Self-Interest and Attitudes

Scholars of public opinion have long grappled with the role of self-interest in policy attitudes—that is, the extent to which attitudes reflect considerations that “bear directly on the material well-being of individuals’ private lives” (Sears et al. Reference Sears, Lau, Tyler and Allen1980, 671). While some major theories of political behavior (e.g., Hobbes Reference Hobbes1973) assume a dominant role of self-interest in politics, many studies have found that the actual influence of self-interest on attitudes is limited, and secondary to “symbolic” factors like ideology and other-regarding or “sociotropic” concerns (Franko, Tolbert, and Witko Reference Franko, Tolbert and Witko2013; Funk Reference Funk2000; Huddy, Jones, and Chard Reference Huddy, Jones and Chard2001; Lau and Heldman Reference Lau and Heldman2009; Legerski and Berg Reference Legerski and Berg2016; Sears Reference Sears1997; Sears and Funk Reference Sears, Funk and Mansbridge1990; Sears, Hensler, and Speer Reference Sears, Hensler and Speer1979; Sears et al. Reference Sears, Lau, Tyler and Allen1980). A less optimistic strain of scholarship (e.g., Bartels Reference Bartels2008; Hicks, Jacobs, and Matthews Reference Hicks, Jacobs and Matthews2016) questions self-interest from the perspective of macropolitical outcomes, given that nonwealthy voters do not seem motivated to protect their own economic interests. Miller and Ratner (Reference Miller and Ratner1998), in a series of experiments, find that people overestimate the importance of self-interest to others’ attitudes. Similarly, Kim (Reference Kim2014), in a critical essay, argues that self-interest does not live up to our perceptions of it as a motivating force.

Other recent studies have found a more clearly defined role of narrow self-interest in policy attitudes, particularly when self-interest is seen in terms of labor market risk and instability (Alt and Iversen Reference Alt and Iversen2017; Cavaillé and Trump Reference Cavaillé and Trump2015; Chong, Citrin, and Conley Reference Chong, Citrin and Conley2001; Cusack, Iversen, and Rehm Reference Cusack, Iversen and Rehm2006; Darke and Chaiken Reference Darke and Chaiken2005; Doherty, Gerber, and Green Reference Doherty, Gerber and Green2006; Franko Reference Franko2016; Gingrich and Ansell Reference Gingrich and Ansell2012; Häusermann, Kurer, and Schwander Reference Häusermann, Kurer and Schwander2016; Margalit Reference Margalit2013; Owens and Pedulla Reference Owens and Pedulla2014). Even in these cases, however, the influence of self-interest is far from overwhelming, and it seems only to emerge under specific circumstances: when the stakes of a policy for the individual are large, clear, and salient (Chong, Citrin, and Conley Reference Chong, Citrin and Conley2001; Doherty, Gerber, and Green Reference Doherty, Gerber and Green2006).

To be sure, the importance of self-interest to political behavior, such as voting and mobilization, is fairly well-established (e.g., Campbell Reference Campbell2002). This may be due to the fact that self-interest can serve to increase the importance of attitudes (Boninger, Krosnick, and Berent Reference Boninger, Krosnick and Berent1995; McCombs Reference McCombs1999), if not necessarily the content of those attitudes (see also Miller and Ratner Reference Miller and Ratner1998). Self-interest also seems to affect issue attention, another necessary component in political mobilization (Bolsen and Leeper Reference Bolsen and Leeper2013).

At the macro level, much of the scholarly and political attention to self-interest is concerned not so much with individuals as with groups. While idiosyncratic self-interest concerns might affect individual behavior and attitudes in any number of ways, such concerns only matter politically insofar as they affect the attitudes and behavior of groups. This is reflected in much of the aforementioned literature comparing self-interest with other-regarding motives, from studies that sought self-interest effects among white Americans on the issue of school busing (Sears, Hensler, and Speer Reference Sears, Hensler and Speer1979) to comparative welfare state research analyzing the response of different economic classes to growing inequality (Alt and Iversen Reference Alt and Iversen2017). This focus on groups introduces a complication, however—affinity for in-groups is a powerful force in public opinion in its own right—empirically similar to but conceptually distinct from self-interest (Conover Reference Conover1984, 764). Disentangling narrow self-interest, or concern with one’s own material circumstances per Sears et al. (Reference Sears, Lau, Tyler and Allen1980), from sympathy or concern for those similar to the individual is difficult to do in typical observational and experimental research designs.

