Academic journal article Political Research Quarterly

Elite Discourse and American Public Opinion: The Case of Welfare Spending

Academic journal article Political Research Quarterly

Elite Discourse and American Public Opinion: The Case of Welfare Spending

Article excerpt

Popular support for American welfare policies dipped sharply during the mid-1990s. The purpose of this article is to determine why this pronounced, but temporary, shift in public opinion occurred. We use data from the CPS National Election Studies to examine several explanations for temporal variability in citizens' attitudes toward welfare spending. Our results show that these changes follow similar variations in media content. Individual-level opinion change was also based entirely upon political motivations (e.g., ideology and partisanship) rather than economic beliefs or racial attitudes. We argue that this provides evidence, admittedly somewhat indirect, that elite rhetoric guides and shapes mass opinions.

Popular support for the welfare system in the United States has never been very high (Feldman and Zaller 1992). Nevertheless, the American public has always accepted the idea that the government has the responsibility to provide for the basic needs of the poor and less fortunate elements of society (Gilens 1999). But, public opinion turned sharply against welfare policies during the mid-1990s. Then, it rebounded back to previous levels by 2000. The purpose of this article is to determine why this pronounced, but temporary, shift in public opinion occurred. Specifically, we use data from the CPS National Election Studies (NES) to examine several explanations for temporal variability in citizens' attitudes toward welfare spending. Our results show that these changes were based entirely upon political motivations, especially individuals' ideological and partisan reactions to the predominant tone of political discourse in the mid-1990s. We argue that this provides evidence, admittedly somewhat indirect, that elite rhetoric guides and shapes mass opinions.


Many analysts gauge public opinion on welfare by examining citizens' preferences toward government social spending (Cook and Barrett 1992; Gilens 1999). Apparently, people believe that the level of public funding for welfare programs is a clear, unambiguous indicator of governmental commitment toward those policy areas.1 In fact, Jacoby (1994) argues that the term "government spending" is, itself, synonymous with welfare spending in the public mind. Specifically, individual attitudes toward spending on programs that assist needy groups (e.g., the poor, Blacks, the homeless, children, etc.) exhibit a consistent internal structure.2 In contrast, spending attitudes on other governmental programs (e.g., education, AIDS research, environmental protection, etc.) show very little coherence; people react to non-welfare policies in a piecemeal, idiosyncratic manner.

Both Gilens (1999) and Cook and Barren (1994) show that popular support for welfare varies over time. But, their analyses only extend into the early 1990s. As a result, they do not pick up an interesting pattern that occurred shortly thereafter. From 1992 to 1996, public opinion turned against government spending across the board. By 2000, however, citizens' attitudes returned to their previous (i.e., circa 1992) levels. The "dip" in public support was particularly pronounced when it came to expenditures for welfare, Food Stamps, and the poor (Weaver 2000).

Table 1 provides the empirical evidence for this temporal pattern, showing data from the 1992, 1996, and 2000 CPS National Election Studies. In each of those years, the NES asked respondents whether federal spending should be increased, decreased, or kept about the same in each of a series of policy areas. The table entries summarize the responses for separate policies across the years. Specifically, the numbers indicate the proportion who favor increased spending minus the proportion who favor decreased spending in each area. The table shows all policies that were included in the NES interview schedule in all three years. In almost every area, preferences were less favorable toward spending in 1996 than they were in 1992. The only exception to this pattern is child care, where support for spending stayed the same across the four-year period. By 2000, spending preferences in most areas increased markedly over their 1996 levels. The only exceptions to this trend involve spending on AIDS research and programs to control crime.

The two rightmost columns of Table 1 also show the sizes of the differences across successive time periods. The values for Food Stamps, welfare, and programs to help the poor are particularly interesting. These three areas show some of the largest declines between 1992 and 1996: -0.24, -0.20, and -0.14, respectively. Then, they show some of the largest increases in support from 1996 to 2000: +0.20, +0.24, and +0.08. None of the other programs show a similar pattern of sharp decrease followed by an increase up to, or exceeding, the initial 1992 levels of spending support.

As mentioned earlier, spending attitudes on these programs exhibit a coherent structure. Accordingly, the separate responses for Food Stamps, welfare, and poor people can be combined to create a single summary measure of spending preferences on public assistance programs. The three panels of Figure 1 show the distribution for this summary variable in 1992, 1996, and 2000. The changes in this variable, of course, mirror those already observed for the individual policy areas. In 1992, the distribution is unimodal and centered at the neutral midpoint of the scale, with a very slight imbalance toward the pro-spending side. In 1996, the central tendency shifts in the opposite direction. The distribution also becomes markedly asymmetric, with far more people favoring decreased rather than increased spending. In 2000, a slight anti-spending tendency remains, but overall, the distribution shifts back nearly to its original shape and location.3

Again, the major question we intend to confront is: Why do these changes occur in welfare spending preferences? One possibility can be dismissed almost immediately. The observed changes are not due to "noise" fluctuations in the responses. For one thing, there is the general nature of the trend. While the move toward decreased spending is strongest in welfare policies, it does exist in virtually all other policy areas as well. At the same time, the size of the change is quite large. The differences in the means of the preference scores-for the separate policies, as well as the summary variable-are statistically significant at any reasonable level of certainty.4 For these reasons, we are confident that the observed variable represents real change in public opinion. It is not random movement due to measurement or sampling error.

