INTRODUCTION
The Great Recession of 2008 has left millions of individuals unemployed, struggling with shrinking incomes and facing heightened levels of economic insecurity. The consequent demand for larger social spending has brought to the fore the public debate over the proper role of government in providing support for the unemployed and the needy. This debate, of course, is not new but rather one on which many individuals hold long-standing ideological views. Yet these views may have been challenged by the hardships experienced during the financial crisis. How do individuals’ preferences on welfare policy shift in response to changes in their personal economic circumstances? Are the welfare policy preferences of voters predominantly a function of their ideological dispositions, or are they mostly shaped by their material interests at a given time period?
To put this point in starker terms, consider a hypothetical case of two otherwise similar individuals, one positioned ideologically on the left and the other on the right, who lose their jobs at the same time. The question that arises is whether the same downturn in their personal circumstances leads to a convergence in policy preferences, whereby the right-leaning individual becomes significantly more supportive of welfare assistance, or whether their different ideological dispositions yield two distinct responses, in line with their previously held views.
The importance of understanding the factors that shape individuals’ attitudes on welfare assistance has been underscored by growing evidence that the preferences of voters on the issue are also an important cause, and not just an outcome, of governments’ choice of welfare policy (e.g., Brooks and Manza Reference Brooks and Manza2007; Haman and Kelly 2010; Lupo and Puntusson Reference Lupu and Pontusson2011; Pierson Reference Pierson1994). As part of this research, much scholarly attention has been given to the claim that economic self-interest is a key determinant of voters’ attitudes on welfare and redistribution; people's position in the labor market, exposure to the risk of layoff, and their financial standing are the main factors associated with this line of explanation (Alesina and La Ferrara Reference Alesina and Ferrara2005; Bean and Papadakis Reference Bean and Papadakis1998; Burgoon Reference Burgoon2001; Cusack et al. Reference Cusack, Iversen and Rehm2006; Iversen and Soskice Reference Iversen and Soskice2001; Rehm Reference Rehm2009, Reference Rehm2011). Although these studies use varied sources of data to support this claim, the strength of the empirical evidence they provide has come under question for a number of reasons. First, the evidence is based almost exclusively on analysis of cross-sectional survey data in which scholars find correlations between measures of survey respondents’ economic standing and their views on social policy. But with this type of evidence a causal link between the two measures remains unclear: It could be, as the authors argue, that individuals’ labor market circumstances shape their attitudes on welfare policy; Yet it is also plausible that an unobservable characteristic—such as people's upbringing, or the influence of their parents—explains their preferences on welfare provision and their standing in the labor market. Further doubt about the material self-interest explanation arises from the fact that several other studies that rely on similar cross-sectional data reveal a very weak empirical association between measures of individuals’ economic circumstances and their attitudes on the welfare programs from which they are deemed to benefit (Lynch and Myrskyla Reference Lynch and Myrskyla2009; Mughan Reference Mughan2007; Taylor-Gooby Reference Taylor-Gooby and Taylor-Gooby2001).
A second strand of explanation shifts the focus away from economic self-interest and instead emphasizes the impact of political ideology as a determinant of people's attitudes on welfare assistance. Differences in ideological dispositions between voters on the left and the right on issues such as equality, fairness, and the appropriate role of government are argued to be central factors underlying individuals’ welfare preferences (Alesina and Glaeser Reference Alesina, Giuliano, Bisin and Benhabib2004; Bean and Papadakis Reference Bean and Papadakis1998; Feldman and Zaller Reference Feldman and Zaller1992; Fong Reference Fong2001; Funk Reference Funk2000; Linos and West Reference Linos and West2003). Yet as some studies recognize, the two strands of explanations—those focused on economic self-interest and those on ideology—may not be mutually exclusive. For one, ideology may be an outcome of material circumstances rather than a separate source of influence. Alternatively, ideological dispositions may be a systematic moderator of the way in which the material interests of individuals shape their political views (e.g., Lodge and Taber Reference Lodge, Taber, Lupia, McCubbins and Popkin2000; Redlawsk Reference Redlawsk2001). Thus, due to these potentially interactive mechanisms, it is empirically difficult to separate the roles of self-interest and ideology and causally identify the impact of personal economic circumstances on people's preferences on welfare policy.
To address these empirical challenges, this article exploits a within-subject research design that tracks individuals during an extended period of economic volatility. More specifically, I use an original panel study that consists of four waves of surveys in which the same national sample of respondents was contacted for repeat interviews between July 2007 and March 2011. In these repeat interviews, detailed information was collected not only on respondents’ changing labor market circumstances but also on their political attitudes. Utilizing this rich longitudinal data, covering periods both before and after the eruption of the financial crisis, I estimate how individuals’ preferences on welfare policy shift in response to the personal experience of three types of economic shocks: a substantial drop in household income, a subjective decrease in perceived employment security, and the actual loss of a job.
The central finding of the analysis is that voters’ preferences regarding welfare policy are strongly affected by changes in their own economic circumstances. In particular, the loss of employment is found to have a major effect, increasing the average probability of support for greater welfare spending by between 22 and 25 percentage points. These results are shown to be robust to a broad range of empirical specifications. In addition, a set of placebo tests lends further credence to the identification strategy by demonstrating that the views of individuals personally harmed by the economic shocks changed substantially with respect to welfare policy but not with respect to largely unrelated policy domains such as global warming or cultural values. This suggests that a change in personal material considerations, rather than a general disorientation in attitudes, accounts for the link between the experience of an economic shock and the shift in people's welfare preferences.
The analysis also reveals that the experience of the economic shock does indeed lead to a convergence in the welfare preferences of harmed individuals who prior to the shock held distinct political views. In particular, I find that in response to a personal economic shock such as layoff, Republicans and Independents grew significantly more supportive of welfare assistance, while among Democrats the effect was much smaller. This differential impact is only partially due to a “ceiling effect,” namely to the fact that most Democrats were supportive of welfare expansion even before the crisis.
Finally, I find that with the passing of time, as job losers regain employment, their support for the expansion of welfare spending decreases significantly. This shift in attitude among the re-employed is more frequent among voters on the right. These patterns suggest that while economic shocks can have a sizable effect on welfare preferences of individuals, the effect is probably not a reflection of a profound conversion in their political world view. Rather, the attitudinal change appears to reflect a more provisional response to an immediate and sometimes temporary need, and as such can be fairly short lived.
The article's findings contribute to the growing literature on the political economy of social policy (e.g., Huber and Stephens Reference Huber and Stephens2001; Mares Reference Mares2003; Rueda Reference Rueda2007). With the accumulation of evidence that voters’ preferences are an important factor shaping governments’ choice of welfare policy (e.g., Brooks and Manza Reference Brooks and Manza2007; Haman and Kelly Reference Hamann and Kelly2010), there is need for a clearer understanding of the influences that shape the public's welfare attitudes as well as for better estimates of the magnitude of these influences. By testing and quantifying the impact of various personal economic circumstances on voters’ attitudes, this study provides a stronger empirical foundation for explaining variation in mass support for more expansive welfare policies both across publics and over time.
