Mechanisms

Did welfare reform transform public aid in a way that improved public views of aid recipients, the poor, and welfare policy (Path A)? This prediction receives little support from the survey data. Here, it is important to return to the distinction between public satisfaction with a policy and feedback effects that yield changes in underlying political attitudes. In 2001, about 61% of those who knew about welfare reform said it was working well and, of these, 64% said the most important reason was that “the law requires people to go work” (NPR/Kaiser/Harvard). Beyond this satisfaction, however, we find considerable stability in perceptions of welfare, welfare recipients, and the poor. (Unless otherwise noted, results can be obtained from the cited polling organization and Bowman 2003.)

A work-promoting policy design does not seem to have allayed public anxieties about welfare dependency. In 1989, 64% of the public thought “welfare benefits make poor people dependent and encourage them to stay poor” (Gallup). We have no identical question for the TANF period, but available data do not suggest that reform banished negative beliefs about welfare dependency. In 2003, 71% of Americans agreed that “poor people have become too dependent on government assistance” (Pew). In 2001, 2002, and 2003, a steady 40% of women and 50% of men agreed with the hostile statement that “most welfare recipients are lazy cheats” (DDB Needham Worldwide). In 2002, 53% of Americans stated that “the current welfare system” makes things worse “by making able-bodied people too dependent on government aid” (Pew). Perhaps most remarkable in light of work-conditioned aid, 49% endorsed the statement that “poor people today do not have an incentive to work because they can get government benefits without doing anything in return” (Greenberg, Quinlan, Rosner).

Wording differences make it impossible to know how strongly these results indicate stability. But responses to questions with consistent wording offer corroborating evidence. For example, “In your opinion do you think most people who receive money from welfare could get along without it if they tried, or do you think most of them really need this help?” The “could get along without it” view was endorsed by 40% in 1986 (CBS/NYT), 44% in 1992 (NBC/WSJ), and 44% in 2001 (NPR/Kaiser/Harvard). Similarly, work requirements have not diminished the belief that lack of effort explains why poor people are poor. “In your opinion, which is more often to blame if a person is poor—lack of effort on his part, or circumstances beyond his control?” From 1988 to 1990, 35% to 40% of Americans endorsed “lack of effort,” whereas 37% to 45% endorsed “circumstances” (Gallup). Between 1998 and 2001, 40% to 48% chose “lack of effort,” whereas 38% to 45% chose “circumstances” (Gallup; CBS; NPR/Kaiser/Harvard). NES “feeling thermometer” scores also allow us to assess changes in Americans' feelings toward welfare recipients and the poor. For welfare recipients, the average score “warmed” from 50.7 (1980–1990) to 53.6 (1998–2004). By contrast, the score for the poor “cooled” from 72.5 (1980–1990) to 70 (1998–2004). These changes are statistically significant. But they are cross-cutting and, with standard deviations running in the 15 to 20-point range, they are not substantively large.

Did policy reform negate welfare as an object of public concern (Path B)? The National Election Studies (NES) time series contains two relevant measures. The first indicates the percentage of respondents each year who volunteered “welfare” when asked, “What do you think are the most important problems facing this country?” From 1976 to 1986, welfare was named by 8% to 12% of NES respondents (mean $=$ 9.2). In the political lull between the Family Support Act of 1988 and Clinton's pledge to end welfare, this number fell to 4.7 in both 1988 and 1990. The reform campaign produced a large spike, starting at 7% in 1992, rising to 16.8% in 1994, and then 26.6% in 1996. After reform, salience declined to 8% in 1998 and then 4.8% in 2000. The NES series for this item ends here. Thus, we are left to conclude only that welfare's salience in 2000 (95% C.I. = 3.4 to 6.2) was discernibly lower than in the 1976–1986 period and equivalent to the 1988–1990 period.