The previous literature has identified several social factors that affect public support for welfare. For example, the state of the economy may have an influence, although the exact nature of its effect remains open to disagreement. Some analysts contend that economic conditions have a direct effect. Stated simply, as times get better American citizens become more generous toward the needy; as the economy slows down, they prefer reductions in welfare benefits (Heclo 2001; Stevenson 2001; Durr 1993). Alternatively, economic conditions and welfare attitudes may be inversely related. When the economy is healthy, there is less need for people to rely on governmental assistance. When times are tough, public benefits help needy people deal with problems that are beyond their own control (Gilens 1999). Regardless of its direction, there should be some discernible connection between the economy and public opinion toward welfare.

Public reactions toward welfare are also shaped by feelings about program beneficiaries. While Americans are typically generous and sympathetic toward poor people, these feelings are delimited by a widespread belief that welfare recipients are unwilling to work or get ahead on their own. Thus, it is the "undeserving poor" themselves whose laziness and indifference toward the norms of individualism and self-reliance lead to a dependence on governmental largesse (Teles 1996; Mead 2001; Weaver 2000). Among people who maintain such beliefs, hostility toward the welfare system is a natural reaction (Mead 2001).

The situation is complicated even further by the influence of race. Many whites apparently believe that African Americans do not subscribe fully to the predominant work ethic of American culture (Bobo and Smith 1994; Mink 1998). And, popular lore grossly overestimates the number of Blacks among the welfare population (Lieberman 1998). Thus, racial feelings interact with cultural norms to accelerate negative orientations toward programs that provide public assistance to the needy (Gilens 1999).

While feelings about the economy, welfare recipients, and racial minorities help shape citizens' welfare attitudes, it is not entirely clear that they could account for the changes in these attitudes that occurred during the early 1990s. While the economy did improve, there was widespread public misperception about economic conditions (Hetherington 1996). Therefore, it is doubtful that it would generate such systematic movement in attitudes. Furthermore, there is no reason to believe that opinions about welfare recipients or African Americans changed very much during this same time period. Hence, they cannot account for systematic temporal variation in welfare attitudes.

There is one relevant factor that did vary markedly during this time period: Elite political discourse. The amount and type of attention devoted to welfare by public officials, the media, interest groups, and academics escalated dramatically by the mid-1990s (Heclo 2001; Mead 2001; Weaver 2000). In order to obtain tangible indicators of elite discussion on this issue, we examine information on welfare coverage from two prominent media sources: (1) The major television networks' evening news broadcasts; and (2) the New York Times. In order to demonstrate the overall amount of attention to this issue, Figure 2 presents two line plots showing the total number of welfare stories appearing in each source, for each year from 1990 to 2000.5 Clearly, a similar pattern occurs in each one.

First, consider the network news broadcasts which are shown in Figure 2a. From 1990 through 1993 there was very sparse coverage: A total of only 32 stories appeared on the evening news shows. From 1994 through 1996, coverage exploded, with 211 such stories appearing during this period. Then, from 1997 through 2000, the issue receded and only 30 stories were included in the broadcasts.

Figure 2b shows the yearly numbers of stories appearing in the New York Times.6 Once again, there was relatively little coverage during the early 1990s; the total for the 1990-1993 period is 319. In this case, the mid-decade peak lasted a year longer than in the broadcast media-the total number of stories from 1994 through 1997 is 1,295. And, the dropoff during the late 1990s was not quite as pronounced even though it is still clearly discernible; only 485 stories appeared during the last three years.

Along with the sheer salience of the welfare issue during the mid-1990s, the content of elite discourse on this topic also reveals an interesting pattern. The tone of the discussion became markedly more negative (Thompson and Morris 1995; Bryner 1998). Weaver (2000: 171) describes it succinctly as follows:

The crucial change in this period was in the balance of messages that Americans heard about welfare reform . . . the dominant messages "priming" attitudes toward welfare reform concerned the evils of the system rather than the welfare of children.

The empirical evidence provides unambiguous support for this statement. Figure 3 shows the results from content analyses of welfare coverage in the television evening news broadcasts and the New York Times from 1990 to 2000. In each display, the solid line traces the number of stories that present a critical or harsh picture of the welfare system. The dotted lines show the number of stories that depict a positive or sympathetic view of welfare and its recipients.7

Figure 3a shows thé distribution of stories from the network news. The salience of negative information on welfare clearly increased sharply during the middle of the decade. For example, in 1995 there were almost six times more negative stories than positive stories in the news. Another obvious feature is that, for nine of the eleven years, negative stories outnumber positive reports. The two exceptions occur in years when there was no virtually no coverage at all-1991 with zero stories and 2000 with one story in each direction.

Figure 3b presents the results for the New York Times' coverage of welfare. In this case, there were more positive than negative stories in 1990 and 1992. Coverage was almost perfectly balanced in 1991, 1993, and 1994 with roughly equal numbers of positive and negative reports in each of those years. But, negative stories became highly prevalent from 1995 through 1997. For example, negative reports outnumbered the positive by a fivefold margin in 1996. While negative newspaper coverage still dominated at the end of the time period, the disparity with positive coverage was not as great as it was during the middle of the decade.

Clearly, there was enormous variation in elite discourse on topics surrounding welfare across this time period. Both media sources exhibit the same mid-decade spike in news coverage along with the marked shift toward a negative tone. Indeed, the magnified salience of this issue is almost entirely due to television and newspaper stories that took a critical view of welfare. It cannot be coincidence that the timing of these changes in media content corresponds perfectly to the sharp decrease in public support for welfare spending.

Specific statements from, and the general orientations of, political actors during this period provide further evidence regarding the nature of elite discourse. There was an extremely high level of bipartisan criticism and a general consensus that the existing welfare system needed to be changed (Mead 2001). From the Democratic side, President Clinton raised the idea of welfare reform during the 1992 campaign when he promised to "end welfare as we know it" (Clinton and Gore 1992). Thereafter, the president repeatedly stated that Congress should pass "national welfare reform legislation" (Bryner 1998). He also used the federal waiver authority to allow welfare reform to proceed on a state by state basis (Weaver 2000).