The article also adds to the ongoing research on the relative roles of ideology and material self-interest in the formation of individuals’ political preferences (e.g., Bartels Reference Bartels2008; Fiorina Reference Fiorina1981; Gelman et al. Reference Gelman, Shor, Bafumi and Park2007; Malhotra and Margalit Reference Malhotra and Margalit2010). As these studies attest, there is still much disagreement on the relative influence of the two forces. In the context of welfare policy, the analysis presented here suggests that the two could be usefully thought of as continuously present influences, but as having varying impacts over time: whereas political ideology is an important factor accounting for voters’ baseline policy stance, self-interested considerations arising from changing material circumstances do often outweigh individuals’ prior ideological dispositions and bring about a sizable, even if temporary, shift in preferences.
The rest of the article proceeds as follows. The next section reviews the debate over the sources of voters’ social policy preferences and draws predictions from the main theoretical approaches. The subsequent section describes the data and the empirical strategy used to test these predictions. The results section, which follows, is composed of two parts: the first focuses on estimating the impact of personal economic shocks on individuals’ welfare preferences; the second compares the response to the shocks among individuals with different prior ideological leanings. The final section discusses the broader implications of the findings.
THEORETICAL BACKGROUND
A great deal of research in recent years has focused on explaining people's attitudes on the inter-related issues of government welfare assistance and redistribution.Footnote 1 Scholars have offered evidence tying individual attitudes on welfare policy to a range of factors, but the bulk of research on social policy preferences centers on two broad forms of explanation: those focused on people's own economic interest and those emphasizing ideological factors.Footnote 2 The arguments centered on self-interest typically assume that individuals are risk averse and thus, when facing less certain or lower future revenue streams (e.g., through the possible loss of one's job), they grow more supportive of higher levels of social assistance provision (Bean and Papadakis Reference Bean and Papadakis1998; Cusack et al. Reference Cusack, Iversen and Rehm2006; Iversen and Soskice Reference Iversen and Soskice2001; Rehm Reference Rehm2009, Reference Rehm2011). To support this logic, these studies put forward evidence that is predominantly based on single snapshots of cross-sectional data, finding correlations between various measures of individuals’ economic circumstances (e.g., employment in an insecure job) and their attitudes on social policy.Footnote 3 Yet as noted earlier, the causal interpretation of this evidence is unclear; while the former may be the cause of the latter, as scholars often assume, it is quite plausible that other unobservable factors account both for individuals’ economic circumstances (e.g., the type of job they take) and their views on welfare assistance. In other words, this empirical association may be spurious.
Indeed, a set of studies finds no evidence for the alleged causal relationship between individuals’ economic interests and their attitudes on welfare policy. For example, a study of survey data from 11 European countries finds that the expected beneficiaries of various pension programs are no more likely than nonbeneficiaries to oppose retrenchment in those programs (Lynch and Myrskyla Reference Lynch and Myrskyla2009). Analysis of survey data from the U.S. and Australia also reveals no significant association between various measures of individuals’ economic insecurity and their attitudes on relevant social policies (Mughan Reference Mughan2007). These “contrarian” studies also rely on the same type of cross-sectional data and thus confront the same inferential difficulties. Yet they are supported by research on other policy domains, which indicates that individual attitudes on government policies tend to reflect sociotropic concerns rather than individual calculations of material self-interest (e.g., Citrin and Green Reference Citrin, Green and Shapiro1990; Kinder and Sears Reference Kinder and Sears1981; Sears and Funk Reference Sears and Funk1990). In sum, the evidence to date on the causal impact of individuals’ own economic circumstances shaping their preferences on social policy is highly inconclusive.
A second approach links preferences on welfare and redistribution to political ideology.Footnote 4 This line of explanation is again not unique to the study of welfare policy: ideological dispositions have long been described as key factors shaping individuals’ policy views, independent of considerations centered on material self-interest (e.g., Campbell et al. Reference Campbell, Converse, Miller and Stokes1960; Stonecash Reference Stonecash2000). Indeed, research on long-term trends in public opinion reveals marked and persistent differences in welfare policy preferences across ideological lines throughout the 20th century, whereby voters on the left tend to be more supportive of expanding welfare programs than voters on the right (Shaw and Shapiro 2005; Shapiro Reference Shapiro2009). These ideological differences may be an outcome of the observed divergence in the beliefs of voters on the left and right regarding the degree to which people's economic fortunes are attributable to external factors (e.g., luck, the system) as opposed to individual characteristics (e.g., hard work, ambition) (Alesina and La Ferrara Reference Anderson and Pontusson2005; Evans Reference Evans1997; Fong Reference Fong2001). A stronger belief that economic hardships are due to one's own doings may lead to lesser sympathy to the plight of the downtrodden, and consequently reduce support for redistributive policies (Alesina and Giuliano Reference Malhotra and Margalit2010; Alesina and Glaeser Reference Alesina, Giuliano, Bisin and Benhabib2004).
While these accounts center on one form of influence—material interests or ideological dispositions—as the main determinant of views on welfare policy, it could be of course that the two influences are interlinked. Instead of treating material interests or ideological dispositions as separate influences, a third approach is to treat the two influences as connected through a process of learning. A change in economic fortunes could lead, for example, to a new calculus of where one's own self-interest lies. It could also lead to an updating of the beliefs underlying the political ideology that one subscribes to. For example, an individual who is laid off may learn new information about the spread of risk in the labor market and consequently update her views on the need for social insurance programs. Indeed, much of the work on voters’ assimilation of new information and on attitude change centers on this type of a learning process (Gerber and Green Reference Gerber and Green1999; Page and Shapiro Reference Page and Shapiro1992). The key aspect in this “learning” account is that the assimilation of the new information leads to an enduring change in views or preferences, one that extends beyond the individuals’ short term circumstances.
To illustrate the differences between the theoretical approaches described above, we can compare their predictions regarding the impact that changes in economic standing are likely to have on people's social policy preferences. Returning to the hypothetical scenario raised earlier of the two individuals losing their jobs, let us now assume further that after a spell of unemployment each manages to find a new job. The question is whether, and in what way, their welfare preferences will change from their “baseline” (i.e., pre-shock) level. Figure 1 outlines three different trajectories of attitude change that an individual may experience. These trajectories are not exhaustive, but are arguably the most plausible in this context. The vertical axis of each graph denotes the probability of support for expanded welfare assistance and the horizontal axis denotes time.
FIGURE 1. Change in Economic Fortunes and Trajectories of Preferences on Welfare Policy
In the left-most panel (Figure 1(a)), the attitude of the individual remains stable, irrespective of the changes in economic fortunes that the individual experienced. In contrast, in the second panel support for expanded welfare increases from the baseline following the loss of the job, but then decreases soon after the individual regains employment. Finally, in the panel on the right, support for welfare grows following the loss of the job, but then remains stable at the higher level, even after the individual's own circumstances improve. The predictions derived from each of the theoretical accounts in the literature can be mapped onto one of the patterns described above, though potentially to more than one. I begin by laying out the more straightforward predictions of each theoretical account and later return to assess possible alternatives.
