A BRIEF OVERVIEW OF THE SURVEYS

Early opinion polls covered a wide range of topics and provide a treasure trove of insights into the dynamics of public opinion and political behavior. For example, most Gallup polls asked about presidential vote intention or recent presidential vote choice, with little variation in question wording. Congressional vote was included in about one-quarter of the polls. Presidential approval was included in over 70 polls between 1936 and 1945, again with little variation in wording. Gallup polls also asked a party identification question at least 39 times between 1937 and 1945.

The surveys also included numerous questions probing respondents' domestic policy attitudes. For decades, historians and political scientists have debated the sources of the halt of New Deal liberalism and the reasons that a conservative coalition of southern Democrats and Republicans rose to prominence in the U.S. Congress while Britain and other advanced democracies were moving toward a more comprehensive welfare state (see, e.g., Farhang and Katznelson Reference Farhang and Katznelson2005; Brinkley Reference Brinkley1995). On the whole, this literature has been largely divorced from the study of public opinion and mass political behavior. We believe that changes in mass attitudes—in particular, concerns about the growing influence of labor unions—played an important role in defining the limits of New Deal liberalism (Schickler and Pearson Reference Schickler and Pearson2009; Schickler and Caughey Reference Schickler and Caughey2010). These new data promise to put such claims to a clearer test. For example, a series of polls investigated attitudes toward labor unions, including approval of the sit-down strikes of 1937, concerns about wartime strikes, and views on the Congress of Industrial Organizations (CIO) and American Federation of Labor (AFL) leadership. An item asking whether the landmark Wagner Act of 1935 should be “revised, repealed, or stay unchanged” was repeated six times in polls during the early 1940s, as anti-union proposals were high on the congressional agenda. The surveys also included questions about respondents' more general attitudes toward New Deal liberalism, such as whether the New Deal ought to be extended to new policy areas, kept unchanged, or reversed. Gallup, OPOR, and NORC polls also included numerous questions tracking public reactions to the domestic side of the war effort, such as price controls, as well as questions about proposals for postwar expansion of the welfare state. For example, an August 1944 NORC survey asked over 2,500 Americans a series of questions about their medical expenses, health insurance status, and views on what the government should do to improve access to medical care (see Mickey and Schickler Reference Mickey and Schickler2008).

In addition, a host of surveys allow us to explore the public's attitude toward foreign relations. These surveys open new paths into well-trod fields of inquiry. While a number of political scientists have conducted systematic studies of public opinion concerning international interventions (Foyle Reference Foyle1999; Holsti Reference Holsti2004; Mueller Reference Mueller1973; Mueller Reference Mueller1994; Sobel Reference Sobel2001), these analyses have focused on the Cold War and post–Cold War periods, almost completely ignoring World War II. Berinsky (Reference Berinsky2007; Reference Berinsky2009) has explored the structure of support for war and found that in many ways, World War II was a war like any other. For instance, before U.S. entry into the war, Democrats who paid close attention to politics were more willing to adopt Roosevelt's aggressive foreign policy position and argue that the United States should risk war to aid the Allied countries. As Republicans' level of attention to elite discourse increased, on the other hand, they became less likely to support aiding the Allies. However, once the United States entered World War II in December 1941, elite discourse unified behind the president. Regardless of an individual's political predispositions, citizens with higher levels of political information expressed greater support for administration policies than did citizens with less information. As in other areas of politics, a unified government led to a unified public.

In sum, the public opinion data from the 1930s and 1940s hold great promise for illuminating scores of vital questions in the field of political behavior. But can the data be trusted? It is to this question that we turn next.

QUOTA SAMPLING AND WEIGHTS

The central problem in interpreting frequencies on variables of interest is that many of the survey samples do not represent certain groups in proportion to their population share. But although the quota-controlled sample data were collected in ways that appear haphazard from a modern vantage point, the data collection process introduced predictable deviations between the characteristics of the sample and that of the population. We can therefore employ methods designed to account for these measurable differences to make reasonable inferences about the U.S. population. Berinsky (Reference Berinsky2006) describes these methods in greater detail.

The present study employed a model-based poststratification weighting scheme.Footnote ⁴ We preferred to use cell weighting, because this approach is simple and requires minimal assumptions. When possible, the aggregate opinion results presented in this article were weighted through the cell-weighting method. In some cases (see the following), we used raking methods. It should be noted that researchers employing these different weighting methods will produce similar pictures of the shape of public opinion from this era (see Berinsky Reference Berinsky2006 for examples and details).

In aggregate analysis, we weighted the data on education levels, occupation, phone access, and quota category variables such as gender and region that could be matched to census data. When education levels were available, we created cell weights using education, gender, region, and (if available) race.Footnote ⁵ If the survey did not contain a measure of education, we created cell weights using a dummy variable for “professional” occupation, gender, region, and (if available) race. If neither education nor a reliable occupation variable was available, we created raking weights based on the marginals of gender, region, phone access, and race (when available). The necessary population counts for the 1940 census were drawn from the Integrated Public Use Microdata Series (Ruggles et al. Reference Ruggles, Sobek, Alexander, Fitch, Goeken, Hall, King and Ronnander2004), and aggregate census records of phone access were drawn from the U.S. Census Bureau's (1975) Historical Statistics of the United States, Colonial Times to 1970. Even when weighting made only a modest difference in our conclusions, it nonetheless provides more confidence that our estimates are not attributable to problematic sample design.