Fortunately, the NES series also allows us to follow the number of respondents each year who volunteered “welfare” in response to any variant of the question, “Is there anything in particular that you [like/dislike] about the [Democratic/Republican] party?” The observations here are slightly higher but follow a similar pattern. From 1976 to 1986, “welfare” was named as a basis of party evaluation by 7.3% to 17.7% of respondents (mean = 12.3). After remaining high at 17.1% in 1988, we see a dip in 1990 (11.9) and 1992 (11.4), followed by a spike in 1994 (23.2) and 1996 (19.6). In 2000, the percentage (10.4%; 95% C.I. = 8.5 to 11.6) falls below the AFDC-period norm and, in 2004, it falls to 4% (95% C.I. = 2.8 to 5.3), a level well below the low point under AFDC (7.3 in 1978).

Causal inference here is complicated by the possibility that welfare was pushed off the agenda after 2000 by other issues—most notably, national security and the Iraq War—rather than by the abolition of AFDC. Indeed, other domestic spending issues with little relation to PRWORA also declined in salience between 2000 and 2004. For example, although Social Security was a major domestic policy issue during this period, its salience dropped between 2000 and 2004 at a rate (7.8 to 2) almost identical to that of welfare (10.4 to 4). Thus, it seems plausible that agenda crowd-out accounted for some of welfare's declining visibility. Nevertheless, a variety of competing issues came and went in the pre-reform era, and none neutralized welfare's salience to this extent. Although outstanding causal questions remain, we consider these results consistent with the negation variant of the progressive revisionist thesis.

In sum, then, we find evidence of issue negation (Path B) but no transformation of views on welfare policy, welfare recipients, and the poor (Path A and, hence, Paths C and E). After 1996, welfare faded from the public agenda, but underlying images of welfare and policy-relevant groups shifted little.

The Indirect Channel: Deracialization

The General Social Survey (GSS) includes a number of measures that can be used to assess the prediction that welfare reform would deracialize poverty and welfare politics. Although this diversity precludes the satisfaction of a single definitive test, it allows us to avoid an overly tidy conclusion that hinges on the validity of a single measure.

We can begin by examining changes over time in the extent to which white preferences for spending on blacks are related to white preferences for spending on welfare and the poor. To control for demographic differences over time, we estimate the relationships with controls for age, sex, education, marital status, and family income. (For full models underlying all results reported in this paper, as well as supplemental analyses, please see http://www.polisci. wisc.edu/$\sim$soss/public_transformed.pdf) In the 1984–1991 period, whites who opposed efforts to “raise the living standards of blacks” were significantly more likely to oppose efforts to “raise the living standards of the poor” (b=.270, p=.00). This relationship grew stronger in the 1998-2004 period (b=.334, p=.00; test of difference t=3.4, p=.00). The relationship between preferences for “spending to assist blacks” and “spending to assist the poor” held steady across the 1984–1991 period (b=.278, p=.00) and the 1998-2004 period (b=.260, p=.00; test of difference t=-.84, p=.40). And finally, whites who opposed efforts to raise black living standards were significantly more likely to oppose welfare spending in both the 1984–1991 period (b=.168, p=.00) and the 1998–2004 period (b=.149, p=.00; test of difference t=-1.01, p=.34). Thus, white Americans' tendency to equate welfare and aid to the poor with forms of aid targeted to blacks seems not to have been disturbed by welfare reform.

In addition to policy associations, the GSS time series includes two measures that allow us to assess change in the relationship between stereotypes of black work effort and white preferences for spending on public aid. The first, available in every GSS from 1985 to 2004, asks respondents to agree or disagree that black-white economic disparities exist because “most blacks just don't have the motivation or will power to pull themselves up out of poverty.” The second, available in the GSS only between 1990 and 2004, asks respondents to place “most blacks” on a seven-point scale running from hardworking to lazy. We find that neither measure exhibits a significant change in trend related to the passage of welfare reform.

To test for changes in stereotype influence, we constructed multivariate models predicting white opposition to spending on welfare and aid to the poor. The models consist of (1) respondents' sex, age, education, family income, marital status, party identification, liberal-conservative identification, and support for individualism, (2) dummy indicators for the reform and postreform periods, with the pre-1992 era serving as the baseline, (3) one of the two stereotype measures, and (4) interactions between the period dummies and all included variables. In models that employ the laziness measure, our tests depend on an AFDC baseline that consists of only one pre-1992 year: 1990. The motivation measure, available for 6 years in the 1985–1991 period, offers a more substantial baseline for tests of pre-post differences.