If welfare preferences are driven primarily by myopic considerations of self-interest, one would expect to observe an inverted-U (“transient change”) pattern: following the loss of a job and an increased reliance on welfare services, individuals’ support for greater welfare assistance would rise. But then, as the individual finds new employment and is less dependent on such assistance, that support would drop. In contrast, if political ideology is the overwhelming influence shaping preferences, a change in one's own material standing is not expected to bring about a meaningful shift in views; whether laid off or re-employed, support for welfare assistance is predicted to stay very close to the baseline rate (i.e., to follow the “no change” pattern). Finally, a “learning” scenario suggests that as an individual learns firsthand about the employment risks in the labor market and the attendant hardships, she would, on average, update her prior beliefs and exhibit greater support for unemployment assistance. Then, after finding new employment, that support should remain high as a result of the learning that occurred in the previous period (i.e., the “persistent change” pattern).Footnote 5
As noted earlier, past research has shown that left and right voters differ in their level of support for welfare spending (e.g., Bean and Papadakis Reference Bean and Papadakis1998; Jaeger Reference Jaeger2008). Yet importantly, the pattern of predicted change in each of the three approaches described above is not dependent on individuals’ prior ideology, even if starting at different baseline rates. Returning to the question raised earlier of whether views on welfare policy will converge among the different partisans who experience a common economic shock, the “ideology” account that predicts a “no change” pattern suggests that no such convergence will occur, as each partisan will maintain their prior views. In contrast, both the myopic self-interest and the learning accounts imply greater, or even complete, convergence in the welfare preferences of left and right wing voters who experience the shock.Footnote 6
All three approaches offer ex ante plausible accounts of human behavior. Which approach best explains actual behavior is ultimately an empirical question. Using novel longitudinal data, the subsequent sections examine how welfare preferences respond to various changes in individuals’ economic fortunes during a four-year period. This analysis should allow us to assess the validity of various alternative explanations for the observed patterns, and help deduce which of the theoretical accounts offer the most explanatory insight.
DATA AND MEASUREMENT
The data presented here are based on four surveys administered by YouGov/Polimetrix, of Palo Alto, CA.Footnote 7 The first survey was carried out in July 2007 and included a national sample of 3,000 respondents.Footnote 8 The sample was constructed using the “closest neighbor” matching methodology described in Rivers (Reference Rivers2007).Footnote 9 The resulting sample was then ranked by stratum to match the age, gender, race, and education marginals in the 2006 American Community Survey. A key feature of the survey is that data were collected on the employment status and income of both members of married couples, as well as respondents’ subjective perceptions of their job security and future job prospects. Crucially, the survey also included items pertaining to respondents’ political views and preferences on economic and social policy matters.
In April 2009, May 2010, and March 2011 three additional waves of the survey were fielded, each time inviting the same set of respondents to participate again. Given that the sample of returning individuals in each wave was not fully representative of the general population on a number of dimensions, as is often the case with re-interviews, the same matching methodology was again used to augment the panel in constructing a broader sample of respondents that matched the marginals of the U.S. population from the Current Population Survey of the corresponding period. In total, the experiences and attitudes of 6,229 respondents are analyzed in this study, 3,049 of whom were interviewed multiple times (1,603 individuals were interviewed twice, 1,044 interviewed in three of the waves, and 402 respondents were interviewed in all four waves).
Table A1 compares the characteristics of the samples in each wave with the sample in the American Community Survey along key demographics.Footnote 10 The table shows that the characteristics of participants in the panel emulate those of the national population quite well on a broad range of categories. Nonetheless, the panel under-represents the 18–34 age group and high-school drop outs, and is slightly poorer than the general population average, an imbalance that is more noticeable among female respondents. These differences should serve as a qualification when drawing conclusions about the average impact of the crisis on the broad U.S. public.
A concern associated with panel data of this type is that attrition from the panel is nonrandom and could thus lead to biased samples: people who agree to be interviewed multiple times may differ in meaningful ways from those who refuse to do so. To examine the potential severity of this issue, Table A2 compares the characteristics of the respondents as a function of the number of interviews in which they participated. While other characteristics not captured in this comparison may still differentiate the groups classified by the number of successful recontacts, the table suggests that the groups do not significantly differ in terms of key demographic characteristics, including age distribution, income, educational attainment, and employment status. This does not mean that the returning panel participants can be considered as fully representative of the population at large; however, it does indicate that the panel participants are, at a minimum, not a biased sample with respect to their distribution along the main demographics.Footnote 11
The key dependent variable in the subsequent analyses (henceforth the “main question”) is respondents’ answer to the following item: “Do you support an increase in the funding of government programs for helping the poor and the unemployed with education, training, employment, and social services, even if this might raise your taxes?” Responses were located on a five-point scale: 1. strongly support; 2. somewhat support; 3. neither support nor oppose; 4. somewhat oppose; 5. strongly oppose. Capturing respondents’ preferences on welfare policy is problematic and ideally requires the use of a broad battery of survey items. While this single question has drawbacks, it arguably captures an important element of the politically salient tradeoff between expanding the social safety net and the potential costs of such an expansion.Footnote 12
Since a key question in this study is the effect of individuals’ prior political ideology, I account for respondents’ partisan affiliation in their first interview.Footnote 13 I code respondents’ initial partisanship based on their answer to the following question: “Generally speaking, do you think of yourself as a . . .?” Possible answers to the question include the following: 1. Democrat; 2. Republican; 3. Independent; 4. Other; and 5. Not Sure. To avoid losing observations, I code all respondents that selected “other” or “don't know” (11.8% of respondents in the 2007 survey) together with the Independents.Footnote 14
Respondents were asked about their total pretax income in the preceding year (and that of their spouse) with answers recorded on a 10-point scale. Each response category represents an income band (e.g., $40,000–$50,000). To calculate the percentage change in household income over the panel period, I transform these bands into their currency midpoints.Footnote 15 Since unemployed respondents were not asked to report their income, I assign them the average intake in 2009 of $15,000. Household income is then calculated using the sum of the two main earners reported by the respondent.Footnote 16
The article examines the impact of three economic “shocks” on respondents’ preferences on welfare assistance. The first shock is job loss, which prior research indicates can affect political preferences (e.g., Margalit Reference Margalit2011). I categorize the variable as “1” for any individual who across two sequential surveys was employed during the first but unemployed during the second. This category thus refers only to newly unemployed individuals; it excludes those who were unemployed in both periods (henceforth long-term unemployed) as well as individuals who were “not employed” in the previous period (as opposed to unemployed), i.e., students, retirees, and homemakers. The analysis also controls for job loss of spouse, an indicator variable that denotes instances in which a respondent's spouse was employed during the previous wave of interviews but was unemployed in the subsequent survey.Footnote 17
The second shock variable is job less secure, a binary measure that denotes a substantial worsening in respondents’ subjective sense of job security. As Anderson and Pontusson (Reference Anderson and Pontusson2007) note, concerns about job security vary on two dimensions: the degree to which respondents perceive the possibility of a job loss as likely (“cognitive job insecurity”) and the degree to which they actively worry about this possibility (“affective job insecurity”). I focus on the former and compare respondents’ answers over subsequent surveys to the question “Looking forward to the next three years, how confident do you feel about being able to keep your current job?” Answers ranged from (1) “very confident” to (4) “not confident.” The variable job less secure takes the value “1” if the subjective sense of job security dropped by two levels or more, and “0” otherwise. I choose this more conservative coding scheme to increase the likelihood of identifying individuals that experienced a substantial decline in job security rather than a general trend of growing pessimism. Among individuals who transitioned from having full, or part-time employment in the previous wave of the survey to being unemployed in the next wave (and were thus not asked the job security question in the latter wave), the job less secure variable was coded as “1” only if they reported in the previous wave that they were at least “somewhat confident” of being able to keep their current job over the next three years.