The questions of self-interest and in-group affinity loom large in the case of senior citizens and “entitlement reform” in the United States. It has become a well-worn cliché to refer to Social Security as the “third rail of American politics” (a phrase often attributed to former House Speaker Tip O’Neill, or one of his aides [Safire Reference Safire2007]), and threatening Medicare, the national health insurance program for seniors, is considered just as politically dangerous. These programs are politically strong, the story goes, because senior citizens in the modern era are politically strong (meaning well-organized and politically active) and motivated to defend them based on self-interest (though again, much academic literature questions this point—see Campbell (Reference Campbell2009), Jacobs and Shapiro (Reference Jacobs and Shapiro1999), Lynch and Myrskylä (Reference Lynch and Myrskylä2009), and Rhodebeck (Reference Rhodebeck1993); see Campbell (Reference Campbell2002) for compelling evidence that self-interest motivates seniors’ mobilization, if not their attitudes). In recent years, this has put Republican politicians in an awkward position—while their policy preferences point toward fundamental reform of old-age “entitlement” programs (both to contain costs and to reduce the role of government in Americans’ lives), they must grapple with the fact that the Republican Party is increasingly reliant on older Americans for support.

One way of squaring this circle is to offer to exempt or “carve out” current or soon-to-be seniors from the pain of reforms. Former President George W. Bush included such an age-based carve-out in his 2005 proposal to privatize Social Security, and former Speaker Paul Ryan included one in his proposal to convert Medicare into a “premium support” program.Footnote ¹ In cases like these, the rubber meets the road on the question of self-interest as opposed to symbolic or sociotropic concerns. Do such carve-outs work in building support among older Americans for major policy change?Footnote ² The failure of Bush’s Social Security plan to garner support among seniors suggests some reason for doubt—Gallup polls conducted before and after the State of the Union address in which the plan was announced suggest that older Americans had a much stronger negative reaction than others (Jones Reference Jones2005). Perhaps seniors did not know or understand that the plan would not have affected them and reacted out of their own perceived self-interest. Alternatively, it could be that in-group affinity for other older Americans who will soon come to rely on the same old-age programs as current retirees contributes to those retirees’ hostility to such policy changes, even if the retirees themselves would be held harmless.

While seniors’ self-interest has long been a subject of attention and debate in American politics, recent developments have raised the salience of self-interest for a much younger group of citizens. The rapid growth of college tuition and student loan debt (Ciluffo Reference Ciluffo2017) has led to growing demand among young Americans for some form of relief. Recognizing this demand, virtually every major Democratic politician has endorsed a higher education affordability initiative (from debt relief to free college), and discussions of what to do about the issue have loomed large in the 2020 Democratic presidential primary (Kreighbaum Reference Kreighbaum2019; Nova Reference Nova2019).

There has been a clear assumption on the part of these politicians that championing this issue would draw greater support from young voters. Indeed, the Democratic candidates who have staked out the most aggressive positions on college affordability and debt (Bernie Sanders and Elizabeth Warren) have drawn the greatest support from younger voters. Attributing causation is difficult, however, as the left-leaning ideology of these candidates is in sync with younger Americans in general, who are much more likely to support radical change than older Americans, even within the Democratic Party (Levitz Reference Levitz2020). High support among the young for free college proposals could be evidence of ideology rather than self-interest or of sociotropic concern born out of recent experience: perhaps those struggling with student debt want to help those younger than themselves to avoid falling into the same circumstances.

Austerity in social insurance for the elderly and student debt relief have something in common—major changes along these lines may not apply to all Americans, even within the relevant subpopulations. Just as social insurance austerity proposals have typically involved a “carve-out” for current or soon-to-be retirees for political reasons, attempts to grapple with the student debt problem may have to focus on a subset of debtors (or potential future debtors) for fiscal reasons. This means that both policy areas are ideal for isolating the true effect of self-interest, since in both cases it is reasonable to assume some sort of arbitrary age threshold above which one is either exempted from austerity measures or excluded from debt relief. The two cases also offer variation in terms of the age of the relevant subgroups and of the nature of the potential impact on those affected (costs versus benefits), which increases the confidence with which consistent results can be generalized.

