A Delicate Balance: Party Branding During the 2013 Government Shutdown

Abstract

Strong party brands help congressional parties elect candidates, maintain or gain majority control, and advance their policy agendas. Because successful branding efforts depend on consistent messaging, party leaders try to choose issues that most members are willing to promote. But what do leaders do when a party majority pressures them to take up issues that harm the brand for others? We investigate the 2013 government shutdown as a branding event. House Republican leaders instigated the shutdown after learning that a majority of Republicans would not vote for a clean funding bill. However, instead of highlighting the issues that led to the shutdown, they publicized the party’s efforts to resolve it. Party leaders sought to exploit the fact that party brands have both position and valence components to simultaneously address the demands of the party base and the electoral concerns of members representing competitive districts.

Keywords

party brand government shutdown social media text-as-data political communication

Introduction

Voters use party labels as informational shortcuts to infer positions, values, and attributes to candidates (Campbell, Converse, Miller, & Stokes, 1960). Party branding research starts with the premise that citizens develop their perceptions of parties by observing the actions of party elites. Much of this research further assumes that citizens identify with, and are more likely to vote for, the party they perceive to be closest to them ideologically.

Because brands are based on perceptions, congressional party leaders behave strategically in setting legislative agendas and publicizing actions. They try to advance issues that instantiate a vision of a prototypical Republican or Democrat favored by voters (Lupu, 2013). Yet an unresolved question is how leaders are able to promote ideological brands that benefit most party members? As elected party leaders, they must be responsive to the party majority while helping the party retain or attain majority control. The first objective implies that leaders should advance issues that reflect the preferences of the party median, while the second implies that they should be attentive to the electoral needs of the chamber median (Cox & McCubbins, 1993; Krehbiel, 1998).

One approach to mitigating such potential conflicts is to keep coalition-dividing issues off the agenda or out of the public eye (Cox & McCubbins, 2005; Sellers, 2010). But what do leaders do when a party majority demands action on a salient issue that threatens the brand for other members of the coalition? We argue that leaders are sometimes able to reframe such events through their public communications to promote a more inclusive brand. Brands are not based solely on issue positions. Stokes (1963) finds that voters also care about broader societal conditions such as the state of the economy. Cox and McCubbins (2005) assume that parties are rewarded for their ability to get things done in the legislature. In each case, voters are judging parties on their perceived effectiveness rather than their perceived positions.

We illustrate this argument by investigating how Republican leaders and members communicated about the 2013 government shutdown on Twitter. We use automated textual analysis to examine mentions of policy and legislative competence in more than 11,000 tweets sent by Republican leaders and lawmakers during the shutdown. With their congressional majority at stake, Republican leaders emphasized the party’s dedication to resolving the shutdown instead of emphasizing the policy positions that brought it about. This emphasis on party effectiveness allowed rank and file members to support a unifying branding message, while permitting more extreme members to bring up policy in their own communications with constituents.

Components of Party Brands

Voters who identify with a party are more likely to turnout and support its candidates (Campbell et al., 1960; Jacobson, 2012). The importance of party affiliation in elections has led to substantial scholarship investigating how voters develop their perceptions of parties, how candidates use party labels in campaigns, and how elites attempt to shape the “party brand” (e.g., Butler & Powell, 2014; Groeling, 2010; Grynaviski, 2010; Neiheisel & Niebler, 2013; Sellers, 2010; Snyder & Ting, 2002).

Brands are markers that simplify choices (Keller & Lehmann, 2006) Much of the research on party branding assumes that voters seek to place parties on a left–right ideological continuum (Downs, 1957). Thus, the goal of branding efforts is to shape voters’ perceptions of these positions. For example, Grynaviski (2010) argues that “party government depends on voters believing that party leaders will follow through on their commitment to pursue their party’s program” (p. 50). He portrays voters as Bayesians who base their perceptions of party positions on observations of elite behavior. Additional research confirms that stronger position-based brands lead to greater electoral success for a party’s candidates (Woon & Pope, 2008).

However, other studies argue that voters judge parties on more than just their issue positions. In 1963, Donald Stokes published a critical review of An Economic Theory of Democracy in which he distinguished between “position-issues” (“those that involve advocacy of government actions from a set of alternatives over which a distribution of voter preferences is defined”) and “valence-issues” (“those that merely involve the linking of the parties with some condition that is positively or negatively valued by the electorate”) (p. 733). Stokes found that voters often link parties with broader societal conditions (such as the state of the economy). Valence branding has since come to mean “any non-policy advantage a candidate or party might have” (Stone & Simas, 2010), such as name recognition (Groseclose, 2001), trustworthiness (Stone & Simas, 2010), and governing ability (Butler & Powell, 2014).