Examining the sources of white welfare opposition, we find little evidence that stereotype effects weakened after welfare reform. Whites who believed that “blacks lack motivation” were significantly more likely to oppose welfare spending from 1985 to 1991 (b=.64, p=.00). The TANF Period interaction term indicates no change (b=-.02, p=.87), leaving the relationship intact in the 1998–2004 period (b=.61, p=.00). Our second specification also fails to produce evidence of deracialization. The effect of the “black laziness stereotype” is actually insignificant in 1990 (b=.11, p=.25) and increases slightly over time (b=.18, p=.11) to yield a statistically significant relationship in the postreform period (b=.29, p=.00).

Relative to “welfare,” “assistance to the poor” has historically drawn stronger public support and has been less central to racialized poverty discourses (Gilens 1999; Schram 1995). Even so, we find a significant relationship in the 1985–1991 period between white preferences for aid to the poor and the belief that “blacks lack motivation” (b=.51, p=.00). The TANF Period interaction is insignificant (b=-.07, p=.71), leaving this relationship substantially the same in the 1998–2004 period (b=.44, p=.00). By contrast, results using the “black laziness stereotype” suggest a pattern of change. In 1990, the only baseline year where we can observe it, this stereotype is a significant predictor of white opposition to assistance for the poor (b=.22, p=.03). This effect then weakens over time—not enough to make the relationship in 1998–2004 statistically distinct from the relationship in 1990 (b=-.14, p=.23), but enough to render the laziness stereotype statistically insignificant in the 1998–2004 period (b=.08, p=.19).

Because they pool years into periods, these pre-post tests benefit from larger samples in each time unit. Separate-year analyses force us to rely on samples that are smaller and more variable, but they also allow us to examine whether racialization has become less stable under the TANF program. Using this approach, we find little evidence that white welfare preferences have become less racialized. Counting results for both our GSS measures, we find that stereotypes of black effort are significant predictors in 5 of 6 separate-year analyses in the 1985–1991 period and 7 of 8 analyses in the 1998–2004 period.

For white opposition to spending on the poor, however, we find a more intriguing pattern. The Black Motivation Stereotype is significant in three of five models in the 1985–1991 period and two of four models in the 1998–2004 period. Notably, the insignificant results in the postreform period appear as the final two observations: 2002 and 2004. The samples available in 2002 (n=182) and 2004 (n=175) are small relative to earlier years, and the 2004 coefficient is comparable in size to the coefficient that is significant with a larger sample in 2000. But this weak indication of change is strengthened by results for the Black Laziness Stereotype. Significant in 1990 as a predictor of white opposition to spending on the poor, this measure yields significant results in only 1 of 4 years in the 1998–2004 period: 2000. Thus, it seems plausible that these stereotypes exerted less influence on preferences for aid to the poor after welfare reform. With similar samples, the GSS shows no similar disruption of stereotype effects for “welfare.”

In sum, we find mixed results for the deracialization hypothesis. The public remains as likely as earlier to equate welfare and generic aid to the poor with targeted efforts to assist blacks. Likewise, we find no reduction in the influence that black-effort stereotypes exert on white welfare preferences. By contrast, for aid to the poor, we find a mixed pattern: no discernible change in analyses that pool years into periods but evidence of weakening effects in separate-year analyses.

End Goals: Public Generosity and Partisan Advantage

Ultimately, the progressive revisionist thesis promised greater public support for efforts to help the disadvantaged and for the Democratic Party. One way reform might generate the first of these effects would be to strip the taint of “welfare” away from anti-poverty efforts. As we have seen, welfare retained negative connotations for large segments of the public in the postreform era: it remained associated with dependence, laziness, and aid to blacks. One critical question, then, is whether the quieting of welfare disputes weakened the relationship between disdain for welfare and resistance to helping the poor.