Hypotheses

Based on the existing literature and my own theoretical expectations, I generate a set of hypotheses on the effects of narrow self-interest considerations on policy attitudes.Footnote ³ Note that since austerity reforms visit costs, while student debt relief delivers benefits, the expected effects are in opposite directions across the two cases for each hypothesis.

The first and most basic expectation is that proposals that either minimize the population subject to austerity or maximize the population receiving benefits will be more popular than those that visit austerity broadly or bestow benefits narrowly. Both self-interested and sociotropic concerns point in this direction.

Of course, for self-interest to be a meaningful factor in policy attitudes, concern about how a policy would affect oneself must have an effect beyond the general preference for less austere or more generous policies.

The second hypothesis is necessarily focused on true self-interest: the exclusion or inclusion of the individual him or herself. One goal of this study is to distinguish between such “true” self-interest and sociotropic concern for others similar to oneself in situations where oneself is not directly affected. This latter dynamic is labeled “social identification” by Fowler and Kam (Reference Fowler and Kam2007). This leads to the third hypothesis.

Which of these two forces—true self-interest or social identification—is more important to policy attitudes? Here, the existing literature is somewhat ambiguous—some work emphasizes the importance of other-regarding considerations over self-interest (Funk Reference Funk2000; Huddy, Jones, and Chard Reference Huddy, Jones and Chard2001), particularly when individuals are considering others like themselves (Conover Reference Conover1984, 764), while some studies suggest that self-interest could override other concerns in situations like this where the stakes are large and clear (Chong, Citrin, and Conley Reference Chong, Citrin and Conley2001). The next set of hypotheses expresses these competing expectations.

In addition to sociotropism, “symbolic” considerations like left-right ideology are often found to be more important to policy attitudes than self-interest. On the other hand, ideology may not be a strong predictor of policy attitudes in every situation. In the present cases, both status quo Medicare (Norton, DiJulio, and Brodie Reference Norton, DiJulio and Brodie2015) and alleviating student debt (Demos 2018) are broadly popular ideas that are not especially polarized. This leads to a second set of competing hypotheses.

True self-interest effects, if they exist, should vary predictably across individuals. Specifically, the literature suggests that self-interest effects are largest when the stakes are high (Chong, Citrin, and Conley Reference Chong, Citrin and Conley2001). This certainly seems to be the case when it comes to mobilization among seniors surrounding Social Security, as Campbell (Reference Campbell2002; Reference Campbell2003) finds that poor seniors, who rely on the program the most, are the most active. Similarly, Ozymy (Reference Ozymy2012) finds that poorer college students are more active on student loan issues than their higher-income peers. If this is also true at the level of attitudes, we should observe larger self-interest effects among those who would be placed at greatest risk by cuts or who would stand to benefit the most from new benefits. In the present study, I focus on specific forms of insecurity that would make a respondent more reliant on the policy in question (concerns about economic security and affording medical expenses in retirement and concerns about being able to pay back student loans).

Research Design

Many existing studies seek to quantify self-interest by making observational comparisons of policy opinions across demographic and economic groups (e.g., Alt and Iversen Reference Alt and Iversen2017; Lau and Heldman Reference Lau and Heldman2009; Rhodebeck Reference Rhodebeck1993; Sears et al. Reference Sears, Lau, Tyler and Allen1980; Tedin, Matland, and Weiher Reference Tedin, Matland and Weiher2001). While these studies contribute valuable evidence, the pitfalls of this strategy are clear: since many other factors differ between demographic groups aside from self-interest, the resulting estimate could be an artifact of these other factors. Control variables offer only a partial solution, since they can only account for factors that are measured and measured accurately. The appeal of a randomized experiment, of course, is that it can potentially offer an estimate of self-interest that is uncontaminated by measurable and unmeasurable differences between economic and demographic groups.

When one considers the reality of group-based affinity, however, designing an effective experiment becomes difficult. If I feel a special affinity for my social group, a randomized experimental prompt that tells me a given program will help (or harm) my group may activate my affinity for my group as well as my narrow self-interest. In this sense, the experimental treatment would “bundle” unintended considerations with the self-interest considerations the researcher intends. One possible solution would be to tell the respondent that she, personally, would be helped (harmed), while others like her would not. While this approach could produce interesting results, it suffers from a lack of realism (public policy does not exempt specific individuals) and makes the intent of the experiment obvious to the respondent (the researcher wants to know how selfish you are) in a way that could trigger social desirability bias.

