Blame, Responsibility, and the Tea Party in the 2010 Midterm Elections

Gallup, “Gallup Daily: Obama Job Approval,” http://​www.​gallup.​com/​poll/​113980/​Gallup-Daily-Obama-Job-Approval.​aspx.

It is important to note, however, that others have concluded that Tea Party endorsements did not improve the chances of electoral victory for Republican candidates (Bond et al. 2011; Jacobson 2011b; Karpowitz et al. 2011).

Supporting this observation, a survey of Tea Party supporters found that while most Republicans expressed dissatisfaction with Washington politics, Tea Party supporters were more likely to express anger toward Washington. (New York Times/CBS News Poll, National Survey of Tea Party Supporters, April 5–12, 2010, http://​documents.​nytimes.​com/​new-york-timescbs-news-poll-national-survey-of-tea-party-supporters?​ref=​politics).

Los Angeles Times Poll, October 17–19, 1994, http://​articles.​latimes.​com/​print/​1994-10-21/​news/​mn-53022_​1_​times-poll.

Respondents were selected from YouGov’s PollingPoint panel, an opt-in Internet panel, and then matched on a set of demographic and political characteristics to a random sample (stratified by age, gender, race, education, and region) from the 2005–2007 American Community Survey. The sample was weighted using propensity scores based on age, gender, race, education, news interest, voter registration, and non-placement on an ideology scale. This method produces a sample that looks similar to a probability sample on the matched characteristics, but may still differ in unknown ways on unobserved characteristics. Research comparing samples using this method to telephone and face-to-face surveys finds that such samples are similar in many ways (Yeager et al. 2011), although there remain concerns about generalizing results to a broader population (AAPOR 2010; Pasek and Krosnick 2010).

Respondents were also given the option of selecting the answer “I’m not sure” in response to this question. Thirty-nine respondents selected this option, and they are omitted from our analysis.

This quantity and those that follow are computed using team module weights.

More detailed information is reported in the “Appendix” section. Because respondents who selected “blame both” were slightly less educated and knowledgeable about politics, we further tested the alternative hypothesis that these voters supported Republican candidates because of a lack of sophistication. We found several reasons to discount this alternative hypothesis. First, respondent education, political sophistication, and interest in politics are included as controls in all multivariable models. Second, because respondents were given the option of answering “I’m not sure” in response to the blame attribution question, there is little reason to believe that the respondents who selected the “blame both” option did so because they were unable to select a substantive answer. Third, as an additional robustness check, we compared these “I’m not sure” and “blame both” respondents across other questions in the survey to see if the “blame both” respondents were expressing non-attitudes when given the opportunity on other attitudinal and knowledge measures. Large percentages (around 45–60%) of those who answered “I’m not sure” on the blame question also selected “I’m not sure” in response to other attitudinal and knowledge questions, including questions asking them to place the Democratic and Republican Parties on an ideological scale and evaluate the Supreme Court and their states’ governors. The “blame both” respondents answered “I’m not sure” at much lower levels on these same questions (5–11 %), rates comparable to those who blamed either the Democrats or Republicans. Finally, closer scrutiny of the congressional vote by sophistication among the “blame both” respondents finds no evidence that the less sophisticated were simply casting a “nature of the times” vote for the GOP; rather, it was the most sophisticated respondents who were the most likely to vote GOP, suggesting they were cognitively capable of both blaming both parties but holding the Democrats responsible. We thank an anonymous reviewer for suggesting that we examine the relationship between blame attribution and political sophistication more closely. These results are available from the authors upon request.

This analysis classifies partisan leaners as independents.

We estimated a multinomial logit model predicting respondent selection into blame categories, with “blame Republicans” as the base category. In this model, party identification, Obama approval, issue attitudes, educational attainment, and interest in the campaign were predictive of blaming both parties for the state of the nation (relative to blaming Republicans). These model estimates are available from the authors.

This quantity was computed by summing the number of pieces of legislation for which a CCES respondent expressed “support,” and dividing this quantity by the number of items for which each respondent offered a directional response (“support” or “oppose”).

Herszenhorn and Hulse (2009).

Krugman (2009, 2011).

The correlation between Tea Party support and the “blame both” variable is −0.03. A majority of “blame both” respondents indicated that they were not supporters of the Tea Party movement (56 %). Among those respondents who “blamed both” and supported the Tea Party, most of these individuals did not participate in Tea Party events. These results can be found in the “Appendix” section.