In the subsample of the GSS that contains the welfare preference question, two items measure willingness to help the poor: a 5-point scale indicating opposition to the idea that government “should do everything possible to improve the standard of living of all poor Americans” and a 7-point scale indicating opposition to the idea that government should “reduce the income differences between the rich and the poor, perhaps by raising the taxes of wealthy families or by giving income assistance to the poor.” To test whether these policy preferences have become less tied to welfare attitudes, we use a regression analysis that includes controls for respondents' sex, age, education, family income, marital status, party identification, liberal-conservative identification, and support for individualism. For the first measure, we find a significant relationship to welfare opposition in the 1984–1991 period (b=.307, p=.00) that, if anything, increases slightly in the 1998–2004 period (b=.345, p=00; test of difference: t=.71, p=.48). We find the same pattern using the second measure (1984–1991: b=.316, p=.00; 1998–2004: b=.388, p=.00; test of difference: t=.82, p=.41). Both results prove robust in analyses of subsamples restricted to white respondents and whites making below the median income.

Despite its lingering association with welfare, the public's willingness to invest in the poor could have risen as welfare became a less salient issue. Comparing means across periods, we find no evidence of such a shift. Opposition to improving “the standard of living of all poor Americans” actually increased significantly in the years after welfare reform (mean_1984–91 = 2.81, mean_1998–2004 = 2.98; t = 8.57, p = .00), as did opposition to reducing “income differences between the rich and the poor” (mean_1984–91 = 3.57, mean_1998–2004 = 3.83; t = 7.86, p = .00). For the key spending preferences, we find continuity. There was a negligible increase in opposition to spending on aid to the poor (mean_1984–91 = 1.42, mean_1998–2004 = 1.44; t = 1.68, p = .09). We observe no significant changes in preferences regarding aid to blacks (mean_1984–91 = 1.93, mean_1998–2004 = 1.93; t = .41, p = .68) and welfare (mean_1984–91 = 2.29, mean_1998–2004 = 2.21; t = -1.04, p = .30). All five results are robust in analyses of subsamples restricted to whites and to whites below the median income, and yearly observations corroborate the analyses based on periods.

Thus, in the postreform era, the tendency to attribute poverty to lack of effort has held steady, feelings toward the poor have grown slightly cooler, efforts to aid the poor have remained associated with “welfare,” and willingness to aid the poor has stayed the same or diminished. Finding little change in the public climate for policymaking, we turn to partisan and electoral dynamics. In the years leading up to welfare reform, revisionists argued that Democrats labored under unacceptable disadvantages as a result of their association with “permissive” welfare. By signing a tough reform bill, President Clinton would clear the way for Democrats to gain support among Americans who opposed permissive public assistance.

Figure 2 offers support for some aspects of this argument. The “balance” trend line for each party indicates the proportion of NES respondents in each year that cited welfare as a reason to like a specific party minus the proportion citing it as a reason to dislike that party. The bars represent welfare's salience to the public, as indicated by the NES “important problem” question. In most years from 1976 to 2004, the Republican Party shows a net positive evaluation on this issue; the Democratic Party shows a net negative. The changing distance between the lines clearly supports those who argued that Democrats suffered a substantial disadvantage on welfare in the late 1970s and early 1980s. At its widest in 1982, there was a 20.2-point gap favoring the Republicans. At the end of the 1980s, however, we see parallel shifts in partisan advantage and issue salience. In the wake of the Family Support Act, amid the much-discussed “new consensus” in welfare policy, public attention to welfare wanes and the Republican advantage on this issue evaporates. This state of affairs proved ephemeral, however. After Clinton reopened the issue with his 1992 campaign pledge, and Republicans responded, welfare reemerged on the public agenda and the Republican Party promptly reestablished its advantage. Indeed, the Republican issue advantage follows the trend in issue salience very closely. It reappears with Clinton's pledge in 1992; it expands markedly in 1994; it peaks in 1996; and then falls away in the years after reform.