In this study I exploit the real-world use of age-based carve-outs in policy proposals and use carefully recruited samples in narrow age ranges to avoid the pitfalls of these other strategies. The basic approach is to present respondents within a narrow age range with a description of a policy proposal that would confer benefits or burdens on people in their demographic and economic group, which randomly includes a birthdate-based carve-out that falls roughly in the middle of the age range of the group, excluding all people born before that date from change. This results in four groups: (1) born before the cutoff, carve-out; (2) born before the cutoff, no carve-out; (3) born after the cutoff, carve-out; and (4) born after the cutoff, no carve-out.

Groups (1) and (2) are the only groups that have a self-interested stake in the presence or absence of the carve-out, but the possibility of group-based affinity means that self-interest is not necessarily the only relevant difference between groups (1) and (2): the carve-out may also activate considerations about others similar to the respondent. This is why groups (3) and (4) are crucial to the design: they offer a test of the effect of the carve-out when self-interest considerations are not present, allowing me to estimate the effect of group-based affinity and subtract it from the difference between group (1) and (2).

Subject to some modest and realistic assumptions, then, the difference in treatment effects between older (groups (1) and [2]) and younger (groups (3) and [4]) respondents represent a “clean” estimate of self-interest. The key assumption is that group-based affinity considerations are the same across the birthdate-based cutoff, meaning that people born on one side of the cutoff identify with people born on the other as much as they do with someone born on the same side. My approach of recruiting very narrow age ranges (two to three years) and using arbitrary cutoffs makes this assumption much more plausible: my baseline feeling of social connection to people born before December 31, 1959 should not be meaningfully different if I myself was born in 1959 or 1960.

This study consists of two experimentsFootnote ⁵ embedded in online surveys of American respondents in narrow demographic categories appropriate to the selected policy proposals: 58- to 60-year-olds for the Medicare reform study and 23- and 24-year-olds who reported having student loan debt for the student debt relief study (see below for more information on these samples). In addition to ensuring that group-based affinity is similar across respondents, sampling from these narrowly defined and directly relevant populations is more appropriate for the study of self-interest than sampling from the national population.

In each experiment, respondents read a description of the policy proposal and were asked to rate their approval or disapproval on a 7-point scale.Footnote ⁶ In the randomly selected control groups, the description stipulated that the proposed change would apply to all eligible Americans regardless of age. For the treatment groups, the description included a cutoff date; individuals born before this date would not be affected by the change. The cutoff dates fall roughly in the middle of the age range of each sample, meaning that about half of respondents in the “carve-out” condition in each experiment would be personally affected by the described change while about half would not.

The wording of the two experimental questions is as follows, with the experimental manipulations in bold. The Medicare reform proposal is based on former Speaker Ryan’s proposal, while the student debt relief proposal is based on one of Senator Sanders’s 2016 campaign proposals.

Some further discussion of question-wording choices is in order. In designing both prompts, I took steps to avoid potential confounding via “background assumptions.” As Dafoe, Zhang, and Caughey (Reference Dafoe, Zhang and Caughey2018) argue, observed effects in a survey experiment may be due not to the direct or intended effect of the treatment on the dependent variable but to the treatment’s effect on important assumptions that also affect the dependent variable. They demonstrate a number of such scenarios, including that respondents assume a country described as a “democracy” in an experimental prompt is also majority white and Christian. Thus, in order to ensure that experimental prompts are truly affecting the dependent variable in the hypothesized manner, it is important to fix information for respondents. I have done so in two ways.

