Reaching Out or Pulling Back: Macroeconomic Conditions and Public Support for Social Welfare Spending

Social Security Administration. 2007. Statistical supplement to the Social Security bulletin, 2006. http://​www.​ssa.​gov/​policy/​docs/​statcomps/​supplement/​2006/​supplement06.​pdf

US Bureau of the Census. http://​www.​census.​gov/​compendia/​statab/​tables/​07s0462.​xls

Such a focus is not unique. While some scholars who study attitudes towards the social welfare state take a more general view, analyzing attitudes towards government spending on services generally or in the government’s role in guaranteeing jobs and a good standard of living (e.g., Berinsky 2002; Feldman and Zaller 1992; Jacoby 2000), many others focus primarily on programs for the poor (e.g., Feldman and Steenbergen 2001; Gilens 1999; Jacoby 1994; Schneider and Jacoby 2005; Smith 1987). Further, some research has identified clear differences in the underpinnings of Americans’ views on programs representing these various categories (e.g., Cook and Barrett 1992; Kam and Kinder 1999; Winter 2006).

This would not necessarily be the case for programs that include an explicit cost-of-living adjustment mechanism. Such a mechanism is in place for Social Security but not for several other social welfare assistance programs such as AFDC/TANF and Food Stamps.

An extension of this argument would suggest that citizens would respond to inflationary pressures generally: they would call for increased spending across the board, not just on social welfare policies. We address this possibility below.

The NES, at various points in time, has included other specific mentions that are relevant to social welfare policy, including assistance to the poor, unemployment assistance, and assistance to the homeless. We have not used these measures because they were included in limited years or were asked with inconsistent wording. The Welfare Programs item is available in the 1992, 1994, 1996, and 2000 and the Food Stamps item is available biennially from 1984–1996 and in 2000.

This statement uses average opinion on Food Stamps and on Welfare Programs. Spending on reducing crime, public schools, Social Security, AIDS research, child care, protecting the environment, financial aid for students, space exploration, and even spending on blacks all garnered more support than Food Stamps and Welfare Programs.

Jacoby (1994) combines spending on “assisting blacks,” “food stamps,” “Medicaid,” “unemployment compensation,” and “Social Security” into a social welfare scale. These specific program attitudes are used to predict general attitudes towards government spending (the services and spending item). Jacoby (2000) uses seven of the federal spending items (homeless, poor people, child care, unemployed, assisting blacks, food stamps, and welfare programs) to form a scale of social welfare spending preferences (note that social security is not included in this analysis); these social welfare preferences are contrasted with the more general services and spending item, which, Jacoby characterizes as “a general and nonspecific presentation of the [government] spending issue” (p. 753, FN 3). Feldman and Zaller (1992) analyze the services and spending item and the guaranteed jobs item, noting that these items have been “the primary NES social welfare issue items for the last decade” (p. 276, FN 5). Furthermore, they argue that these items relate well to other questions on “domestic spending priorities” and “general attitudes toward economic equality and redistribution” (p. 276, FN 5). Finally, the items have the virtue of containing the open-ended probes that form the core of the contribution of the article.

The correlations between the four items appear in the Appendix.

Family budget data “tally the cost of food, housing, transportation, clothing, personal items, medical care, and taxes for an urban family of four” (Berry et al. 2000, p. 552). We use Berry, Fording and Hanson’s data instead of the Bureau of Labor Statistics’ Consumer Price Index (CPI) because the BLS CPI provides only regional, rather than state-level, information.

In a departure from previous work, we also included inflation in both its raw and its quadratic form. The two theoretical arguments we address suggest either a positive relationship between good macroeconomic conditions and public support for social welfare (the procyclical pattern) or a negative relationship between good macroeconomic conditions and public support for social welfare (the countercyclical pattern). We added a layer of complexity to these statements in arguing that the relationship between macroeconomic conditions and public support for social welfare might be better modeled using a quadratic, because the marginal effect of changes in macroeconomic conditions might depend upon the level of macroeconomic conditions. Inclusion of a quadratic allows us to identify whether the effect of inflation on social welfare policy is monotonically increasing or monotonically decreasing, or whether it is nonmonotonic (that is, if the effect is positive up to some point and negative thereafter). This specification also allows us to examine whether the effect of inflation on support for social welfare spending varies with the level of inflation itself. This quadratic specification is especially appropriate for the indicator of inflation, as target levels of inflation (e.g., in countries such as the United Kingdom, Canada, and New Zealand (Bernanke and Mishkin 1997) and endorsed by U.S. Federal Reserve Board Chairman Ben Bernanke (Geewax 2005)) suggest that interpretation of inflation levels with respect to “good” or “bad” macroeconomic conditions is not unidirectional. Instead, too low or too high levels of inflation may be undesirable, as low inflation rates may destabilize the economy (Delong 2000; Reifschneider and Williams 2000; Saunders 2000). A one-unit shift in inflation, say from 1% to 2%, may have a different effect on support for social welfare policy compared with a one-unit shift in inflation from 4% to 5%. Inclusion of the quadratic permits the size and direction of the effect of inflation on support for social welfare policy to vary with the level of inflation. The estimated coefficient for the quadratic term was not significantly different from zero in either model; hence we include a linear specification.

We also examined the effects of unemployment at its lagged level, as a raw change between the current and previous year, and as a percentage change between the current and previous year. The effect of unemployment was indistinguishable from zero in each of these specifications. On the same idea that the effect of changes in unemployment may be conditional upon the level of unemployment, we included a quadratic here as well. In all models, the quadratic term on unemployment never obtained statistical significance and was substantively tiny.