The third and final shock I examine is income drop, a binary variable that takes the value “1” if the household income decreased by at least 25% in the period between the two surveys and coded as “0” otherwise. The reason for choosing the 25% threshold, rather than simply any income decrease, is twofold: first, there is likely to be some degree of measurement error associated with the fact that income is recorded in ranges rather than by the exact dollar figure. If one were to use any income decrease as the threshold (e.g., of 1% or greater), the measurement error problem is likely to be more severe. Second, I am interested in examining whether individuals that confront a significant decline in well-being change their policy preferences. While still arbitrary, a drop in income of 25% or more is likely to represent such a significant shift in household circumstances.Footnote 18
Table 1 presents the key statistics pertaining to changes in respondents’ economic standing during the four waves of the study with respect to the three shocks. As the table shows, the rate of unemployment among working-age respondents increased between the first two waves of the survey from 5.7% to 8.3%, a shift that was slightly smaller than the actual rate in the U.S. population (4.6–8.9%).Footnote 19 This increased rate remained almost unchanged in the 2010 study and decreased somewhat by early 2011. Furthermore, among the individuals that were employed during the first wave of the survey, 16.8% reported two years later that they were either unemployed, not employed, or employed in a lower capacity than in the preceding period (e.g., shifted from full-time to part-time work). In addition, many of the individuals who were unemployed at the time of the first wave (in 2007) continued to suffer from the adverse labor market conditions. In fact, only one in four had a job (even part time) at the time of the survey in 2009. Indeed, the phenomenon of long-term unemployment is one of the widely discussed features of the Great Recession.Footnote 20
TABLE 1. Economic Standing of Sample Respondents, by Wave
Note: Cell entries correspond to the distribution of respondents across categories in each survey wave.
The middle panel of Table 1 compares respondents’ sense of job security over the four waves. Particularly notable is the fact that between the first two waves—i.e., the one before and the one shortly after the eruption of the crisis—there was a sharp decline of about 16 percentage points in the share of individuals that felt “very confident” in being able to keep their job over the next three years (from 51.7% to 35.3%). At the same time, the share of respondents insecure about their labor market prospects almost doubled, rising to just over 33% in July 2009. Notably, in the surveys carried out in 2010 and in early 2011, levels of job insecurity remained similarly low.
Finally, the table shows that a nontrivial portion of the sample experienced a squeeze in their household income: whether as a result of a family member losing a job, due to salary cutbacks in the workplace or because of lower profits earned by self-employed business owners, respondents’ household income registered a drop of 4% in the immediate aftermath of the crisis. However, this figure masks substantial divergence: whereas the household income of over 80% of the sample remained intact or even grew somewhat, about 9% of the sample experienced a drop of 25% or greater.Footnote 21
The sheer magnitude of the Great Recession means that the affected segment in the sample population is quite sizable, allowing for a meaningful assessment of the impact that personal economic shocks have on people's political attitudes. Table A3 presents the numerical counts of respondents in each category of labor market experiences.
RESULTS
Attitudes on Welfare Policy: Before and After
I begin by examining the pre-crisis “baseline” rate of support for greater welfare spending as reported in 2007. Figure 2 presents the distribution of policy preferences among the full sample and among the two main groups of partisans. The distribution of responses in the full sample shows an overall, albeit slight, majority of respondents in support of an increase in welfare provision (56%). Yet the graph also highlights the extent to which welfare spending is a partisan issue in the U.S.: whereas 81% of Democrats were supportive of greater welfare spending, the corresponding figure among Republicans was 22%. In fact, 60% of Republicans opposed such an increase, a figure that is about 10 times greater than the rate of opposition among Democrats. Predictably, the preferences of Independents were located within the range of the two partisan camps, with just over half (54%) supportive of welfare spending expansion (not shown in the graph).Footnote 22
FIGURE 2. Attitudes on Welfare Spending in 2007, by Partisan Identification
Note: The vertical axis denotes the share of respondents that answered the question: “Do you support an increase in the funding of government programs for helping the poor and the unemployed with education, training, employment, and social services, even if this would raise your taxes?” The national survey was fielded by Polimterix/YouGov in July 2007.
How have these attitudes changed following the eruption of the Great Recession? Table 2 analyzes the shift in attitudes on welfare spending between the period shortly before the crisis (July 2007) and the first survey taken after the eruption of the crisis (April 2009). The results indicate that among those who before the crisis opposed the expansion of welfare spending, one in four grew more favorable of such a policy: by 2009, about 13% reported being neutral (“neither support nor oppose”) while another 12% became supportive of increased welfare spending. However, the reverse process was just as pronounced: among those who initially supported an expansion of welfare spending, 14% described their view in the latter survey as neutral, and another 11% became opposed. Finally, among individuals who were neutral on the issue prior to the crisis, about 24% have grown more opposed while 38% have become more supportive of the expansion of welfare spending.
TABLE 2. Change in Welfare Spending Preferences: Before and After the Eruption of the Crisis
Note: Percentages are calculated by row. Numbers in parentheses are cell counts.
When accounting for the initial distribution of respondents across the different cells in the table, one finds that public opinion after the eruption of the crisis was overall less supportive of increasing welfare spending than in the precrisis period (from 55% to 47%).Footnote 23 As Figure 3 shows, this decline in support occurred among partisans of all ideological persuasions, though the patterns of the shifts differed: Among Democrats, support for welfare expansion dropped almost nine points shortly after the crisis but steadied onwards, while among Independents the decline was consistent and also the steepest. The drop in support among Republicans was smaller in absolute terms (4.8 percentage points) but fairly large in relative terms once considering the low baseline.Footnote 24 The general drop in support for expanded welfare assistance at a time of a painful recession might seem surprising. One explanation might be that voters became increasingly concerned about the specter of growing budget deficits and the consequent possibility of future tax hikes. Such concerns may have brought about a general shift against government spending of any type, including on welfare programs.Footnote 25 Yet does this downward shift mean that the economic hardships experienced by many had no effect on increasing their support for welfare assistance? In the next section I explore this question in some detail.