Welfare as a Reason to Like vs. Dislike the Democratic and Republican Parties Comparison of Party Balances and Issue Salienceover Time

Several points merit note here. First, in periods of higher salience, the welfare issue redounded to the Republican Party's advantage, regardless of whether leading Democrats were defending welfare (pre-1988) or demanding its reform (1992–1996). Thus, contrary to the transformative variant of the revisionist thesis, if any party was in a position to claim credit for ending permissive welfare, it was the Republicans. Second, the periods that follow major legislative actions (1988–1990 and 2000–2004) are characterized by dissipation: public attention drops and the party balance measures drift toward each other at the neutral point on the scale. Third, public concern and Republican issue advantage were not constants under AFDC; they were variables tied to elite interaction. Fourth, whereas bipartisan legislative action neutralized the welfare issue briefly after 1988, it did not prevent the issue's resurgence or the reestablishment of partisan advantage. When elite conflict returned, so did the earlier patterns. Thus, just as the welfare issue can be taken off the table, it can be put back on. A key question, then, is whether PRWORA changed underlying political dynamics (or constrained political discourse) in a more fundamental and durable way.

The welfare issue became less salient after 1996, but did this change produce dividends for the Democrats? One way to interpret the progressive-revisionist strategy is to say that it aimed to make partisan competition less “about” the most divisive targets of social spending: blacks and welfare recipients. Success in this strategy would be indicated by an electorate in which preferences for welfare spending and assistance to blacks became less predictive of partisanship. Because opponents of such spending would feel more comfortable identifying as Democrats, the correlation between attitudes toward these “stigmatized” policies and party identification would diminish.

The GSS time series suggests no such pattern of change. Outside the South, the bivariate correlation between party identification and welfare preferences was r = .21 between 1980 and 1991, and it remained r = .21 in the 1998–2004 period. In the South, where conservatives were moving toward the Republican Party during this time, the correlation rises across the two periods from r = .18 to r = .24. A stronger version of the same patterns emerges for the relationship between party identification and preferences for spending on aid to blacks. Outside the South, the average bivariate correlation between these measures was r = .21 between 1980 and 1991; it rises to r = .26 in the period from 1998 to 2004. In the South, the correlation rises from r = .19 to r = .32. None of these patterns suggests that welfare reform helped the Democrats win over opponents of stigmatized public assistance.

Indeed, welfare opponents who lived outside the South identified with the Democratic Party at a rate that was 84% of the general electorate between 1980 and 1991 but only 77% of the electorate between 1998 and 2004. Democratic identification among welfare opponents in the South fell from 89% of the general electorate (1980–1991) to 73% of the general electorate (1998–2004). Outside the South, people who opposed aid to blacks identified as Democrats at a rate that was 79% of the general electorate in the AFDC era (1980–1991) but only 73% of the general electorate in the TANF era (1998–2004). Within the South, the rate of Democratic identification among those who opposed aid to blacks fell from 86% of the general electorate (1980–1991) to 63% (1998–2004). Thus, opponents of these two forms of spending remained just as numerous in the postreform era as they were before; survey data suggest that “welfare” symbolized the same things it had in the past to these opponents; and even if one isolates Americans outside the South, public aid opponents were at least as hesitant to identify with the Democratic Party after welfare reform as they had been before.

If we shift our attention from party identification to electoral choice, a similar story emerges. Because issues related to national security and the Iraq War dominated the election of 2004, diminishing the influence of domestic social policy issues in general, 2004 offers an uncertain basis for assessing relevant changes. Instead, we focus on the two presidential elections immediately prior to Clinton's welfare pledge (1984 and 1988) and the election four years after reform (2000). The 1984 election took place under AFDC, in a year when the welfare issue was salient and favored the Republican Party over Democrats. The 1988 election was also contested under AFDC, but in this year welfare was much less salient and the Democrats enjoyed parity on the party balance measures. The 2000 election was contested after welfare reform, in a period when welfare was once again low on the public agenda and the party balance measures were in parity. How did the Republican advantage among welfare-opposing voters shift across these elections? If we take (1) the Republican candidate's share of the two-party vote among welfare opponents, and divide it by (2) this candidate's share of the two-party vote among all voters, we obtain a ratio indicating how much better the Republican performed among welfare opponents than among the electorate as a whole. Outside the South, this ratio was 1.15 in 1984, 1.20 in 1988, and 1.19 in 2000. In the South, this ratio was 1.32 in 1984, 1.16 in 1988, and 1.21 in 2000. These patterns of stability hold even if we restrict our samples to white voters or to white voters below the median income.