OPINION TREND DATA

To demonstrate the temporal richness and promise of the data, we present a series of figures that chart the data on certain questions that were repeatedly asked during this period. We list the question wordings for these items in the appendix. For series for which we had at least 15 measures of opinion, we present smoothed trend lines along with the individual data points.Footnote ⁶ For the other data series, we present the individual data points connected by interpolated lines.

One important question to answer is how our weighting changes the picture of public opinion in earlier times. To demonstrate the effects of weighting, we first present two time trends. The first, party identification (figure 1), is a case in which the use of weights shifts the distribution of opinion, increasing the estimate of Democratic Party identification by an average of 4%. This result is not surprising, because Southerners tended to identify with the Democratic Party and were significantly underrepresented in the polls from the 1930s and 1940s. However, our weights do not always alter estimates of public opinion. On questions relating to the use of military force, the sampling problems in the early polls produced biases that often largely offset one another: Southerners and women were both underrepresented, but the former tended to favor intervention, while the latter were more dovish. We can see this trend clearly on questions concerning opinion on the draft, presented in figure 2. The unweighted marginals turn out to be close to the mark, but only because the sampling problems led to biases in opposite directions. For example, weighting responses to the draft question never changes the aggregate support by more than half a point, and, on average, the unweighted support for compulsory service is the same as the weighted estimate.Footnote ⁷ It is worth noting, however, that in both the case of party identification and military intervention, the weighted and unweighted marginals vary over time in almost identical ways. Thus, while the use of weights affects the marginal distribution of some of the variables in our datasets, the effect of these weights is largely consistent over time. Thus, for the other series presented here, we present only the weighted trends.

Figure 1

Democratic Party Affiliation

Figure 2

Every Able-Bodied Man Should Serve in the Armed Forces

Figures 3, 4, and 5 expand our scope to three other survey trends from the 1930s and 1940s: presidential approval, support for helping England, and support for unions. A few points are worth noting here. First, many of the data series are quite rich. For instance, the presidential approval question was asked 76 times during this time period. Some of the issue questions also allow us to trace changes in opinion over time. The “help England” question, for instance, was asked 23 times in 1940 and 1941. Second, the time series seem to move in meaningful ways. For instance, reflecting the growing elite consensus about the importance of helping England in the year before Pearl Harbor, the “help England” series rises over time.

Figure 3

Presidential Approval

Figure 4

Support for Unions

Figure 5

Support for Helping England over Staying out of the War

Most important, these data allow scholars to examine the political significance of changes in opinion during the 1930s and 1940s. For instance, the data speak to the debate over macropartisanship, which in the past has been limited to surveys from 1953 onwards. A substantial increase in Roosevelt's approval rating from 1938 through the end of 1941 is evident (see figure 3). This sustained increase, which coincided with the rise of the war issue on the national agenda, corresponds to much less—if any—movement in Democratic partisanship (see figure 1). Roosevelt's approval peaks soon after Pearl Harbor but then enters a slow decline, although the rating does remain above the levels of the late 1930s. A year after Pearl Harbor, his rating still stood at approximately 75%, reflecting a roughly 20-point increase over his 1938 approval rating and a 15-point increase over his typical ratings in late 1939. By comparison, Democratic Party identification stood at about 50% in late 1942, which is barely above the party's standing in 1938–39. Democratic identification appears to fall slightly in 1943 and early 1944, but by the time of Roosevelt's 1944 reelection, it once again stands at approximately 50%.

Turning to the domain of domestic policy, while the “support for unions” series (figure 4) shows that most Americans expressed basic support for unions in principle, actual support dropped by about 10 points from 1937 to early 1941 before rebounding slightly during the war. We suspect that the other questions, which measure attitudes toward the Wagner Act and wartime strikes for shorter periods of time, might provide greater insight into the dynamics of mass attitudes toward labor. Indeed, the 10-point decline in abstract support for unionization corresponds to a period in which the polls indicated considerable popular disapproval of union tactics, such as the emergent sit-down strike movement of 1937 and strikes occurring in the defense industries in the run-up to the war (see Schickler and Caughey Reference Schickler and Caughey2010).

In the realm of foreign policy opinion, the most noteworthy pattern is the massive increase in support for interventionist policies in the years before Pearl Harbor. Majority support for the policy of helping England, even at the risk of war with Germany, was attained during the 1940 election and continued to rise before Pearl Harbor (see figure 5; for further discussion, see Berinsky Reference Berinsky2009). This increase in support can also be seen in the trends regarding support for instituting a draft. Support for the notion that all able-bodied men ought to serve in the armed services doubled from 1939 to early 1941 (see figure 2).

FUTURE RESEARCH

The survey questions about presidential approval, vote choice, and union support, and the foreign policy series discussed here are only a small fraction of the questions available in the database. Extensive series also exist on topics ranging from racial politics to economic conditions, congressional vote choice, court packing, social welfare policy, labor policy, war, and prohibition. We present several of these additional series in an online appendix, available at http://igs.berkeley.edu/schickler/docs/publicopinion30s40s.pdf. We hope that this brief review of certain time trends sparks interest in opinion poll data from the 1930s and 1940s. As scholars and journalists consider the possibility that a deep economic crisis will again redraw the contours of American politics, the rich data that exists on public opinion in an earlier economic crisis can serve as a useful point of departure in considering public responses to bold economic experiments. We hope that scholars in a variety of fields will use this understudied resource to more deeply investigate this important period and help shed light on new theories of politics and public opinion.