FIGURE 3. Support for Expansion of Welfare Spending, by Original Party Identification
Note: The vertical axis denotes the share of respondents that either “somewhat” or “strongly” support an increase in welfare spending.
Personal Economic Circumstances and Support for Welfare Policy
To estimate whether economic circumstances affect individuals’ preferences on social spending, I examine the attitudes of respondents as a function of their personal exposure to an economic hardship. Figure 4 presents an unconditional comparison of those who did and those who did not experience the shocks. As the graph shows, individuals whose job has become significantly less secure are somewhat more supportive of welfare assistance than individuals whose job had remained about as secure (51% versus 47%). A similar gap also differentiates those whose household income dropped significantly and those whose incomes did not. In contrast, the graph indicates that the loss of a job was associated with a sizable increase in support for welfare assistance: individuals who lost their job were significantly more likely to support the expansion of welfare spending than individuals who remained employed (59% versus 47%, p=0.006). Strikingly, the graph also shows that the rate of support for welfare expansion is about as low among individuals who remained employed as it is among the newly re-employed. This graph thus suggests that the boost in support for welfare assistance among job losers may be quite short-lived, dissipating soon after individuals find new employment.
FIGURE 4. Change in Economic Conditions and Support for Increased Welfare Spending
As noted, this comparison does not take into account any individual-level characteristics that may distinguish between the harmed individuals and the other respondents. The next analysis thus introduces controls for a range of potentially relevant individual-level characteristics. For ease of interpretation, I present results of an ordinary least squares (OLS) estimation in which the dependent variable Welfare i,t is a transformed five-point scale that takes the maximum value 1 if respondent i strongly supports expanding welfare provision in time t and 0 if she strongly opposes it. One can therefore interpret a regression coefficient as a 100*β percentage point change in the dependent variable associated with the predictor.Footnote 26 The model controls for key demographic variables—age, gender, income, education, race, and marital status—and takes the form
\begin{eqnarray}
\emph{Welfare}_{i,t} &=& \alpha + \beta_1\emph{Welfare}_{i,t-1} +\beta_{2}\emph{Shock}_{i}\nonumber\\
&& +\, \gamma\emph{Demographics}_{i,t} + \phi\emph{SurveyWave} + \epsilon_{i}.\nonumber\\
\end{eqnarray}
Several features of this specification should be noted. Given that our primary interest is understanding how the experience of a significant economic setback alters people's welfare policy preferences, the estimated model examines change in people's support for expanding welfare spending rather than their absolute level of support. The model specification does so by controlling for Welfare i,t−1, a measure that denotes respondent i’s level of support for welfare expansion in the previous period (t − 1).Footnote 27 The baseline category in each specification is an individual who was employed both at the time of the previous survey and during the time of the next survey. In each model, the effect of at least one of the three shocks is estimated. The model also includes Demographics, a vector of individual-level characteristics (income, education, gender, and marital status) as well as a set of indicator variables denoting respondents’ employment status: long-term unemployed is a binary variable that takes the value 1 if the individual was unemployed in both the previous period and the next period of the study, and 0 otherwise; similarly, newly re-employed denotes individuals who were unemployed in the previous period and have found employment by the time of the next survey. In all estimations reported in the article, standard errors are clustered by respondent.
Table 3 presents the results of this estimation. The specification presented in the first column estimates the impact of a loss of a job on the probability of support for increased welfare assistance. It indicates that an individual who recently lost a job is expected to experience, on average, a shift equivalent to about 9.5 points (on a 100 point scale) from their prior stance in the direction of greater support for welfare spending. This change differs quite significantly from the effect associated with being long-term unemployed, for which we find only a small, negative and statistically insignificant change in support for welfare assistance. Also notable is the pattern seen in Figure 4, whereby the newly re-employed were significantly less likely than recent job losers to support welfare expansion (p<0.01), and on average reported views similar to individuals who remained employed throughout. This result holds also when one excludes from the category of newly re-employed those individuals who lost their job before the financial crisis broke out.Footnote 28 I return to discuss the newly re-employed category in greater detail below.
TABLE 3. Personal Economic Shocks and Support for Welfare Assistance
Note: Standard errors clustered by respondent. All regressions include controls for respondents’ marital status, race, and income (coefficients not reported). † significant at 10%; * significant at 5%; ** significant at 1%.
The results also indicate that Democrats were significantly more likely, and Republicans significantly less likely, than Independents to increase their support for welfare expansion. In addition, individuals with higher levels of educational attainment were also more likely to become more pro-welfare.
Column (2) shows that a major drop in household income does not correspond with a large increase in support for welfare spending. The effect associated with the shock is small and the standard errors are a good deal larger (β income drop = 0.005, σ = 0.018). Furthermore, column (3) indicates that a significant decrease in a respondent's sense of job security is associated with increased support for welfare spending, though the effect is smaller than the effect associated with the loss of a job (substantively, about a four-point change). However, when all three shocks are included in the model (column (4)), the effect of job loss is the only one that remains precisely estimated (p<0.01).Footnote 29
Is the effect of the shock driven solely by concerns about one's own experiences, or do changes in the labor market standing of the other household members matter as well? Column (5) provides a partial answer to this question, showing that the layoff of a spouse is also associated with a fairly sizable effect, equivalent to almost a five-and-half-point increase in probability of support for welfare expansion. Notably though, the effect is smaller than that associated with the loss of one's own job, and also less significant statistically (p = 0.101).
Yet while these results indicate that changes in people's economic circumstances can have a sizable impact on their support for welfare assistance, the effect we observe may also be a reflection of the fact that those individuals who experienced the shocks reside in areas more severely hurt by the crisis. If that is the case, the effect we observe may at least partially represent a sociotropic concern about the impact of the crisis on the broader public rather than a response to the personal experience of the setback. To examine this possibility, I re-estimate the same model but add controls for the unemployment rate in the respondents’ county at the time each survey was taken, and also include fixed effects for state of residence and year of the survey (state × year). The additional controls should help account for some of the temporal variation in local circumstances. Column (6) presents the results and shows that the estimates are not sensitive to the inclusion of these additional controls for local conditions, as the estimated effects of the economic shocks remain almost unchanged. In fact, controlling for other factors, the coefficient of county unemployment rate is negative, indicating that higher unemployment locally is associated with a slight decrease in support for welfare expansion. The evidence thus suggests that the observed increase in support for welfare spending occurs primarily in response to the change in respondents’ own circumstances rather than in response to weak economic conditions in their area of residence.Footnote 30
Does an Economic Shock Cause a Switching of Views on Welfare?