We also examined the effects of GSP at its current level, its lagged level, and as a raw change between the current and previous year. The effect of GSP was indistinguishable from zero in each of these specifications.

Within the time frame analyzed, the correlation between state-level inflation and state-level unemployment was −0.32, the correlation between state-level inflation and change in per capita GSP was 0.043, and the correlation between state-level unemployment and change in per capita GSP was −0.009.

Support for social welfare ranges from 0 (conservative) to 1 (liberal). The datapoints refer to the year-state averages across both indicators (i.e., the average opinion on Welfare Programs, within a state for a given year, combined with the average opinion on Food Stamps within the same year and state). When only one indicator is available, the dot represents the year-state average for the one indicator.

The reader may notice a few outlying observations in our state-level data. Results run without the outlying observations on inflation (California respondents in 1990) are nearly identical. Results run without the outlying observations on unemployment (West Virginia in 1984, 1986, and 1992; Louisiana in 1986; Alabama in 1984; and Michigan in 1984) were substantively similar.

We conceptualize economic interests as falling into the following 2 × 2 matrix:

A considerable literature has examined pocketbook versus sociotropic determinants of electoral behavior (e.g., Kinder and Kiewiet 1979, 1981), using both objective and subjective indicators (see Lewis-Beck and Stegmaier 2000 for a review).

The disabled (3% of the sample) often receive Supplemental Security Income (SSI) and benefits from the Social Security Disability Insurance; they are excluded from analysis.

Note that the use of sociotropic information does not mean that individuals are altruistic: “The distinction between pocketbook and sociotropic politics is not equivalent to the distinction between a self-interested and an altruistic politics” (Kinder and Kiewiet 1981, p. 132). It simply means that contextual information about the state of the economy shapes political judgments.

We estimate each of our models with ordered probit, since the dependent variables are ordered categorical indicators. We include yearly fixed effects to allow for intercept differences across years, attributable to factors that we do not explicitly control in our model (such as political discourse, i.e., in the Welfare Reform era). We include regional dummies to capture time-invariant regional differences in the political culture surrounding social welfare policy. We are not able to include state-fixed effects due to the paucity of respondents from some states. These relationships could be modeled using hierarchical modeling or random effects models that attribute shared variance to individuals nested within states. We deal with this shared variance using robust standard error adjustments, clustering by year-state, which produce results that are quite similar to those produced using hierarchical techniques (Kam and Franzese 2007).

The predicted probabilities, based on the combined model, are graphed setting values to their means (or modes for categorical variables).

Note that this sociotropic measure is suboptimal, since we believe that most of the action occurs at the state-level. The NES only included a state-level economic assessments item in the 1992 survey.

Tables 3 and 4 provide results from ordered probit models, where respondents who reported “Don’t Know” or “Haven’t Thought About It” are set to the midpoint. We also estimated the models with selection equations, following Berinsky (2002). Our estimates of the (null) effect of state-level economic conditions remained unchanged. We also extended the time-period of our analysis. For the Guaranteed Jobs item, this meant that we went back to 1976; for the Services and Spending item, this meant that we went back to 1982. In so doing, we had to omit egalitarianism from the models (but since egalitarianism and the state-level economic indicators are not correlated, this should not have biased the estimated coefficients on the state-level economic conditions). The results were similar: null results for each of the state-level economic indicators.

These statistically insignificant coefficients could be attributable to collinearity between the multiple interaction terms. So, we estimated models with an interaction between black affect and only one of the macroeconomic indicators. The results were similar to those reported in Tables 5 and 6: a negative (but insignificant) interaction between feelings towards blacks and inflation appears when it comes to spending on Welfare and a positive but significant interaction between feelings towards blacks and inflation when it comes to spending on Food Stamps. The other interactions were insignificant. Finally, as a more direct test of whether changes in the composition of welfare recipients was responsible for more supportive views of welfare spending, we included several measures directly relating to the race of welfare recipients (e.g., the raw percentage of welfare recipients who were black; changes in the percentage of welfare recipients who were black). None of these were significantly related to support for Welfare Programs or Food Stamps.

Gilens’ analysis consists of an autoregressive (AR(1)) model, with a dummy included for the period between 1983 and 1988, in which there was an anomalous period of GDP growth. GDP is not analyzed alongside other macroeconomic indicators (unemployment or inflation) and no explicit justification is given for analyzing GDP instead of the other indicators.

Aggregate public opinion measures result from combining General Social Surveys, Roper polls, and two Consumer Attitudes and Behavior surveys. Like Gilens, Kluegel uses an AR(1) model to estimate the relationship.

We also analyzed the effects of national economic conditions—inflation, unemployment, and change in per capita gross domestic product. These analyses are a bit perilous, given the relatively few years that are available for analysis, and given the high degree of collinearity between the indicators. One of the symptoms of high collinearity is unstable coefficient estimates—especially estimates that change signs across models. We found that to be the case in our estimates. Hence, here are the estimates from the models where each economic indicator appears on its own. Notice that inflation is the only one that has a consistently positive effect, consistent with our state-level story. Unemployment rates have a negative relationship to Welfare Programs and a positive relationship for Food Stamps. The effect of productivity is significant but shifts direction: greater productivity is associated with less support for Welfare Programs (that is, the worse things are, the more support there is), but greater productivity is also associated with more support of Food Stamps.