In sum, we find that reform made welfare less salient as a basis for party evaluation, but decreased salience does not seem to have yielded broader political gains for the Democratic Party. Welfare opponents remained just as numerous after 1996 as in the AFDC era, and these individuals became no more likely to identify with the Democratic Party or vote for Democratic candidates.

Proposition 1

When a policy exists as a potent but distant symbol for mass publics, the details of its material design will seldom anchor public understandings. Unaffected publics will rarely pay close attention to changing policy realities and, absent the introduction of new frames of understanding, new facts on the ground will rarely force the abandonment of old myths about the state. Accordingly, the potential for program-design change to produce significant opinion change will be quite low.

Progressive revisionists were right to see “welfare” as a potent symbol, but they overestimated the extent to which welfare's symbolism depended on the AFDC program's design features and material effects. Writing in 1995, Nathan Glazer was closer to the mark when he argued, “the issue has become what welfare symbolizes, not what it is. Welfare has come to stand for the rise of a permanent dependent population cut off from the mainstream of American life and expectations…[and] for the problems of the inner-city black poor…. Ending ‘welfare as we know it’ seems to promise some relief from these social disorders” (Glazer, 21). Indeed, in the era before reform, researchers repeatedly found that public perceptions of welfare had only the loosest relation to the AFDC program's actual features. A 1994 poll, for example, found that most Americans thought that AFDC—which, at its peak, cost less than 5% of Social Security—was one of the two largest items in the federal budget (Hacker 2004a). Under AFDC, most Americans held beliefs about welfare that were demonstrably at odds with policy facts; most tended to be confident about the accuracy of their beliefs; and most tended to resist correcting their beliefs when supplied with accurate information (Kuklinski and Quirk 2000).

In such a context, legislative reform can easily change a policy's design without creating a tighter bond between the design and the policy's image among unaffected publics. In the TANF era, evidence suggests that 40% to 50% of Americans have never had a firm cognitive grip on the material nature of welfare reform. In August 1996, as the drama surrounding PRWORA reached its apex, 44% of the public said they did not know enough about the legislation to say whether President Clinton did the right or wrong thing by signing it (Weaver 2000, 338). When asked in 2001 whether welfare had been reformed in any significant way over the past five years, 50% of respondents answered either “No” or “Don't Know” (NPR, Kaiser, Harvard). Despite heavy media coverage of falling caseloads, 42% of Americans reported in 2002 that they had not seen any stories indicating that the number of people on welfare had declined (authors' data). Moreover, the 8% of the public that answered “Don't Know” when asked whether “welfare reform has been a success” in February 2002 (Future of Children, Packard Foundation) swelled to a remarkable 32% only a month later when a separate poll asked whether “the current welfare system work[s] better or worse than the system in place before 1996” (Pew).

With unaffected publics paying little attention, old narratives soldiered on. Despite the new conditionality of aid, 49% of Americans in 2003 continued to endorse the view that “poor people today do not have an incentive to work because they can get government benefits without doing anything in return” (Greenberg, Quinlan, and Rosner). Interestingly, a 2001 poll found that people below the poverty line—a more directly affected minority—were actually the most aware of welfare reform, despite their lower levels of education and news attention (NPR, Kaiser, Harvard). Thus, one explanation for the limited effects of reform is that material policy changes drew little attention from unaffected publics and, hence, forced few Americans to abandon familiar beliefs about a mythic government activity.

Ironically, this explanation suggests that issue negation—seen by some revisionists as a potent feedback mechanism—may have actually made opinion effects less likely. Once the welfare issue was off the table, mass publics became less likely to encounter new narratives about the terms of welfare and the behaviors of recipients; liberal elites had fewer opportunities to engage in strategic priming of policy considerations. Many liberal elites were happy to see the negative discourse surrounding “welfare” fade from the public stage after 1996 (e.g., Marshall 2002). Erasure, however, is not the same as counter-argument. It leaves unaffected publics undisturbed in their beliefs. There is a general lesson to be learned here: when a policy is a familiar condensation symbol for mass publics but is distant from their daily lives, the quieting of elite political arguments will typically abet the ideational status quo.