The specification in Equation (1) estimated the likelihood of an increase (or decrease) in support for welfare spending along a five-point scale. The specification did not distinguish between a shift within a policy stance (e.g., of someone changing their preference from being “strongly” to “somewhat” opposed to the expansion of welfare spending), and an actual switch across stances (e.g., from being opposed to becoming supportive). To estimate whether the economic shocks had the effect of bringing about an actual switch in policy position, I estimate a probit model that takes the form
\begin{eqnarray}
&&{\rm Pr}(\emph{ProWelfare}_{i,t} =1) = \beta_1\emph{Welfare}_{i,t-1} +\beta_{2}\emph{Shock}_{i}\nonumber\\
&&\quad +\, \gamma\emph{Demographics}_{i} + \phi\emph{SurveyWave} + \epsilon_{i}
\end{eqnarray}
The dependent variable ProWelf i,t is a binary measure that takes the value 1 if respondent i supports expanding welfare provision at time t, and 0 otherwise. This specification is thus more demanding—and arguably, more politically meaningful—in that only an actual shift in preferences from nonsupport to support (and vice versa) is estimated.
Table 4 presents the estimated marginal effects.Footnote 31 The results reveal empirical patterns similar to those obtained from the previous estimation. Most notably, the loss of a job is again associated with a sizable and statistically significant effect: an increase of about 24 percentage points in the likelihood of becoming a supporter of greater welfare spending.Footnote 32 The findings regarding the other two shocks are also similar to those observed earlier: a drop in job security is again associated with a notable increase in support for welfare spending (10.4%, p <0.05), while the effect associated with a drop in household income is small and statistically indistinguishable from zero. When controlling for all three shocks in the same estimation (columns (4)–(6)), only the effect associated with a job loss remains significant. As before, the loss of the spouse's job is associated with a sizable effect, however, its magnitude is appreciably smaller (40–50%) than the effect associated with the loss of one's own job and is imprecisely estimated.
TABLE 4. Economic Shocks and Transition in Support of Welfare Assistance
Note: The coefficients of the probit analysis are estimated marginal effects (∂F/∂xk); that is, the marginal effect on Pr(y = 1) given a unit increase in the value of the relevant (continuous) regressor (xk), holding all other regressors at their respective sample means. Standard errors are clustered by respondent. All regressions include controls for respondents’ marital status and race (coefficients not reported). † significant at 10%; * significant at 5%; ** significant at 1%.
Notable again is the fact that the newly re-employed are not more likely (and perhaps even slightly less likely) to support welfare expansion than individuals who remained employed.Footnote 33 The results also indicate that the newly re-employed are, on average, less supportive of welfare spending than individuals who just lost their job (difference significant at p<0.01).Footnote 34 This result is very much consistent with the “transient change” pattern in Figure 1: a large increase in support for welfare expansion following the experience of the shock followed by a significant drop in support after finding a new job.Footnote 35 The most obvious explanation for this inverted-U pattern is the myopic self-interest account, yet, as noted earlier, it is not the only possible explanation. The pattern of attitude change may also be a result of a rapid learning process: following a layoff, individuals update their views about the societal advantages of having an expansive welfare system. However, from their experience of searching for a new job, they learn again, this time about the weaknesses of the welfare system (e.g., encourages idleness among others) and consequently become less supportive of welfare spending. If indeed such a learning process underlies the change in attitudes, individuals who had experienced a layoff in the past should exhibit a different pattern of attitude change from those for whom this is the first such experience, since the former had presumably learned from their prior experience of a layoff. Yet that is not what we observe.
The analysis in Table 5 examines whether the loss of a job has a different impact on the welfare preferences of individuals who experienced a layoff in the past than on those who have not had such an experience. Column (1) presents results from the same specification used in the first column of Table 3.Footnote 36 Adding a control in column (2) for a past experience of a layoff indicates that such an experience is not associated with a different attitude on welfare assistance, as the effect of previously laid off is statistically indistinguishable from zero. In column (3) the effect of a recent job loss is interacted with the past experience of a layoff, and again the results show no empirical relationship between previous layoffs and change in voters’ current welfare attitudes. Finally, in column (4) I examine whether the timing of the previous layoff event, when interacted with experiencing a new job loss, is associated with a different effect on respondents’ welfare attitudes. The fact that none of the relevant coefficients even approaches statistical significance indicates that a past history of a layoff has no discernible effect on changes in respondents’ attitudes. This seems, prima facie, as fairly strong evidence against the alternative explanation that attributes the transient change in attitudes to a “learning” process in the sense described earlier.Footnote 37
TABLE 5. Past and Recent History of Job Loss and Support for Welfare Assistance
Note: All models include the full set of controls from the benchmark specification in Table 3 (coefficients not reported). “Laid off in past” is a binary variable coded as 1 if the respondent had previously experienced a layoff and 0 otherwise. † significant at 10%; * significant at 5%; ** significant at 1%.
Robustness: Placebo Specifications and Reverse Causality
Is the adverse change in individuals’ economic standing the cause of the shift in their welfare preferences? The use of panel data tracking individuals over time, rather than reliance on a single snapshot of cross-sectional data, means that concerns about unobservable variables accounting for the changes in preferences are significantly diminished. To further increase confidence in the identification strategy, this section subjects the main result to two additional tests. In the first, I look at whether job loss is associated with a shift in voters’ preferences on other policy issues; the expectation is that in domains unrelated to economic policy, the trend of change in the preferences of the harmed individuals would not deviate significantly from those who were unharmed. To examine whether this was the case, I exploit the fact that respondents were repeatedly asked in subsequent surveys a set of questions on other policy domains. I then estimate the same specification as in Equation (2) but with a different dependent variable: instead of analyzing the change in respondents’ attitudes on welfare policy, I estimate respondents’ attitudes on far-removed topics—global warming, cultural values and border security (see the Online Appendix for exact wording of questions). As before, all estimations also include as a regressor the respondent's answer to the same question as recorded in the previous interview.
Table 6 reports the coefficients pertaining to the parameters of interest. Entries in the top row denote the marginal effect of a job loss on respondents’ policy stance. In the first column the dependent variable is the same as in column (1) of Table 4, namely support for expansion of welfare spending. As we have seen before, job loss is associated with almost a 24-percentage-point increase in the likelihood of having a pro-welfare stance, an effect that is highly significant in statistical terms. In contrast, columns (2)–(4) show that job loss is not associated with any meaningful change in views on the importance of dealing with global warming, with protecting American values from foreign cultural influences, or with the perceived importance of border protection from security threats. In sum, this analysis indicates that job loss is associated with a major change in respondents’ preferences only in a policy domain directly related to the economic setback experienced by the individual but not with corresponding changes in any of the other policy areas.Footnote 38
TABLE 6. Placebo Specifications and Tests for Reverse Causality
Note: The coefficients of the probit analysis are estimated marginal effects (∂F/∂xk). All regressions include all controls for respondents’ marital status, race, education level, and income, and fixed effects for survey wave (coefficients not reported). Standard errors clustered by respondent. Columns (1)–(4) also control for respondents’ employment status. † significant at 10%; * significant at 5%; ** significant at 1%.
A second informative robustness test examines the possibility of a reverse causal relationship, namely that people's prior welfare preferences reflect the expectation of confronting in the future a serious economic setback. To test for this possibility, the dependent variable in columns (5)–(7) is the experience of the shock (coded as a 1 if experienced the shock and 0 otherwise). The results indicate that respondents’ views on welfare assistance in the preceding period were not associated with the experience of any of the shocks in the subsequent period, alleviating the concern that welfare preferences are entirely endogenous to the anticipation of an economic shock.