First, the control prompt explicitly informs respondents that all eligible Americans would be affected rather than offering a “true” control that excludes all information about eligibility. I chose this approach to ensure that respondents in control did not simply assume some sort of carve-out provision would be included, as this is a common feature of policy proposals (particularly entitlement reform proposals). In experimental terms, such respondents would fail to “comply” with the treatment (or control, as the case may be) and bias estimates of the true causal effect downwards. Second, I have also made the stakes explicit for beneficiaries or potential beneficiaries in both experiments—Medicare reform would visit costs on those affected, while student debt relief would extend benefits. This guards against the possibility that respondents in treatment and control groups would make different assumptions about the consequences for those affected.Footnote ⁹

Both samples were recruited with the aid of a survey research firm (SSI, now Dynata). For the Medicare study (conducted April 19–28, 2016) the sample consisted of 2,020 US-located respondents born between the dates of January 1, 1955, and December 31, 1957. Of these respondents, 1,206 were born after the cutoff date of December 31, 1955.Footnote ¹⁰ For the student debt relief study (conducted July 12–22, 2016), the sample consisted of 1,215 US-located respondents born between January 2, 1991, and December 18, 1993, all of whom reported owing at least some student debt.Footnote ¹¹ Of these respondents, 635 were born after the cutoff date of June 30, 1992. By targeting these relatively precise ranges of people with a potential stake in the described policies, I ensure that self-interest considerations, where operative, are clear and salient. Respondents were randomized into treatment and control conditions as described above, without regard to age. I also worked with Dynata to recruit two similar but smaller samples (roughly 500 respondents each) for a pair of follow-up studies in March of 2020, which I discuss briefly in the text and in more detail in Appendix E.

The dependent variables in the analyses that follow are respondents’ approval of the described policy measured on a 7-point scale from “strongly disapprove” to “strongly approve.” The key independent variables are indicators of treatment group and respondent age and interactions between the two.

Testing the ideological hypotheses (H5a and H5b) requires a measure of ideology, which I capture with a 7-point scale (“conservatism”) ranging from “very liberal” to “very conservative.”

Testing H6 on policy reliance requires measures of respondents’ financial circumstances and the extent to which the described proposal would either put them at greater risk or alleviate that risk. For the Medicare study, I asked respondents about their self-reported health, their confidence in their ability to afford their medical costs in retirement, and their confidence in affording a comfortable retirement in general.Footnote ¹² These ordinal items scaled well (alpha = 0.78), so I rescaled each to range from 0 to 1 and added them together to create a single measure of health and retirement insecurity, a continuous variable ranging from 0 to 3 (the mean and median of this measure are both 1.5, with a standard deviation of 0.74). For the student loan study, respondents answered a question on whether they agree or disagree (on a 5-point scale) with the statement “I will be able to pay off my student loans.”Footnote ¹³

Both surveys also included a battery of demographic and political questions to use as control variables where necessary. These including party identification, gender, income, education,Footnote ¹⁴ and race/ethnicity. The full text of both surveys can be viewed in Appendices A and B, and descriptive data for both samples can be viewed in Appendix C.Footnote ¹⁵

Discussion and Conclusion

This study makes a number of contributions to our understanding of the attitudinal effects of self interest in general and the effect of “carve-out” provisions in particular, the latter being an under-studied phenomenon in politics and public policy.

First, both experiments make clear that narrowly defined self-interest can indeed play a role in attitude formation as well as behavior and mobilization. Consistent with existing literature, I find effects in a situation where the personal stake in policy is clear and salient, and these effects are more pronounced when individual need is greatest (though the findings of the follow-up study suggest the latter results may not be robust). In terms of substantive size, these self-interest effects dwarf group-based altruism (i.e., concern for others affected by a cutoff) and ideological concerns, though caution is warranted with respect to generalizing the latter finding, given that ideological effects were small in both studies. It appears that the most pronounced effects of self-interest came at the extreme ends of the approval scale—increasing or reducing the proportion of respondents who selected “strongly approve” or “strongly disapprove”—rather than at the threshold between support and opposition. This is consistent with past findings that self-interest motivates behavior more clearly than attitudes—one possible mechanism for this dynamic would be that self-interest increases attitude intensity and therefore mobilization (Boninger, Krosnick, and Berent Reference Boninger, Krosnick and Berent1995; McCombs Reference McCombs1999).