Proposition 2

Mass publics will tend to value policies of the distant-visible type less for their tangible achievements than for what they affirm about “us.” Such policies function for unaffected publics as expressions of group values, independent of their effects on targets. The act of doing—of endorsing a principle as official government policy—is valued for what is says about who we are, what we stand for, and what we expect of one another. When such affirmations challenge traditional arrangements or take sides in an existing societal conflict, feedbacks producing mass arousal become more likely. By contrast, when a new policy affirms dominant, widely held values—rewarding individuals who live up to them, organizing practices to be consistent with them, justifying state actions by reference to them, forcing individuals to comply with them, punishing individuals who deviate from them, and so on—one should expect it to reinforce rather than disrupt existing patterns of mass opinion and behavior.

This proposition suggests two distinctions ignored by progressive revisionists: (1) the symbolism of a policy action versus the symbolism of its effects and (2) state actions that reinforce consensus versus those that disrupt consensus or take sides in conflict. By recovering these distinctions, one can identify key flaws in the promise that welfare reform would rehabilitate the image of public aid recipients by giving them a “work attachment” (Kaus 1992; Mead 1992). The prediction began from a sound premise: Americans do tend to value work and use it to evaluate deservingness. From here, however, the prediction relied on a more questionable narrative. By creating a work attachment, the TANF program would move aid recipients into the more positively valued group known as “workers.” Recipients would then qualify as full members of the deserving “us” who “play by the rules.” Greater public generosity would follow.

The problem with this account is that it trains attention on eventual changes in target-group behavior while ignoring welfare reform as an expressive action in its own right. The predicted symbolic effects flow entirely from the public's observation of rising work levels (a policy effect) and not at all from the act of reforming welfare itself (a policy action). Symbolic politics theory suggests that the reverse is more likely the case when policy actions are salient and effects are distant (Edelman 1964). Welfare reform mattered to Americans primarily in the act of doing—that is, as a state action that evoked and affirmed dominant societal values. Its subsequent effects on recipients' lives and behaviors occurred off in “other” communities and elicited little interest from unaffected publics.

By identifying welfare reform as an act affirming widely held values, we gain access to several insights that help explain opinion continuity in this case. If one asks how welfare reform affirmed work and personal responsibility, one cannot answer “by celebrating welfare recipients as exemplars of these values.” Quite the contrary, the dramaturgy of this event turned on images of recipients as violators of these values (Hancock 2004; Sparks 2003). State action affirmed commitments to work and responsibility by demanding accountability from those who deviated from norms and expectations.

This context provided the frame of reference that supplied work's meaning. “Work,” in this instance, was not a prior status indicating deservingness, nor was it a trait that could simply be attached to group members. It was the behavioral standard that had been violated, the value that demanded action to hold a problematic group accountable. In this sense, the revisionist language of a “work attachment” elided a crucial distinction between policies that reward work and policies that compel work as a condition of aid. Although the former honor freely chosen behavior, the latter imply that individuals must be forced to behave. Policy actions that make a show of requiring expected behavior will tend to reinforce the belief that bad individual behavior lies at the heart of the social problem (Edelman 1977, 149).

This point also helps to explain how revisionists could be right about the popularity of welfare reform but wrong to predict that reform would improve the image of aid recipients. When policy actions affirm majority values by holding violators accountable, they will tend to produce divergent public evaluations of the policy design (positive) and the target group (negative). Majorities may like it when legislators pass tough sentencing laws that hold criminals accountable, but such policies rarely lead publics to view criminals in a positive light. Work requirements followed a similar logic.