Support for reform also existed among congressional Democrats, who were characterized as "talking tough on welfare" (New York Times 1994). Some, such as Senator John Breaux and Representative Harold Volkmer, were long-time proponents of significant change in the welfare system. Other Democrats, such as Senators Joseph Lieberman and John Kerry, initially expressed reservations about the Presidents plan. But, they later voted for the resultant legislation. And, even though a number of congressional Democrats (e.g., Senator Tom Daschle and Representative Richard Gephardt) were opposed to major changes in public assistance, they were not particularly vocal or active in support of their own positions. Hence, they seemed to acquiesce in the general spirit of welfare reform (Bryner 1998).

Many Republican congressional leaders also stressed the need to overhaul existing public assistance programs. Welfare reform was a prominent component of their highly-publicized "Contract With America" in 1994.8 When presenting this initiative, Republicans "pledged to 'revolutionize the system from top to bottom' and 'make the tough decisions that must be made'" (1995 House Congressional Hearings, as quoted in Sparks 2003).

In addition, several prominent Republican members of Congress made overt statements about the problems and shortcomings of public assistance. For example, Senator William V. Roth, Jr., of Delaware stated that the current welfare system had failed the "very families it was intended to serve." In a similar vein, Senator John Ashcroft from Missouri argued that "(t)he welfare system, designed as a way of helping people, has become a way of ensnaring them" (Pear 1996). House Speaker Newt Gingrich stated that it would be better to place poor children in "orphanages or group homes" than to raise them in existing welfare families (Winess 1994). Thus, there was vocal support for the general idea of welfare reform across the entire partisan spectrum. This focus completely permeated policy deliberations in the United States, especially during 1996.

To the extent that there were ideological disagreements on this issue, they centered on the precise kind of changes that were necessary. Liberal rhetoric emphasized the plight of needy children. They also believed that better educational opportunities and job training programs were needed for adults in order to get families off the welfare rolls. In contrast, conservatives were fairly united in their desire to end the entitlement basis of the welfare system; therefore, they favored more stringent work requirements for able-bodied program recipients (Epstein 1997; Heclo 2001).

The preceding ideological disagreements involved programmatic aspects of the American welfare system. From a policy perspective, these differences are extremely important. But, they are secondary concerns when welfare is conceptualized as a political issue to which the general public must react (Zaller 1992). For the latter, the more general theme of welfare reform dominated elite rhetoric, media content, and public discussion. In contrast, the specific types of changes advocated by liberals and conservatives comprised technical details that did not attract all that much explicit public attention and/or reaction (Heclo 2001).

The political environment of 1996 created a situation that was highly conducive to elite leadership of public opinion with respect to the welfare issue (Teles 1996). On the one hand, there was the sheer amount of attention devoted to welfare. It comprised a dominant component of the information stream that was transmitted from elites to the public, through the mass media. On the other hand, the content of the rhetoric stressed the negative aspects of public assistance and coalesced around the phrase "welfare reform." The intended message from this phrase was clear-cut: There was something wrong with the current welfare system, and it needed to be changed quickly and in dramatic ways (Weaver 2000).

There is simply no question that welfare reform was an extremely salient stimulus within the political environment that confronted the mass public in 1996. Both the quantity of coverage and the consistent symbolic content of the messages regarding welfare increased the probability of a public response. And, this response is manifested as the sharp decline in popular support for governmental welfare spending.


Fortuitously, the National Election Studies conducted a panel study that spanned the period from 1992 through 1996. This provides a unique opportunity to examine individual-level stability and change in attitudes toward welfare spending. Furthermore, variables routinely included in the NES interview schedule will enable us to test the different explanations for attitude change that were discussed in the previous section. Unfortunately, the panel data do not extend to 2000. This prevents us from examining the "rebound" of welfare spending attitudes that occurred in the latter year. But, we can still investigate the onset of the attitudinal changes that developed during the earlier time period. And, this is particularly important because major policy change was initiated during these same years.

The dependent variable for this analysis is the three-item scale composed of spending preferences on Food Stamps, welfare, and poor people.9 The focus is on temporal change in the individual scores on this variable. It is important to emphasize that a substantial amount of change did occur. Figure 4 shows the histogram for the differences between respondents' 1996 and 1992 scale scores. Negative values of this variable indicate that a person moved in the anti-spending direction, while positive values show pro-spending movement. Clearly, a sizable portion of the public-nearly 50 percent-became more opposed to welfare spending from 1992 to 1996. Approximately one third maintained stable spending preferences, while only about 18 percent favored increases in welfare expenditures. Thus, there is a substantial amount of temporal variation to be explained.

The remaining independent variables in Equation 1 represent individual orientations in 1996. Econ^sub i^ is i's response to the NES question asking whether the nations economy had gotten better or worse over the preceding year. This variable is scored with successive integers from 1 to 5, with larger values indicating that the economy had gotten worse. We use this variable to gauge the effect of economic conditions on welfare attitudes. Of course, these are subjective assessments rather than any objective indicator of economic performance. But, we believe that any impact of the latter on attitudes would have to be filtered through individual perceptions. Therefore, this is the appropriate type of variable to employ in the model.

Blacks^sub i^ is the respondents rating of the stimulus "Blacks" on a 0-100 feeling thermometer scale. Similarly, Recip^sub i^ is the thermometer rating for "people on welfare." These variables tap generalized feelings about the beneficiaries of government welfare spending. They are included in the equation because mass reactions toward welfare policies are clearly affected by citizens' characterizations of the recipient population. We are assuming that specific welfare-related beliefs (e.g., "Blacks do not subscribe to the traditional American work ethic," "welfare recipients are lazy and don't want to work," etc.) are manifested in their overall attitudes toward these groups.