How relevant are these findings to politics in the real world? While “carving out” is a policy and political strategy that elected officials do pursue from time to time, and while the experimental prompts presented here were designed to be as realistic as possible, they were still presented to respondents in an artificial setting. In real life, members of the public learn about policy proposals from politicians, the media, and interest groups seeking to mobilize them in support or opposition—the way these elites present this information, and the way that individuals perceive these elites, is very important in shaping public opinion (Zaller Reference Zaller1992). Would the effects documented here really matter in a more realistic political scenario? I argue that there is reason to believe that they would. Again, existing work (Boninger, Krosnick, and Berent Reference Boninger, Krosnick and Berent1995; McCombs Reference McCombs1999) has established self-interest as a mobilizing force, and the nature of these attitudinal effects is consistent with that—the largest differences between treatment and control involve the extremes of the 7-point scales. To use the Medicare findings as an example, the likely real-world effect of such a carve-out strategy would not be to change the minds of those opposed to austerity, but to blunt some of the fiercest opposition, which would, in turn, reduce the likelihood of mobilization from current retirees. Perhaps a card-carrying member of the American Association of Retired Persons who would otherwise have responded to the organization’s call to write his representatives in opposition to an entitlement reform plan might decide it is not worth his while after finding out that the plan would not affect his own bottom line, even as he continues to believe the plan is bad policy. Perhaps a liberal Democrat with student loan debt who would have joined a march or rally to demand passage of a relief plan that would help her would be less eager to do so after finding out that the plan carves her out.

Moreover, it is easy to see how carve-outs could matter not just as another covariate in a model of attitudes that also includes partisan or interest group elite cues, but as a rhetorical shield or bludgeon employed by those same elites. A Republican proponent of entitlement reform can promise that “current retirees will not be affected,” while a left-wing Democrat calling for a more radical approach to student debt might attack the hypothetical loan relief proposal from this study as “doing nothing for those who are already struggling with debt.” The effectiveness of such messages would depend in part on the partisan identity of the people hearing them, but these findings suggest that self-interested considerations would also magnify or dampen the effects.

Still, these findings should be kept in perspective. It bears repeating here that these populations were very narrowly and carefully selected to maximize the size of self-interest effects. While the self-interest effect in the student debt study was quite large, the effect in the Medicare study was surprisingly modest at less than one half of a standard deviation. Moreover, in both studies, self-interest was clearly far from the whole story—Medicare privatization was overwhelmingly unpopular and student debt relief overwhelmingly popular across the board.

This study also makes an important contribution to the self-interest literature in its distinction between true self-interest and affinity for those similar to oneself, a distinction many studies do not make. Failure to account for the effect of the treatment among those who would not be personally affected would have overstated the apparent effect of self-interest by about 0.08 standard deviations in the Medicare study and 0.12 standard deviations in the student debt relief study (the latter being a statistically significant difference, the former nearly so). While these differences were small, they demonstrate that empathy towards others in one’s social group can bias the measurement of self-interest effects. In the case of the Medicare study, if I had not accounted for group-oriented sociotropism, I would have concluded that the self-interest effect was about 20% larger (0.52 standard deviations rather than 0.44). In both studies, group-oriented sociotropism significantly inflated the effects of the carve-out treatment on strong approval or disapproval. Moreover, the results of the follow-up studies, which included alternative dependent variables tapping respondents’ beliefs about how the policy would affect others, suggest that the distinction between self-interest and sociotropism is surprisingly fuzzy. Together, these findings suggest caution in the measurement of ostensible self-interest effects and, more importantly, a more nuanced view of self-interest and sociotropism. It may be more appropriate to think of these concepts as opposite ends of a spectrum, with points in between representing concern for others similar to oneself rather than generalized altruism.

Of course, this study does have some important limitations. Notably, it examines two policy issues where left-right ideology (and partisanship) plays a relatively small role, at least among the targeted subpopulations who took part in the experiments. A study of more polarizing or divisive policy issues may well find that ideology and partisanship swamp any effects of self-interest on policy attitudes. Moreover, while the difference in effect size between the two studies is surprising given that much existing work in the vein of prospect theory (Kahneman and Tversky Reference Kahneman and Tversky1979) leads us to expect losses to be a more powerful motivator than gains, the numerous other differences between the two samples and studies make it impossible to assign much significance to this difference. Lastly, the findings of the manipulation checks in the follow-up study (Appendix E) suggest that self-interest considerations may affect policy attitudes through complex mechanisms (such as motivated reasoning), which this study was not designed to disentangle. These limitations point to a need for more experimental and observational research on the politics of “carving out.”