EXISTING LITERATURE

Much of the earliest work about the Tea Party was written by journalists who placed a premium on enhancing its general understanding of the Tea Party as a social movement, particularly with respect to its base (see, e.g., Zernike Reference Zernike2010). Academics, meanwhile, debated the individual-level determinants of Tea Party support—namely, whether they tend to have latent racist attitudes (Arceneaux and Nicholson Reference Arceneaux and Nicholson2012; Barreto et al. Reference Barreto, Cooper, Gonzalez, Parker and Towler2011; Williamson, Skocpol, and Coggin Reference Williamson, Skocpol and Coggin2011).

Other researchers explored the electoral impact of the movement. Karpowitz et al. (Reference Karpowitz, Monson, Patterson and Pope2011) demonstrated that an endorsement by a Tea Party group had a negligible impact on the outcome of general elections; however, in some cases, it helped to secure the victory of Tea Party favorites over Republican moderates in primary elections. Jacobson (Reference Jacobson2011) found that the movement energized an older, whiter, more conservative electorate, which helped to deliver a historic victory to the Republicans in the 2010 midterms. Abramowitz (Reference Abramowitz2011) analyzed survey data, which revealed that Tea Party members overwhelmingly identified with the Republican Party and were, by far, the most conservative and most activist bloc within the party. Libby (Reference Libby2013) argued that the Tea Party movement had already an important impact on the Republican Party and was likely to retain its influence for many years to come. Others tested the “Tea Party effect” by evaluating the messaging of Tea Party candidates versus establishment candidates (Livne et al. Reference Livne, Simmons, Adar and Adamic2011).

For this article, we devised a direct test of whether the language and principles associated with the Tea Party movement have had a radicalizing effect on the positions taken by Republican candidates in the 2012 primaries as compared to those of 2008. If we are correct, then the candidates in 2012 voiced positions more characteristic of an anti-Washington Tea Party ideology than the traditional social conservative ideology. We tested this proposition using text analysis, extracting the candidates’ respective positions from their announcements and primary debates in both 2008 and 2012 and announcements in 2012.

HYPOTHESES AND METHODS

Williamson, Skocpol, and Coggin (Reference Williamson, Skocpol and Coggin2011) discussed a range of scenarios likely to emerge as a result of the Tea Party movement and concluded that the most likely outcome was that “the presence of newly elected Tea Party representatives will only reinforce the 2009–10 Republican strategy of total opposition to the Obama agenda.” This suggested that rather than serving as a political sideshow, the Tea Party would radicalize the mainstream Republican position in national (i.e., presidential) elections. Our study directly tested this hypothesis by analyzing shifts in the messaging of Republican politicians in two presidential elections that “bookended” the emergence of the Tea Party movement.

Specifically, we test the claim that the messages of mainstream Republican presidential candidates in 2012 were significantly different than those of Republican candidates in the 2008 primaries. If, as we claim, the emergence of the Tea Party movement in 2009 had a radicalizing effect on the Republican presidential candidates in 2012, then we should expect that they had an overall more anti-Washington (versus traditional social conservative) message in the 2012 versus the 2008 primary elections. Our claim receives even stronger support if we can observe ideological radicalization of individual candidates who ran in both elections (i.e., Mitt Romney and Ron Paul).

Candidates in US presidential primaries are expected to take positions on a number of issues, yet they rarely generate manifestos that reliably address all important issues. However, in the course of presidential primaries, candidates produce numerous shorter written and spoken statements that encompass their positions on all important campaign issues, particularly those that concern their general approach toward governance. We used a subset of these statements to extract the positions of the main Republican candidates in 2008 and 2012.