Party^sub i^ and Ideol^sub i^ are i's party identification and liberal-conservative self-placement, respectively. These variables are measured using the familiar seven-point scales that have been included for many years in the NES interview schedules. They are coded so that larger values indicate stronger Republican identification and more extreme conservative self-placement, respectively. Partisanship and ideology are often characterized as symbolic predispositions that provide long-standing, stable guidance for citizens' issue attitudes (e.g., Sears 1993). In that capacity, however, their influence on 1996 welfare attitudes should be largely indirect and channeled through the lagged dependent variable. Their separate presence in this equation is intended to represent the degree to which people respond to contemporary political events. Self-identified Republicans and conservatives should mirror the stance articulated by Republican congressional leaders, who were openly hostile toward any expansion of the welfare system (Heclo 2001). Expectations about Democrats and liberals are less clear-cut as a result of mixed signals in elite rhetoric. While these groups are the traditional defenders of the American welfare state, that position was clouded by President Clintons active promotion of welfare reform (Weaver 2000). Thus, support for welfare among Democrats is perhaps less pronounced than is usually the case.

Inclusion of a lagged dependent variable on the right-hand side of Equation 1 introduces a complication in the estimation procedure. Specifically, the disturbance term for current welfare attitudes is likely to be correlated with prior attitudes. If so, then the OLS estimates would no longer be unbiased and consistent. In order to avoid this problem, we use two-stage least squares (2SLS). Several predetermined variables are used to obtain predicted values of 1992 welfare spending preferences. The latter are then substituted in the equation for Welf^sub 92,i^ in order to create an independent variable that should be uncorrelated with the equation's disturbance term.10


Table 2 shows the 2SLS estimates for Equation 1.11 The model fits the data quite well, with an R^sup 2^ of 0.423. More important, the specific coefficient estimates allow us to reject immediately some of the hypotheses about the influences on welfare spending.

First, there is not much continuity in welfare attitudes. The coefficient on the lagged dependent variable is 0.329. This value is statistically significant, indicating that past attitudes have some effect on current welfare preferences. However, the coefficient is also statistically different from 1.00. Therefore, current welfare attitudes are not just a simple extension of prior attitudes.

Second, individual assessments of the national economy have no discernible effect. The coefficient for the Econ^sub i^ variable is quite small at 0.069, and it fails to achieve statistical significance. Previous aggregate-level evidence was somewhat contradictory, showing both positive and negative relationships between economic performance and public generosity for the needy. But, any such connections are simply not reflected in these individual-level results. To the extent that 1996 attitudes differed from 1992 attitudes, this was not due to feelings about recent economic conditions in the United States.

Third, citizens' evaluations of Blacks are unrelated to their welfare spending attitudes in 1996. As with the previous variable, the coefficient for Blacks^sub i^ is very small (0.003) and has no significance. Several previous studies have shown convincingly that racial orientations affect the development of attitudes toward both welfare and government spending (Gilens 1999; Jacoby 2000). However, the result presented here shows that feelings about African Americans did not contribute to change in welfare spending preferences from 1992 to 1996.

The three remaining variables all have substantial effects. Consider evaluations of people on welfare: the coefficient for this variable is 0.012, a value that is statistically greater than zero. Precisely as hypothesized, feelings about welfare recipients affect individual reactions toward welfare policies themselves. Again, however, it is important to emphasize that this effect remains after taking prior welfare attitudes into account. Thus, feelings about the beneficiary population help account for temporal change in public opinion toward welfare.

Finally, party identification and ideology both have strong, statistically significant effects with coefficients of 0.116 and -0.288, respectively. The negative signs on these coefficients are perfectly reasonable given the coding schemes used in the variables (i.e., larger values indicate more Republican identifications and conservative self-placements, respectively). This finding is particularly telling precisely because it is probably not due to long-term symbolic influences. As explained earlier, these should be incorporated in the lagged dependent variable. The fact that partisanship and ideology continue to have a sharp impact even after prior attitudes are taken into account suggests that political orientations were a major source of the observed temporal change in citizens' preferences about welfare spending.

In order to summarize the relative magnitudes of the various effects, the right-hand column of Table 2 shows the standardized coefficients. The largest value occurs for the lagged dependent variable (0.343), showing that there is, in fact, a sizable amount of inertia in welfare attitudes. But, there is definitely not perfect continuity over time. Furthermore, the coefficient for ideology is also very large at -0.265. This indicates that a persons liberal-conservative stance could easily deflect spending preferences away from their earlier orientation. The same interpretation holds true for party identification and evaluations about welfare recipients, although their coefficients are somewhat smaller at -0.161 and 0.142, respectively.

Apart from the continuity of prior welfare attitudes, the three variables that show an effect in Table 2 are precisely those that are most likely to emanate from the external political environment. Again, the period from 1992 to 1996 was characterized by an unusual level of governmental and public debate over the American welfare system. Citizens' reactions to the political world, in general, are colored by their own personal orientations. Therefore, the impact of party identification and ideology is completely reasonable. It is also interesting to note that ideology had a stronger influence than partisanship, by a sizable margin. The relative imbalance in these two effects may well reflect the clarity of elite messages during this period. Once again, conservative opposition to the welfare state continued unabated, providing clear guidance for ideologically-based judgments. But, there was bipartisan support for the general objective of welfare reform and this may have decreased the potential for partisan divisions in the mass public's beliefs about welfare.

