', and says the role of money 'is complex and far from completely understood'. No contribution or lobbying-spending variable enters the book's preference-and-outcome regressions in chapters 3-7, so the Introduction's 'money is the root' is stronger than the chapter's own wording. |
What the record shows that it leaves out
- Campaign contributions and lobbying spending are not variables in the preference-to-policy regressions (ch. 3-7). The interest-group index counts how many of 43 groups took a position, not what they spent; the book's money argument is a comparison of participation types in chapter 8, which it calls circumstantial.
- Polled proposals only, found by a keyword search for 'oppose' in two poll databases. The book says narrow industry-versus-industry fights are unlikely to be polled and that its sample excludes the obscure issues that attract lobbying; it could not link its data to Baumgartner and colleagues' 98 randomly selected lobbying issues. Baumgartner's synopsis reports resources explain less than five percent of the difference between winners and losers (exhibit Exhibit 689).
- Who gains from the policy, as distinct from whose preferences it matches. The book states it does not judge whether preferences are wise or policies good (Introduction p. 3;); Elkjaer and Iversen 2020 (exhibit Exhibit 702) use a dataset on who gains from fiscal policy and find alignment with the middle class.
- Income outcomes by party of president are cited, not analyzed. The book mentions Bartels's finding that growth was stronger at every income level under Democratic presidents and that post-tax inequality was flat under Democrats and rose under Republicans (;); the party chapter measures responsiveness to preferences. Bartels 2006 (exhibit Exhibit 704) reports real family income at the 20th percentile growing 2.6 percent a year under Democratic and 0.6 percent under Republican presidents, 1948-2001, with the 95th percentile doing equally well under both.
- The very rich. The book says it has no sufficient data above the 90th percentile and calls the Page-Bartels pilot pending. Page, Bartels and Seawright 2013 (exhibit Exhibit 708) interviewed 83 households in the refined sample; Gilens and Page 2016 and Enns each state the very rich may hold influence the 90th-percentile proxy picks up.
- Descriptive win rates in the disagreement cases. The book reports regression coefficients for proposals with preference gaps above 10 points and no count of how often each side won; the 47 and 53 percent figures for 185 opposing-majority cases come from Bashir and Branham and colleagues (exhibits Exhibit 697, Exhibit 700) and are accepted by Gilens and Page 2016.
Published responses
- Omar Bashir, Research & Politics (October-December 2015): Reproduced the Gilens-Page 2014 main table from the authors' input diagram, simulated the estimator (a true median-income coefficient of about 0.4 was estimated as essentially zero in over 20 percent of trials at a.78 correlation), and counted outcomes in 185 cases where the groups' majorities split: 47 percent for the median-income side. States that the 'oligarchy' label comes from journalistic accounts and that his review does not show the wealthy lack advantages.
- Martin Gilens, Research & Politics (April 2016): Reply to Bashir arguing the critique is misplaced; described through Elkjaer and Klitgaard 2024 (p. 322).
- J. Alexander Branham, Stuart Soroka, Christopher Wlezien, Political Science Quarterly 132(1), 2017: On the roughly 1 in 10 cases where the middle and high-income groups hold opposing preferences, the rich get their way 53 percent against 47 percent (as reported by Elkjaer and Klitgaard 2024, p. 320). Gilens and Page 2016 accept the 47 and 53 figures.
- Peter K. Enns, Perspectives on Politics 13(4), December 2015: Using the Gilens data (1,836 questions by his count), finds the 50th and 90th percentile support correlated at r=.94 (r=.68 where groups differ by more than 10 points, 18 percent of policies) and policy where 'we would expect if policymakers represented the middle class and ignored the affluent'. Affirms the status quo bias in the data and warns against an overly sanguine portrait of inequality; concedes the data cannot test off-agenda policies and that the truly rich may influence policy at the expense of both groups.
- Martin Gilens, Perspectives on Politics 13(4), December 2015: Agrees with Enns that congruence without influence is 'democracy by coincidence' and has political implications; argues that the policies departing from middle-class preferences are important, highly salient issues and that there 'simply is not enough coincidence of policy outcomes and middle-class preferences'. Only p. 1065 was read.
- Gilens and Page, Washington Post Monkey Cage (23 May 2016): Accept the 47 and 53 majority win-rate figures; argue slim majorities should not equal overwhelming ones and report 46 percent adoption for affluent-only strong support against 24 percent for middle-class-only; say the 90th-percentile proxy may attribute to the affluent influence that belongs to the truly rich.
- Mads Elkjaer and Torben Iversen, World Politics 72(2), 2020: Simulate that a regression of policy change on preferences for change can credit the rich with influence when they are better informed; re-test on preferred spending levels and find policy aligns with middle-class interests in fiscal and redistributive policy. State this does not mean inequality is adequately addressed and that critics who make the middle class a sometimes-coequal do not explain why the rich so often get their way.
- Larry Bartels, Unequal Democracy and PS January 2006: Reports Senate responsiveness for 1989-1994 to the middle and upper thirds and none to the bottom third, and describes Gilens's then-unpublished work as a parallel finding. Bartels says his result covers 'a single governmental institution in a single six-year period'. The book thanks Bartels as a manuscript reader (Acknowledgments).
- Mads Elkjaer and Michael Klitgaard, Perspectives on Politics 22(2), 2024: Systematic review of 25 studies and 1,163 estimates: a positive income gradient with large variation, extreme results arising more often in models that enter several income groups together, and a conclusion that the degree of differential responsiveness 'remains highly contested'. Describes Gilens 2012 as finding 'just a slight income gradient' across all policies but responsiveness only to the affluent 'when preferences differ by more than a few percentage points'.
- Where critics and author agree: Gilens, Enns, Bashir, Branham et al. and Gilens and Page agree that (1) the middle-income and affluent groups want the same thing in about 9 of 10 polled cases, so middle-income Americans often get the policies they favor; (2) in the 185 divergence cases the middle-income side wins 47 percent and the affluent side 53 percent; (3) congruence without influence is 'democracy by coincidence' and carries normative weight; (4) the data cannot separate agenda blocking from rejection; (5) the 90th percentile is not the very rich, who may hold influence the proxy attributes to the affluent. They disagree on whether the divergence cases show a tilt big enough to call the middle class uninfluential, on how to weight narrow against strong majorities, and on whether the multivariate model or separate models describe responsiveness better.
Where it connects
- Whose Preferences Pass: The book's 1,779-question dataset and its 90th-percentile result are what the chapter sets against Bashir, Enns, Branham et al. and the 2024 review.
- The Return on Lobbying: Chapter 5 of the book tests interest-group alignment against income-group preferences, and the audit sets it beside Baumgartner's lobbying-and-policy-change study.
- Money Without Names: The book's last chapter ties unequal responsiveness to campaign donating, which is the channel the chapter's donor-disclosure records bear on.