PRIOR IDEOLOGY AND RESPONSES TO AN ECONOMIC SHOCK
Having demonstrated that a change in people's economic circumstances brings about a subsequent change in their attitudes on welfare assistance, the remainder of the analysis explores the degree to which voters’ response to the experience of a personal economic shock varies as a function of their prior ideological dispositions. To do so, I estimate a similar model as in Equation (2), yet this time the experience of the shock is interacted with the original partisan affiliation respondents reported in their first wave of interviews.
The estimated marginal effects presented in Table 7 reveal several findings: first, we see that the impact of job loss on welfare preferences is appreciably smaller among Democrats than among Republicans and Independents (the omitted category). Whereas job loss is associated with a 40.2-percentage-point increase in the probability of support of welfare expansion among Republicans (p <0.001), the effect is only 3.7 percentage points among Democrats, and statistically indistinguishable from zero. A second notable result is that a drop in household income is associated with a heterogeneous effect on different partisans. Whereas the previous analysis found no significant effect of an income drop on the preferences of the population as whole, column (2) shows that among Republicans it is associated with a large increase in support for welfare spending of 19 percentage points (p<0.05), while among Democrats the effect is smaller and negative. Finally, we see in column (3) that growing job insecurity is associated with a larger positive impact on Republicans and Independents than on Democrats (18.7%, 8.2%, and 3.0%, respectively). This gap between Democrats and Republicans is also highly significant in statistical terms.Footnote 39 The final two specifications control for all shocks together, interacted with partisan identification. Here we see again that all three shocks have a larger impact on Republicans than on Democrats, although the effects for income drop and job insecurity are substantively smaller.Footnote 40
TABLE 7. The Impact of Economic Shocks on the Welfare Policy Preferences of Different Partisans
Note: Dependent variable is a binary measure of support for expanding welfare assistance to the needy and the unemployed. The coefficients of the probit analysis are estimated marginal effects (∂F/∂xk), holding all other regressors at their respective sample means. Baseline (omitted) category is Independents who did not experience the shock. Standard errors clustered by respondent. All models include the full set of control variables used in the benchmark specification (coefficients not reported). † significant at 10%; * significant at 5%; ** significant at 1%.
It is not ex ante obvious why the impact of economic shocks on the welfare preferences of right-of-center voters is more significant. One possible explanation might be that Republicans and Democrats differ in terms of other personal characteristics, which may then account for the different responses to economic setbacks. To assess the merits of this explanation, I re-estimate Equation (2) separately for voters of each party. This is akin to interacting party identification with each individual characteristic in the model. The results of this estimation are presented graphically in Figure 5, which shows the probability of a shift in support for greater welfare assistance as a function of respondents’ initial partisan affiliation.Footnote 41 The horizontal axis in each of the charts denotes the level of support for welfare expansion that respondents reported in the previous survey; the vertical axis presents the probability of support for welfare expansion in the subsequent survey. The continuous lines refer to individuals who experienced an economic shock in the period of study and the dashed lines refer to individuals who did not.
FIGURE 5. Probability of support for expansion of welfare spending, by exposure to economic shocks and initial partisan affiliation
Note: The graphs report the probability of support for welfare expansion (on the Y-axis) as a function of the individual's level of support for the policy in the previous period (measured on the x axis along a five-point scale). Each graph corresponds to a different type of economic shock. Results are reported separately for Democrats and Republicans.
The main pattern that the graph illustrates is that the welfare preferences of Republicans harmed by the shocks diverged more sharply from the preferences of their unaffected Republican counterparts than was the case among Democrats. Among the latter, the policy preferences of the individuals who experienced the shock remained very similar to those who did not.Footnote 42 In other words, controlling for differences in the individual characteristics of Democrats and Republicans does not account for the divergence in how partisans’ preferences shift in response to a shock.
A second explanation for the variation in partisan responses might be that the Republicans and Democrats that were harmed—rather than Republicans and Democrats in general—differ in terms of key characteristics which account for the different attitudinal responses to the experience of the shock. Figure A1 compares the characteristics of the different partisans that lost their job during the period when the four surveys were conducted. The graphs show quite clearly that while the job losers were similar across partisan groups in terms of age and confidence in their labor market prospects, Democrats who lost their job tended to be slightly poorer and less educated than their Republican counterparts. Do these differences account for the different shifts in welfare attitudes among the job losers in the two partisan camps?
To explore this possibility, I use propensity score matching to estimate the responses of the different partisans who lost their job. The matching exercise seeks to identify the closest replicate among the control group units of each “treated” unit. For example, I compare the welfare policy views of a job loser to the preferences of those individuals who were most similar on a host of other relevant dimensions, but who remained employed throughout the period.Footnote 43 By comparing the average difference in the welfare preferences of the treated and the matched group, one obtains an unbiased estimate of the average effect of the treatment.Footnote 44
The left panel of Table 8 reports the results of this comparison. Columns (1)–(3) present the net change associated with each economic shock, by respondents’ partisan affiliation in the first wave. While the magnitudes of the estimates shift somewhat, the same pattern emerges as in the previous analyses: Republicans and Independents were much more likely to increase their support for welfare assistance in response to experiencing an economic shock than their partisan counterparts who did not. In contrast, the differences among Democrats are small in magnitude and statistically insignificant. These results indicate that the differences in the partisan responses to the experience of a major economic setback are not simply a reflection of the individual characteristics of the harmed Republicans, Independents, and Democrats.
TABLE 8. Economic Shocks and Support for Welfare Expansion Using Propensity Score Matching
Note: Entries are based on comparisons using propensity score matching. Matching was done on respondents’ previous preferences on welfare, education, gender, age, marital status, race, and previous level of job security. In the left-hand panel (columns 1–3) entries denote the difference in the probability of support for welfare expansion between partisans that experienced the economic shocks and their counterparts who did not. In the panel on the right (columns 4–6) the dependent variable is a binary indicator denoting whether the individual became more in favor of expanded welfare spending or not. Entries denote the difference in the probability of increased support among partisans that experienced the shock and their counterparts who did not. Figures in the right-hand panel are based on matched comparisons only among individuals who did not “strongly support” welfare expansion in the previous period. Calculations estimated separately by partisan affiliation. † significant at 10%; * significant at 5%; ** significant at 1%.
The final explanation I assess empirically for the difference in partisan responses is the presence of a “ceiling effect”: since most Democrats were in favor of welfare expansion even prior to the crisis while Republicans were not (81% versus 22%), the latter had more “room” to move in the direction of greater support for welfare assistance. To explore this possibility, I carry out a second matching exercise in which the dependent variable is an increase in support for welfare expansion (1) versus no increase (0). However, in this analysis I limit the sample to those individuals who did not “strongly support” welfare expansion in the previous period, i.e., the sample now includes only those respondents who could potentially increase their stated support for welfare expansion.Footnote 45 The results of this analysis, reported in columns (4)–(6), provide some support for this explanation, indicating that the probability of increased support for welfare expansion was indeed less divergent across partisan lines once accounting for the ceiling effect. For example, the effect of a job loss on Republicans is now only about twice the effect on Democrats (compared to a sixfold difference in the analysis on the left panel). Furthermore, job insecurity is now associated with an even larger impact on Democrats. These patterns suggest that the ceiling effect accounts for some of the partisan difference in responses to the shocks. Nonetheless, a nontrivial share of the variation remains unexplained. In the concluding section, I discuss other potential avenues for explaining this remaining difference in the partisan responses to the shocks.