Scholarly analysis of texts approaches the problem as either discourse or content. The most widely used text-analytic methods are highly labor-intensive, relying heavily on human judgment. Computer-assisted content analysis is largely free of these problems. ^{Footnote 1} We applied a computer-assisted content-analytic method, known as Wordfish (Slapin and Proksch Reference Slapin and Proksch2008) and as implemented by Lowe (Reference Lowe2013), that relies on factor analysis of word frequencies. The method has been shown to produce valid estimates of positions, ^{Footnote 2} and it has been applied to a variety of political documents. ^{Footnote 3} Each count W of word j in document i is modeled as a draw from a Poisson distribution with rate λ_ij (1). In the following equations, (2) the rate is modeled as a function of the sum of the document fixed effect α_i, the word fixed effect ψ_j, and the product of the word slope β_j with the document position θ_i. The document positions and word slopes are the crucial inferential quantities. The fixed effects account for the fact that some documents are longer and some words are more common.

(1) $${{\rm{W}}_{{\rm{ij}}}}\, \sim \,{\rm{Poisson}}\left( {{\lambda _{{\rm{ij}}}}} \right)$$

(2) $${\rm{log}}\left( {{\lambda _{ij}}} \right)\, = \,{\alpha _{\rm{i}}}\, + \,{\Psi _{\rm{j}}}\, + \,{\beta _j}\,{\theta _i}$$

DATA

Our analysis focused on the pre-Iowa caucus Republican presidential debates because, in this period, the candidates had yet to maneuver in response to competition in the primary battles. Therefore, their positions were more likely to reflect their own ideological positions. For the sensitivity analysis, we used the candidates’ presidential candidacy announcements. The debates and candidacy announcements differ substantially—the debates are spoken words whereas the announcements are prewritten. The sensitivity analysis was restricted to the 2012 cycle because the 2008 cycle announcements were not in a consistent format. ^{Footnote 4}

For the analysis, we selected those candidates who participated in at least two thirds of the televised presidential debates before the Iowa caucus—that is, in at least seven of the 11 pre-Iowa debates of the 2008 cycle or nine of the 13 pre-Iowa debates of the 2012 cycle. Included and excluded candidates are listed in table 1. Tables 5 and 6 in the online appendix (available on the PS website at http://dx.doi.org/10.1017/S1049096514001085) report details about the debates and announcements, including the sources of their transcriptions.

Table 1

Candidates

Candidates who took part in the debates selected for the analysis and the number of the selected debates they have attended. Candidates excluded from the analysis are at the bottom of the list.

We treated all utterances of each candidate in all selected debates as a single document. The resulting compiled texts can be expected to encompass the majority of relevant issues because individual debates tend to focus on a narrow subset.

In our analysis, we made no ex-ante assumptions about how informative each word (i.e., “stem”) is relative to an ideological position other than assuming that the most common function words are not informative. Instead, we used the data and the model to assess how informative each included word is. Thus, we excluded only the most common function words (i.e., “stop-words”) and numbers. ^{Footnote 5} Including primary-specific issues would have artificially inflated the differences between the 2008 and 2012 candidates because related words often are common in one cycle but rare in the other. Therefore, we selected the stems that were used in each cycle by at least two (i.e., 15%) of the selected candidates; we included a total of 1,513 stems in our analysis. ^{Footnote 6}

SENSITIVITY ANALYSIS

To test the sensitivity of the analysis, we compared the ideological dimensions derived from the 2012 candidate announcements with the 2012 debates. ^{Footnote 7} Whereas the debates can be expected to encompass a larger set of specific issues, the announcements—much shorter texts produced at the beginning of a campaign—focus on two very general topics: namely, the general values emphasized by the candidate and the qualities of the candidate. Furthermore, although candidates prepare carefully, the messages they convey in the debates are considerably more improvised than those embedded in their announcements. We selected stems that were used by at least two candidates. Although this analysis did not consider the 2008 announcements, the dimensions extracted from both the announcements and the debates are remarkably close to one another (as shown in figure 3) and strongly correlated: Pearson’s ρ = 0.94 (95% confidence interval: 0.69, 0.99).

Figure 3

Positions of 2012 Candidates

Positions of 2012 candidates extracted from pre-Iowa debates and candidacy announcements.