The most prominent example of an application of valence-branding to legislative organization and behavior is Cox and McCubbins’s (2005) party cartel theory. Because voters reward parties for their accomplishments (rather than positions), all members of the party have an incentive to support strong party leadership agenda control:

The more favorable is the majority party’s record of legislative accomplishments, the better its reputation or brand name will be… The better the majority party’s brand name, the better will be the prospects for (re)election of its various candidates and the better will be the prospects for (re)attainment of majority status. (Cox and McCubbins, 2005, p. 7)

Additional research confirms that voters are more likely to support candidates from a party that gets things done, and that party leaders are more likely to pressure rank and file members on decisions where the party’s governing reputation is at stake (Butler & Powell, 2014).

However, cartel theory also highlights a dilemma for leaders. According to Cox and McCubbins (2005), leaders follow two agenda setting rules: “Thou shall not aid bills that will split thy party,” and “thou shalt aid bills that most in thy party like” (p. 24). All party members benefit when a party has a reputation for getting things done, but attaching the party name to particular policy positions may benefit some while harming others. As a result, leaders may find it difficult to hold their coalitions together. They may be forced to choose between addressing the electoral needs of the moderates who hold the key to retaining the chamber majority,¹ and responding the demands of the party base that holds the key to their own leadership positions.

Party Messaging

When faced with intraparty policy conflicts, how do party leaders promote party brands that are supported by and benefit most party members? Brands can be promoted through actions (as in the cartel theory) and through public communications. Coordinated messaging campaigns in Congress date to at least the 1990s and are now institutionalized within both parties (Evans & Oleszek, 2001; Groeling, 2010; Jacobs & Shapiro, 2000; Sellers, 2010). Sellers (2010) specifically addresses intraparty conflicts in the context of messaging efforts during policy debates.

Successful messaging campaigns depend on party members’ willingness to promote them. The worst outcome for leaders is when members of the coalition decide to promote a message that conflicts with the one leaders are promoting (Sellers, 2010, p. 36). Sellers argues that leaders will avoid bringing up issues that create such intraparty conflicts.² If leaders are not able to keep such issues off the agenda, “the potential defectors must balance the collective benefits from keeping silent and allowing a more unified public presentation from their party, and the individual benefits from publicly stating their own dissent and undermining the party’s collective image” (Sellers, 2010, p. 39).

We propose that leaders still have opportunities to promote unifying branding messages in such contexts. During the 2013 government shutdown, Republican Party leaders were confronted with a divided coalition on a salient issue they could not avoid. Instead of leaving it to members to decide whether they would publicly support or oppose the shutdown, GOP leaders promoted a messaging strategy that emphasized the party’s efforts to resolve it. This valence messaging frame helped to attract broader coalition support than would have been the case if leaders had emphasized the position frame favored by the party base. To test this argument, we investigate how Republican leaders and members communicated about the shutdown on Twitter.

The 2013 Government Shutdown as a Branding Event

On October 1, 2013, the federal government suspended all nonessential functions. This “shutdown” furloughed about 800,000 nonessential federal government employees while another 1.3 million deemed essential reported to work not knowing when they would be paid. The most visible effects included gated National Parks, suspended Head Start programs (which led to a highly publicized US$10 million private donation), reduced veterans’ services, and the possibility of delayed tax refunds (Schwartz, 2013). The estimated longer term consequences included a .25 percentage point reduction in annualized gross domestic product (GDP) growth rate and 120,000 fewer private-sector jobs (Executive Office of the President of the United States, 2013).

House Republican leaders had an opportunity to prevent the shutdown. The Senate passed a clean funding bill (H.J. Res. 59) and sent it to the House on September 30. A bipartisan majority of Democrats and moderate Republicans was prepared to support it.³ However, House GOP leaders refused to schedule a vote before the clock ran out.

House Speaker John Boehner was clearly a reluctant warrior. He later called the shutdown a “predictable disaster” (Memoli, 2014). But he felt he had little choice but to support it as party leader:

When I looked up, I saw my colleagues going this way. You learn that a leader without followers is simply a man taking a walk. So I said, ‘You want to fight this fight? I’ll go fight the fight with you. (O’Keefe, 2014)

Conservative Republicans were elated. They saw a shutdown as their best opportunity to defund the Affordable Care Act, described Boehner’s decision as “wonderful” (Rep. John Culberson [R-TX]) and predicted that “people will be very grateful” (Rep. Michelle Bachmann [R-MN]). Although it might be “painful,” it was still the right thing to do (O’Keefe & Helderman, 2013). Other Republicans angrily warned that the shutdown could cost the House its majority (Rep. Adam Kinzinger [R-IL]; Siddiqui, 2013); predicted that it “ends badly for the American people and the Republican Party” (Rep. Reid Ribble [R-WI]; Weisman, 2013); and described it as the “dumbest idea” ever (Senator Richard Burr [R-NC]; Zwillich, 2013).

Most Americans, including most Republican identifiers, opposed the shutdown. A widely reported CBS poll conducted the day before found that 80% of respondents considered it an unacceptable way to negotiate (Dutton, De Pinto, Salvanto, & Backus, 2013). A poll commissioned by Republican lawmakers a month earlier found that 56% of those planning to vote for Republicans in the next election opposed shutting down the government.⁴ Sixty-three percent of “very” conservative respondents did favor a shutdown, but they were just 10% of respondents.