The third variable with a significant effect is individual evaluations of "people on welfare." Much of the rhetoric during the 1992-1996 period focused on welfare recipients, themselves. Critics emphasized the personal shortcomings and individual failures of program participants, sometimes describing them as "alligators" and "piranhas" (Sidel 1996). Supporters of the welfare system expressed deep concerns about children in poverty and, more generally, characterized welfare recipients as sympathetic figures-victims of circumstances beyond their own control (Mead 2001). If citizens were reacting to this element of elite discourse, then the significant impact of the thermometer ratings for people on welfare makes a great deal of sense.

We are arguing that public attitudes toward welfare spending in 1996 were shaped largely by external factors. If this is the case, then individual reactions should be especially pronounced among those people who were most attentive to the political world. The reasoning behind this expectation is simple: People who pay no attention to political phenomena would be highly unlikely to receive any messages emanating from elite discourse; thus, it would be impossible for them to react to such messages in any way. We would expect different responses from people who do follow political events.

In order to test this hypothesis, we estimate a new model. The dependent variable is still welfare spending attitudes in 1996. The basic independent variables are those that exhibited significant results in Table 2: welfare spending attitudes from 1992, party identification, ideology, and evaluations of welfare recipients. The equation also includes four interaction terms consisting of the previous independent variables multiplied by each respondent's general level of information about politics and public affairs. The latter is based upon the NES interviewer's assessment. It is a five-point scale, receded to range from zero through one, with higher values indicating a greater amount of information. Including the multiplicative terms in the equation will enable us to determine whether the effects of the other variables change across different levels of political information.

The new equation is estimated using 2SLS.12 The complete second-stage equation is shown in the Appendix.13 The R^sup 2^ value is 0.476, a larger value than the one obtained for the previous equation (i.e., from Table 2). While the fit statistics cannot be compared directly to each other, this result does suggest that the effects of the other variables vary with political interest.14

Table 3 shows the conditional effects of the other independent variables for people at the lowest and highest levels of political interest, respectively. The values reported in the table are calculated according to the procedures described by Friedrich (1982). First, let us examine the results among the least-interested set of NES respondents, shown in the left-hand column of the table. In this case, partisanship and feelings about welfare beneficiaries have statistically significant effects in the expected directions. Previous welfare attitudes and ideology have no impact at all.

The results are very different when we turn to the most interested stratum of respondents. The conditional effects for this subset are shown in the right-hand column of Table 3. Here, the only independent variables with significant impacts are previous welfare attitudes and ideology. The coefficients for the remaining two independent variables, partisanship and evaluations of welfare recipients, are tiny in magnitude and statistically indistinguishable from zero.

The information-related differences revealed in Table 3 are very important. Among less-informed citizens, 1996 welfare opinions are completely unrelated to prior stands on this issue. This result is perfectly consistent with the high levels of temporal instability that are known to exist in mass opinions on policy issues (Converse 1970; Hill and Kriesi 2001). For these people, responses to issue questions are apparently generated anew, at each time point. When this occurs, party identification functions in its usual capacity as a symbolic predisposition, providing general cues on how to react to policy-related stimuli. Similarly, opinions about relevant groups (welfare recipients in this case) provide additional easy guidance for the development of issue attitudes. This structure conforms to the more-or-less standard social psychological conception of issue attitudes, which does not really depend upon the receipt of incoming information from the external political environment.

In sharp contrast, the highly-informed respondents exhibit a great deal of continuity in their welfare spending attitudes. A persons stand is a good, albeit not perfect, predictor of their position in 1996. To the extent that they differ systematically from their 1992 positions, it is due entirely to the influence of liberal-conservative ideology. And, once again, this is consistent with the kinds of messages that would be received from elite political discourse. In Congress, Democrats as well as Republicans voiced strong support for the general goal of welfare reform. In so doing, they effectively removed partisanship as a useful cue for guiding citizens' opinions about welfare-but, only among people who were aware of these elite positions.

There was no such ambiguity with respect to ideology. Conservatives clearly wanted to end the entitlement nature of welfare programs and also reduce the amount and scope of government spending on public assistance. Liberals were perhaps not as vocal in their support for welfare; however, their position relative to conservatives remained fairly clear-cut. The influences of these elite positions are reflected in the strong conditional effect of ideology that exists among well-informed citizens. We do recognize that party identification and feelings about welfare recipients probably exert indirect effects on 1996 attitudes through their prior impact on 1992 attitudes. However, neither of these latter variables makes any contribution to attitudinal change, within this subset of the public, across this time period.


The purpose of this analysis has been to examine an interesting and politically consequential temporal pattern in public opinion toward welfare spending. The pattern itself is quite simple: support for spending decreased sharply from 1992 to 1996, but returned to its prior levels by 2000. The explanations for this pattern are more complex.

We argue that elite political discourse is the key factor in accounting for this change in public opinion. Welfare is an ongoing policy problem in the American political system, but it is seldom the immediate focus of issue conflict. It did occupy a central position on the governmental agenda in the mid-1990s. And, this is reflected in public opinion during 1996. According to R. Kent Weaver (2000), elites invested heavily in efforts to influence public perceptions of welfare. "A stream of critiques of AFDC from both Republican and Democratic politicians contributed both to a marked increase in the prominence of welfare reform issues and to increasing dissatisfaction with the current system" (171-72). Our empirical results confirm this assertion very nicely. This is an important result in itself. The level of discussion by public officials does affect the manifest content of citizen attitudes.

It is also important to emphasize that the structure of elite influence varies markedly depending upon individual levels of attentiveness to politics and public affairs. For uninformed citizens, opinions result from a combination of the "usual suspects"-party identification and evaluations of people on welfare. Consistent with standard scholarly explanations, factors like these provide general guidance for citizens to develop their issue attitudes. But, for present purposes, it is equally important to emphasize that partisan and group-based cues can generate change in citizens' opinion, as well.