DISCUSSION
This study provides compelling evidence of the strong impact of personal economic circumstances, particularly the loss of employment, on individuals’ preferences on welfare spending. Beyond documenting and quantifying this significant causal effect, the analysis reveals an inverted U-like shape in the pattern of support for welfare expansion: rising sharply following the loss of a job, but dropping back down as the employment situation improves. Finally, the results show an asymmetry in the partisan response to the economic shocks: a worsening personal economic standing has had a large impact on the attitudes of voters on the right and center, but a much weaker effect on voters on the left.
Earlier in the article I described a set of theoretical accounts pertaining to the link between political ideology, economic interests and attitudes on social policy, and sketched out their divergent predictions regarding the impact that changes in economic fortunes would have on voters’ welfare preferences. The pattern of a transient change in attitudes that mirrors the changes in individuals’ labor market standing is very much consistent with the predictions of a self-interested account of myopic welfare preferences. Furthermore, the analysis also offered evidence that this pattern of attitude change is most likely not a result of an ongoing learning process, perhaps the most plausible alternative explanation.
Nonetheless, it would be misguided to interpret the findings of this study as indicating that voters’ welfare policy preferences are simply a function of self-interested considerations; such an interpretation ignores a set of the findings that, taken together, suggest otherwise. First and as noted, the data reveal a significant partisan difference in the extent to which attitudes change in response to common shocks. Second, despite the dramatic changes in people's economic fortunes during the years of the crisis, a large majority still maintained their prior views on welfare policy. And finally, even among Republican job losers, a group whose support for welfare expansion rose to a rate 2.5 times greater than their co-partisans who remained employed (32% versus 13%), approximately two-thirds still did not support increased spending on welfare assistance. In fact, support for welfare expansion among Republicans who lost their jobs remained about 45 percentage points lower than among Democrats who stayed employed throughout the period. The accumulation of these findings clearly indicates that there is more to preferences on welfare policy than mere self-interest; prior ideological commitments remain an important factor in any account of voters’ policy stance on this issue, even in conditions of great economic turmoil.
Going forward, the findings point to a somewhat different theoretical approach from those currently prominent in the literature on welfare policy preferences. Rather than focusing on a single source of influence (economic interest or ideological disposition), or conceiving of one as simply conditioning the other, the two forces can most usefully be thought of as continuous influences, but as having varying impacts over time, depending on the changing circumstances. For example, the evidence suggests that the relative influence of ideological considerations decreases in times when people face major economic concerns, in which case selfish material considerations gain prominence and lead to attitude change among some. Crucially, the temporal variation in the relative influence of each type of consideration is not random, but rather appears to follow a theoretically predictable pattern. A promising avenue for subsequent research would therefore be to accept the ongoing presence of multiple, at times conflicting influences on people's welfare preferences, and focus on identifying the conditions and timing under which one form of influence gains relative pre-eminence over the others.Footnote 46
As noted, the results of the analysis reveal a difference in the way the economic set-backs affected the attitudes of voters across the political spectrum. This partisan gap is shown to be only partially accounted for by a “ceiling effect,” and controlling for demographic differences does not eliminate the observed partisan gap either. More research is needed to fully explain the factors underlying this pattern of partisan difference. Yet to advance this undertaking, I wish to briefly outline two other explanations that cannot be tested convincingly with the data I have and which strike me as worthy of further examination.
The first holds that partisans who are willing to explicitly depart from a widely shared party stance on a central issue x are likely to (i) hold stronger-than-average views about that issue,Footnote 47 and (ii) support the party due to its position on some other important dimension or due to a longstanding emotional tie (e.g., the party their parents traditionally support). In other words, their partisan affiliation is maintained despite the party's position on issue x. For example, some voters might have a clear stance against increased welfare spending, but nonetheless support the Democratic Party because of its position on other issues such as gay rights or environmental protection. Thus, the small minority of Democrats that were, to begin with, opposed to welfare expansion perhaps represents a hard “core” whose views on this issue are less malleable. If that is the case, it could help explain why those individuals exhibited only a small change in attitudes on welfare policy following a worsening in their personal circumstances. A second explanation builds on the literature on government learning, which argues that policy change typically follows strong external stimuli in the form of disaster or major failures of existing policy (Birkland Reference Birkland1997; Heclo Reference Heclo1974). Smaller failures, in contrast, are argued to instigate less reconsideration and therefore produce little or no change. Since right wing voters, as noted earlier, often attribute the economic fortunes of individuals to their own doings rather than to external forces (e.g., Alesina and La Ferrara Reference Alesina and Ferrara2005; Evans Reference Evans1997), it stands to reason that an economic shock such as being laid off represents a sharper contrast to (or “failure” of) prior beliefs for a voter on the right. Such contrast, if it brings about a more serious reassessment of one's prior political position, might help account for why a greater change in views was registered among voters on the right.
Shortly after the eruption of the crisis, and with looming elections in the U.S., an editorial of the Wall Street Journal warned of “a period of unchecked left-wing ascendancy” and lamented that “the current financial panic may give today's left another pretext to return to those heydays of welfare-state liberalism.”Footnote 48 That forecast, which some dreaded and others hoped for, was soon belied when two years later the Republicans won a resounding victory in the midterm elections, in part due to the Tea Party's vocal campaign against “big government” and “out-of-control” social spending. The results of this study help explain why, despite the meltdown of the financial system and the hardships it inflicted on millions, the Great Recession did not bring about that transformative shift of policies to the left. For one, the uneven impact of the crisis meant a sharp bifurcation in sentiment between the narrower constituency who personally experienced a major economic setback and the broader population that did not. Whereas support for greater government spending on welfare assistance increased among the former, it actually decreased among the rest of the population. Second, prior ideological commitments were not simply thrown by the wayside; rather, they remained a strong dividing line separating support and opposition to an expanded welfare system. Finally, even among those directly hurt by the crisis, the pro-welfare shift seems to have been fairly short-lived, dissipating as personal circumstances improved. In other words, a broad leftward shift in the electorate's preferences simply did not take place. Thus, if change in U.S. social policy will depend on a strong popular demand, it seems fair to conclude that a return to the “heydays of welfare-state liberalism” is probably not much closer today than it was in the more prosperous years before the crisis erupted.
TABLE A1. Comparison of Samples across Key Demographics: The American Community Survey and the Four Panel Waves
TABLE A2. Respondent Characteristics, by Number of Successful Contacts
Comments
No Comments have been published for this article.