We also estimated models using separate issue attitude variables, measuring views toward taxes and spending, abortion, immigration policy, and gun control, but including these variables instead of the issue liberalism scale did not change our conclusions.

The factor analysis yielded a single factor with an eigenvalue greater than 1. This primary factor accounted for 50 % of the variance in the eight policy attitudes used in the analysis. The latent dimension yields scores which range from −1.8 to 1.6, where higher values indicate more consistently liberal opinions on policy issues.

We used a five category variable measuring responses to this item, which incrementally ranges from “Strongly Disapprove” to “Strongly Approve.” Those who selected the “Not sure” option were assigned to the intermediate category. We use separate dummy variables for Democratic and Republican incumbents.

We estimated additional models using a 3-category, a 5-category (with categories for strong Republicans, weak and leaning Republicans, Independents, weak and leaning Democrats, and strong Democrats), and dummy variables for each party identification category. In each of these models we found results analogous to those reported below.

The wording for the campaign interest item was "Some people seem to follow what's going on in government and public affairs most of the time, whether there's an election going on or not. Others aren't that interested. Would you say you follow what's going on in government and public affairs?" Response options included “hardly at all,” “only now and then,” “some of the time,” and “most of the time,” coded in that order. To measure political knowledge, we relied on an item from the pre-election wave which asked respondents to identify partisan majority control in four legislative chambers—the U.S. Senate, the U.S. House, the respondent’s state Senate and state House. Responses were summed from 0 to 4, with higher values indicating more correct answers.

We explored several alternative specifications of the models reported in Table 4. Most notably, we included an indicator for freshmen Democratic members to see if newly elected members faced an especially difficult electoral environment. The variable was not statistically significant from zero and did not affect the parameters reported in the table. We also estimated the model with an indicator if the member voted for the ACA bill; and replicated with an interaction between a respondent’s issue preferences and Tea Party support. Again, these variables were not statistically significant.

Wald test of joint statistical significance for both incumbency variables was statistically significant (\(\chi_{2}^{2} = 17.8, \;p < 0.05)\)).

This result, and those that follow, were obtained by simulation using CLARIFY (King et al. 2000). Covariates were held at the following values: Obama approval (“somewhat disapprove”); Tea Party support (“not a supporter”); Policy attitude scale (“0”); blame variables (“blame both”); family income (“$50,000–$59,999”); education (“2 year degree”); age (“55”); sex (“male”); African-American racial identity (“non-black”). Simulations were based upon the model reported in column 2 of Table 4.

The fact that presidential approval is not statistically significant in models 2 and 3 likely can be attributed to the large number of highly correlated variables in this model. Model 1 demonstrates that presidential approval is a significant predictor of vote choice.

The more complex models in columns 2 and 3 significantly improve the model fit beyond the referendum model reported in column 1 (\(\chi_{11}^{2} = 35.2,\; p < 0.05\), column 2; \(\chi_{12}^{2} = 34.8, p < 0.05\), column 3).

Again, given the potential alternative explanation that those who blamed both might be making a simple-minded “nature of the times” decision based on a lack of sophistication, these controls for these political sophistication variables are especially important. Although these measures are not statistically significant given the other variables in the model, we conducted a series of other analyses more closely examining the relationship between political sophistication, blame, and vote choice, summarized in footnote 9.

For this calculation, we estimated the predicted vote of each respondent based on column 2 of Table 4. This yields a 56.1 % Republican share of the two-party vote. If we recalculate such that those who “blamed both” were evenly divided between voting for Democrats and Republicans, the Republican share of the two-party vote is predicted to be just 52.4 %.

To obtain this result, we first used the model parameters from column 2 of Table 4 to predict in-sample vote choice. Voters whose Democratic vote probability was greater than 0.5 were classified as Democratic votes, while those with a probability of <0.5 were classified as Republican votes. For the counterfactual described in the text, we changed the coefficient for the “blame both” variable to zero. Then we used the remaining model parameters to generate an in-sample prediction for this scenario. Under the counterfactual described in the text, Democratic vote share moves from about 43 % to about 48 %.

e.g., see Roberts (2010), Klein (2010), Salvanto and Gersh (2009).