Elite attempts to alter public opinion among the well-informed stratum of the population must overcome the "inertia" due to the general temporal stability of issue attitudes. But, when they can do so, as was apparently the case in the time period leading up to 1996, opinion changes are quite consistent with the nature of elite cues. When there is clearly identifiable elite conflict, such as differences between liberal and conservative leaders in the 1990s, this is manifested in patterns of opinion change. On the other hand, political alignments that are relatively muted with respect to an issue do not impinge on temporal change in the opinions of well-informed citizens. This was the case with respect to partisanship and welfare during the mid-1990s. Thus, highly-informed people responded directly to the content of the political environment.

Finally, and in a somewhat more speculative vein, direct elite influence on public opinion about a specific issue (welfare spending in this case) is apparently rather short-lived. This is perfectly consistent with Kingdon's (1995) "window of opportunity" concept which holds that policy change is only feasible during a brief period of time. And, it occurs when the orientations of political leaders and the mass public converge. The "punctuated equilibrium" framework suggested by Baumgartner and Jones (1993) is also relevant. According to that perspective, policy entrepreneurs manipulate the images that constitute the public's understanding of a particular issue. When successful at doing so, the result is rapid change in public opinion that occurs separately from long-term, secular trends. We believe that both of these ideas correspond precisely to what happened with welfare during the mid-1990s. Since that time, however, the issue has largely faded from the political agenda, and public opinion has returned to its former levels of support for welfare spending. This may have profound political consequences.

The legislation that enacted welfare reform, the Personal Responsibility and Work Opportunity Act of 1996, was set to expire on September 30, 2002. Although Congress has been considering various proposals to re-authorize the legislation, it has not been able to pass a final version of the bill. So, continuing resolutions have been used to extend funding for the original 1996 welfare reform provisions while the debate over the future direction of American welfare policy continues. Political elites have been unable to reach consensus on how welfare reform should proceed; consequently, they have not yet mobilized in ways that stimulate a clear directional response from American citizens. It remains to be seen whether this "void" in public opinion will have identifiable policy-relevant consequences.


Political Research Quarterly, Vol. 58, No. 3 (September 2005): pp. 367-379

1 Americans certainly voiced reactions to many other aspects of the welfare system apart from spending (Farkas, Johnson, Friedman, Bers 1996; Weaver, Shapiro, Jacobs 1995). However, Weaver (2000: 86) points out that there was widespread, serious ambivalence about specific components of welfare reform. Therefore, "single questions about welfare policy options are likely to provide a misleading reflection of patterns of public opinion." This is consistent with the more general position taken by Zaller (1992) that citizens generally possess opinions on broad policy questions, but not on specific details of governmental programs. Furthermore, Zallers work builds upon Converse (1964), who argues explicitly that public opinion often coalesces around a fairly simple spend-save dimension-precisely the aspect of welfare policy that we are examining here. From a different perspective, questions about federal spending really dominated the welfare discussion during the 1990s. Epstein (1997:12) states that "(t)he contemporary welfare debate is framed by . . . an intensifying competition over scarce public resources," Other analysts make a similar point: Issue conflict centered on the degree to which resources would be devoted to welfare rather than other governmental programs (e.g., Blank 1997; Bryner 1998; Weaver 2000). Thus, our focus on citizen attitudes toward welfare spending is justified by both the nature of public opinion and the substantive content of the welfare issue, itself.

2 It is important to emphasize that "common structure" does not mean monolithic or identical responses to all aspects of welfare and public assistance to needy groups. Jacoby (1994,2000) and others (Gilens 1999; Eismeier 1982; Sanders 1988) have shown that public suppon for government spending varies directly with the perceived legitimacy of program beneficiaries. Americans are willing to spend for sympathetic groups, such as the poor and the homeless. On the other hand, there is more reluctance to provide resources for Food Stamp and welfare recipients.

3 Weaver, Shapiro, and Jacobs (1995) analyze trends in public opinion toward welfare. They employ data extending back in time as far as the early 1970s, obtained from a variety of polling organizations. Their results show that the level of public support for welfare has been relatively stable over the years except for the decline that occurred in the mid-1990s. Weaver, Shapiro, and Jacobs use different data, but they uncover the same temporal pattern that we do. This increases our confidence that the NES data are providing an accurate representation of public opinion toward welfare. Weaver el al.'s research also shows that the opinion movement in the 1990s retains its distinctiveness when viewed across a longer time period. We believe this enhances the significance of the current study precisely because the 1996 decline in welfare support represents a significant departure from what appears to be a longer-term equilibrium in the distribution of American public opinion.

4 Difference-of-means tests comparing the 1992 and 1996 data produce t values of 8.67 for Food Stamp spending preferences, 6.87 for welfare spending, 4.95 for spending on poor people, and 8.39 for the three-item spending preferences scale. The probability values associated with these statistics are all less than 0.0001. For the differences from 1996 to 2000, the t values are -8.49 for Food Stamp spending preferences, -9.75 for welfare spending, -3.90 for spending on poor people, and -9.12 for the three-item spending preferences scale. Once again, the probability values for these test statistics are all less than 0.0001.

5 The information on television news content was obtained from the Underbill University Television News Archive, which is available online at: The information on New York Times content was obtained from the New York Times Index. In each case, we counted the yearly number of stories that appeared under the "welfare" topic heading provided by each of these archival sources in order to produce the values in Figure 2.

6 Clearly, the overall amount of coverage in the New York Times is greater than in the television news. For example, both of these sources were least attentive to welfare during 1990 and 1991. But, during those two years, a total of 127 prim stories appeared compared to only one story on the network news broadcasts. This difference in levels of coverage is to be expected given the nature of the two media. And, more important, it does not compromise our main arguments about temporal variability in media coverage of welfare issues.