Finally, when policy actions function as affirmations of widely held values, their mass feedbacks, if any, will tend to reinforce orientations that already prevail. Indeed, one might go further. When mass publics value a policy action primarily as a sign that government stands behind some cherished principle, the policy action will not satisfy public desires for long, even if its effect is to bring target-group behavior in line with the principle. “The reassurance must be periodically renewed” by policy actions that in the doing convey that the state's value commitments have not flagged (Edelman 1964, 193).

Proposition 4

The material features of distant-visible policies do have consequences for mass opinion but not in a straightforward way. First, the opinion effects of distant policy designs are inevitably mediated: they depend on portrayals in elite rhetoric, mass media, social conversation, and so on. Second, two general classes of material features matter for mass perceptions of public policies: (1) a policy's internal characteristics and (2) its positioning within the larger structure of a policy regime.

Humans perceive objects, not only by examining their individual features, but also by ascertaining their location within meaningful categories and contrasts (McGarty 1999). Indeed, the balance of the two depends partly on the proximity of the object. At a distance, categorical contrasts are easily deployed: the differences between Serb and Croat communities, or between the English and Chemistry departments, are perceived more sharply than the internal workings of either. With direct experience, individuated features and complexities become more salient as complements to such categorical thinking. (This, of course, is why people tend to perceive their own in-groups as more heterogeneous and out-groups as more homogeneous; see Judd and Park 1988.) Applying this general cognitive principle to the question of mass feedback, we arrive at the expectation that more directly experienced policies will tend to be perceived on the basis of relevant political categories and internal design characteristics. As the distance between policy and public grows, internal design characteristics become less apparent, leaving perception more dependent on categorical location and contrast.

In this capacity, state institutions and the structures of policy regimes may be quite important. The establishment of “separate departments of government to deal with…supposedly distinct problems” presents the public with an organizing schema that helps to fix the meanings of particular social problems, social groups, and government activities (Edelman 1977, 26). The U.S. welfare state is a case in point. Its bifurcated structure provides an institutional contrast––“Social Security versus Welfare”––that is frequently cited as a basis for public distinctions between the deserving and the undeserving (Gordon 1994; Schneider and Ingram 1997). The symbolic oppositions conveyed by this contrast have been analyzed by many scholars: contract versus charity, independence versus dependence, white versus black, masculine versus feminine, universal versus particular (Gordon 1994; Schram 1995). Indeed, recent empirical research has underscored the extent to which this contrast maps onto and derives meaning from a racial contrast between black (Welfare) and white (Social Security) (Winter 2006).

Progressive revisionists traced negative views of welfare to the internal design of the AFDC program; they ignored its institutional and symbolic location vis-à-vis Social Security. Reform changed the design of welfare, but it did not displace or reconfigure this fundamental contrast in American social politics. For most Americans, changes within the TANF program have been far less salient than the contrast between this kind of program and the kind they might make use of themselves. Indeed, Social Security's positive image remains held in place today, not just by its own policy design, but also by the idea that it is “not welfare” and its contributing beneficiaries are in some sense “the opposite” of welfare recipients. Such categorical thinking is bolstered by a diversity of contrasts in political discourse and the policy regime itself (e.g., Medicare vs. Medicaid).

Thus, we suggest that material changes to distant-visible policies will be more likely to produce mass feedback effects to the extent that they (1) classify relevant objects in new ways or (2) significantly alter the contrasts made salient by a policy regime. This emphasis on a policy regime's classification system does not restrict effects to rare moments of wholesale institutional change. For example, new legislation may shift a salient group or problem from one side of a regime contrast to another, such as when policymakers moved the “widows and orphans” of deceased breadwinners to the social insurance side of the welfare state in 1939. Alternatively, the addition of new groups or problems to one side of a contrast may change the meaning of the contrast for mass publics, such as when the War on Poverty used public aid programs for controversial new purposes related to African Americans.

As these examples illustrate, “conversion” processes, in which “existing institutions are redirected to new purposes” (Thelen 2003), can have substantial implications for regime classifications. Alternatively, “layering” processes may add new institutional categories on top of old ones in ways that alter classification (Thelen). Or the salience of a classification scheme may change through “drift” if a static pattern of policy development renders old categories irrelevant to new problems (Hacker 2004b).