7 Our classification scheme for the news stories is based upon a distinction made by Weaver (2000). He argues that a "crucial change" occurred in the "balance of messages" presented by elites about welfare reform in the mid-1990s. The debate shifted away from the positive elements of welfare to negative aspects on the system and to how to reform it. We used this basic distinction in our content analysis of elite discussions about welfare during the period. Media stories which emphasized that welfare benefits should be reduced, eligibility limited, or program spending cut, were coded as "negative" messages. News reports that focused on the economic achievements of welfare beneficiaries, the health care gains of welfare families, or the need to increase program spending were coded as "posiiive" messages. Stories which presented an impartial, more objective impression of the welfare system were coded as "neutral." Stories with a negative or positive slant always outnumbered those that look a neutral tone, usually by a sizable margin. In the television news broadcasts, neutral stories never made up more than one-third of the yearly total. In the New York Times, there was never more than 19 percent neutral stories in any year. Full information about the distributions of stories and the content analysis is available upon request. We found additional support for this basic categorization of elite messages in a number of other sources, including Thompson and Norris (1995), Epstein (1997), Blank (1997), Bryner (1998), Heclo (2001), and Sparks (2003).

8 The "Republican Contract With America" contained ten major provisions. The third provision, The Personal Responsibility Act," dealt specifically with reforming the welfare system: The objective was to "overhaul the American welfare system, to reduce government dependency, attack illegitimacy, require welfare recipients to enter work programs, and cap total welfare spending" ("Republican Contract With America" 1994).

9 The use of this multiple-item scale has at least two important advantages. First, it provides a more complete measure of individual feelings about the American welfare system than would any of the individual items that comprise it. This is important precisely because welfare is an extremely complex, multi-faceted social problem that subsumes a variety of more specific themes and concerns (Epstein 1997; Bane and Ellwood 1994). Second, the scale is a more reliable indicator than any of the individual items, so it reduces the detrimental effects of measurement error on the empirical results. The reliability coefficient for the scale (Cronbachs Alpha) is 0.698 in 1992, 0.697 in 1996, and 0.710 in 2000.

10 The predetermined variables used to create the instrument for 1992 spending preferences are the 1992 values of the remaining independent variables from Equation 1 (i.e., retrospective assessments of the national economy, thermometer ratings of Blacks and people on welfare, pany identification, and ideological self-placement), along with thermometer ratings of the federal government, family income, gender, race, and an additive scale measuring the respondents' reliance on public assistance in 1992. The R^sup 2^ for the first-stage regression is 0.365.

11 The small number of observations in Table 2 (n = 331) reveals that there is an enormous amount of missing data on the variables included in this analysis. The total number of respondents in the NES 1992-1996 Panel Study is 2439. Thus, the current analysis only uses about 14 percent of the total sample. The major problem seems to be a sizable number of people who (for various reasons) failed to provide valid responses to almost all of the items used in this study. Restricting attention to only those observations with more nonmissing than missing values in both 1992 (i.e., six or more valid values out of eleven variables used in the First-stage regression) and 1996 (i.e., four or more valid values out of seven variables used in the second-stage regression) still leaves only 595 observations (about 24 percent of the total). With this subset from the sample, we used Honaker et al. Amelia software (1999), and King et al. (2001) to obtain multiple imputation estimates of the missing data. When the latter are used to estimate Equation 1, the precise values of the coefficients and standard errors change somewhat from those reported in Table 2. However, the relative sizes of the effects, the overall fit of the equation, and the substantive conclusions that would be drawn from the results remain virtually identical to those reported in the text. Full details of the multiple imputation results are available from the authors.

12 As with the estimates for Equation 1, there is a severe missing data problem; the analysis is based on only 328 observations. Once again, the multiple imputation estimates obtained for the subset of NES respondents who had more nonmissing than missing data in both years would not change the substantive conclusions drawn from the analysis. Full details of the multiple imputation results are available from the authors.

13 The predetermined variables used to create the instrument for 1992 spending preferences are identical to those described earlier. The R^sup 2^ for the first-stage equation is 0.361. Also, note that the second-stage equation does not include a term for the political interest variable. The latter is omitted for two reasons. First (and most important), there is no substantive reason to believe that political interest has a direct impact on spending preferences. Second, when the variable is included in the second-stage equation, its coefficient is very small and not significantly different from zero.

14 The goodness of fit values cannot be compared directly because Equation 1 is not completely nested within the equation reported in the Appendix and summarized in Table 3. Again, the latter omits the variables from Equation 1 with no significant effects. When the not significant independent variables from Equation 1 are placed back into the equation with the multiplicative term (thereby creating a pair of nested equations), the improvement in model fit is statistically significant. The F statistic for the combined effects of the multiplicative terms is 16.385, with 4 and 312 degrees of freedom. The probability level for this test statistic is less than 0.0001.



Bane, Mary Jo, and David T. Ellwood, eds. 1994. Welfare Realities: From Rhetoric to Reform. Cambridge, MA: Harvard University Press.

Baumgartner, Frank R., and Bryan D. Jones. 1993. Agendas and Instability in American Politics. Chicago: University of Chicago Press.

Blank, Rebecca. 1997. It Takes a Nation: A New Agenda for Fighting Poverty. New York: Sage.

Bobo, Lawrence, and Ryan A. Smith. 1994. "Antipoverty Policy, Affirmative Action, and Racial Attitudes." In Sheldon H. Danziger, Gary D. Sandefur, and Daniel H. Weinberg, eds., Confronting Poverty: Prescriptions for Change. New York: Sage.

Bryner, Gary. 1998. Politics and Public Morality: The Great American Welfare Reform Debate. New York: Norton.

Clinton, Bill, and Al Gore. 1992. Putting People First. New York: Times Books.

Converse, Philip E. 1964. "The Nature of Belief Systems in Mass Publics." In David Apter, ed., Ideology and Discontent. New York: Free Press.

_____. 1970. "Attitudes and Nonattitudes: Continuation of a Dialogue." In Edward R. Tufte, ed., The Quantitative Analysis of Social Problems. Reading, MA: Addison-Wesley.

Cook, Fay L, and Edith J. Barrett. 1992. Support for the American Welfare State: The Views of Congress and the Public. New York: Columbia University Press.

Durr, Robert H. 1993. "What Moves Policy Sentiment?" American Political Science Review 87: 158-70.

Eismeier, Theodore J. 1982. "Public Preferences About Government Spending." Political Behavior 4: 133-45.

Epstein, William M. 1997. Welfare In America: How Social Science Fails the Poor. Madison: University of Wisconsin Press.

Farkas, Steve, Jean Johnson, Will Friedman, and Ali Bers. 1996. Americans and Welfare Reform. New York: Public Agenda Foundation Press.

Feldman, Stanley, and John Zaller. 1992. "The Political Culture of Ambivalence: Ideological Responses to the Welfare State." American Journal of Political Science 36: 268-307.

Friedrich, Robert J. 1982. "In Defense of Multiplicative Terms in Multiple Regression Equations." American Journal of Political Science 26: 797-833.

Gilens, Martin. 1999. Why Americans Hate Welfare. Chicago: University of Chicago Press.

Heclo, Hugh, 2001. "The Politics of Welfare Reform," In Rebecca Blank and Ron Haskins, eds., The New World of Welfare, pp. 169-200. Washington, DC: Brookings Institution Press.

Hetherington, Marc J. 1996. "The Medial Role in Forming Voters' National Economic Evaluations in 1992." American Journal of Political Science 40: 372-95.

Hill, Jennifer L, and Hanspeter Kriesl. 2001. "An Extension and Test of Converse's 'Black and White' Model of Response Stability." American Political Science Review 95: 397-414.

Honaker, James, Anne Joseph, Gary King, Kenneth Scheve, and Naunihal Singh. 1999. Amelia: A Program for Missing Data (Windows Version). Cambridge, MA: Harvard University,

Jacoby, William G. 1994. "Public Attitudes Toward Government Spending." American Journal of Political Science 38: 336-61.

_____. 2000. "Issue Framing and Public Opinion on Government Spending." American Journal of Political Science 44: 750-67.

King, Gary, James Honaker, Anne Joseph, and Kenneth Scheve. 2001. "Analyzing Incomplete Political Science Data: An Alternative Algorithm for Multiple Imputation." American Political Science Review 95: 49-69.

Kingdon, John. W 1995. Agendas, Alternatives, and Public Policies, 2nd ed. New York: HarperCollins.

Lieberman, Robert C. 1998. Shifting Ike Color Line: Race and The American Welfare State. Cambridge, MA: Harvard University Press.

Mead, Lawrence M. 2001. "The Politics of Conservative Welfare Reform," In Rebecca Blank and Ron Haskins, eds., The New World of Welfare, pp. 201-20. Washington, DC: Brookings Institution Press.

Mink, Gwendolyn. 1998. Welfare's End. Ithaca, NY: Cornell University Press.

New York Times. 1994. "Talking Tough on Welfare." October 20.

Pear, Robert. 1996. "Senate Votes to Ease Work Requirement in Welfare Overhaul Bill." New York Times, July 23.

"Republican Contract With America." 1994. U.S. House of Representatives, 104th Congress.

Sanders, Arthur. 1988. "Rationality, Self-interest, and Public Attitudes on Public Spending." Social Science Quarterly 69: 311-24.

Sears, David O. 1993. "Symbolic Politics: A Socio-Psychological Theory." In Shanto Iyengar and William J. McGuire, eds., Explorations in Political Psychology. Durham, NC: Duke University Press.

Sidel, Ruth. 1996. Keeping Women and Children Last. New York: Penguin Books.

Sparks, Holloway. 2003. "Queens, Teens, and Model Mothers: Race, Gender, and the Discourse of Welfare Reform." In Sanford E Schram, Joe Soss, and Richard C. Fording, eds., Race and the Politics of Welfare Reform. Ann Arbor: University of Michigan Press.

Stevenson, Randolph T. 2001. "The Economy and Policy Mood: A Fundamental Dynamic of Democratic Politics?" American Journal of Political Science 45: 620-33.

Teles, Steven M. 1996. Whose Welfare? AFDC and Elite Politics. Lawrence: University Press of Kansas.

Thompson, Lyke, and Donald E. Noms. 1995. "Introduction: The Politics of Welfare Reform." In Donald F. Norris and Lyke Thompson, eds., The Politics of Welfare Reform. Thousand Oaks, CA: Sage.

Weaver, R. Kent. 2000. Ending Welfare As We Know It. Washington, DC: Brookings Institution Press.

Weaver, R. Kent, Robert Y. Shapiro, and Lawrence R. Jacobs. 1995. "Trends: Welfare." The Public Opinion Quarterly 59: 606-27.

Winess, Michael. 1994. "Team in Place, Gingrich Comes Out Slugging." New York Times. December 7, B11.

Zaller, John R. 1992. The Nature and Origins of Mass Opinion. New York: Cambridge University Press.

[Author Affiliation]



[Author Affiliation]

Received: January 6, 2004

Accepted for Publication: September 7, 2004

Author Advanced search


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.