The shutdown finally ended 17 days later when Speaker Boehner scheduled a vote on a new bill (H.R. 2775) to raise the debt ceiling and fund the government at current levels through February 7, 2014—a bill that was opposed by most (62%) House Republicans but passed with the support of a bipartisan majority. Republicans understood that the shutdown was unpopular with many Americans. Most of the bills and resolutions proposing to restore funding for popular programs (such as National Parks and veterans’ services) were Republican sponsored.⁵ We hypothesize that—after demonstrating their allegiance to the party base by instigating the shutdown—House GOP leaders then turned their attention to mitigating its damage to the brand for party moderates. More specifically, we expect to find that leaders enagaged in a messaging strategy that emphasized the party’s effectiveness in resolving the shutdown rather than emphasizing the policy differences that brought it about. Although we expect members from marginal electoral districts to pay little attention to policy, and tea party members to do the opposite, we expect to find broad support within the caucus for the more inclusive valence frame promoted by the leadership.

Party Branding Through Social Media

We investigate how Republicans communicated about the government shutdown on Twitter. Politics-related Twitter research has rapidly advanced from basic descriptive analyses of usage patterns, to serious investigations of whether tweets can be used to infer preferences and predict behavior (Barbera, 2015; Barbera & Rivero, 2014; Cummings & Wang, 2010; Gayo-Avello, 2012; O’Connor, Balasubramanyan, Routledge, & Smith, 2010; Tumasjan, Sprenger, Sandner, & Welpe, 2010). Research involving congressional Twitter communications are more limited. Golbeck, Grimes, and Rogers (2010) and Hemphill, Otterbacher, and Shapiro (2013) report that lawmakers use Twitter to share information, advertise and credit claim. Barbera (2015) estimates member ideology using information about lawmakers’ Twitter followers in lieu of roll call votes.

Twitter has attributes of a powerful medium for party messaging. The 140-character limit encourages succinct themes. In contrast to press releases, members use it to communicate directly with their followers (or anyone who subscribes to a hashtag). Research also indicates that reporters and bloggers increasingly turn to Twitter to source news stories (Oriella PR Network, 2012). Finally, as press officers increasingly rely on programs such as Hootsuite and Buffer to distribute content across multiple social media platforms (Twitter, Facebook, Instagram, etc.), Twitter posts are increasingly representative of members’ broader communications activities (Casas & Morar, 2015).

Twitter has also become a common form of legislative communication among lawmakers. In October 2013 (the month of the shutdown), all but 10 members of Congress had accounts. The average Republican lawmaker had 8,521 followers or subscribers (for a total of 2,471,090). Former presidential candidate Sen. John McCain (R-AZ) had the most followers (around 1,800,000), followed by House Speaker Boehner (570,000), Marco Rubio, Ron Paul, and Michelle Bachmann.

Method

Our first objective was to identify tweets by Republican lawmakers that were about the shutdown. The next was to label shutdown-related tweets for whether they included references to policy or the party’s efforts to resolve the shutdown. We first manually annotate a random sample of tweets and then use a portion of this sample to train a supervised machine learning algorithm. The predictive accuracy of the algorithm is then tested on a held out set of manually labeled cases. Finally, the algorithm is used to automatically label the remaining unlabeled tweets.

We used the Twitter API to retrieve 11,505 tweets by 286 Republican members of Congress sent between September 23 and October 20 (the shutdown started on October 1 and ended on October 17).⁶ Two researchers then labeled 1,000 randomly drawn tweets for four nonmutually exclusive dichotomous variables: whether the tweet was about the shutdown; whether it mentioned policy; whether it mentioned party competence; and whether it blamed Democrats for the impasse. The blame category was not part of our original research design but emerged during the labeling process. Grimmer and King (2011) also found “partisan taunting” to be an important theme in members’ press releases. The agreement between the annotators (after adding the partisan blame theme) was 96% for the shutdown variable, 96% for the position variable, 90% for the performance variable, and 92% for the blame variable. Table 1 provides examples from each category.

Table 1.

Example Tweets for the Four Categories.

Name	Message
Not about shutdown
Doug Lamborn (C0-5)	With Colorado Springs constituent Jennie Dangers and her newly-adopted daughter Elizabeth.
Cathy McMorris (WA-5)	Always great to be home getting good local food in Spokane. @SweetFrostings, Dominis and its not even lunch yet!
Policy
Walter Jones (NC-3)	As the layers of the #Obamacare onion are peeled back, we’re getting a better sense of just how much it stinks.
Marsha Blackburn (TN-7)	Federal public debt accounts for 73% of national debt. We’ve got to get this under control or our grandchildren will face bigger crises.
Effectiveness
Gregg Harper (MS-3)	The House has passed three bills to keep government open. Now the #SenateMustAct
Steve Southerland (FL-2)	Burning the midnight oil: expecting votes to go past midnight tonight as House works to avert #shutdown and ensure troops are paid. #sayfie
Partisan blame
Tim Huelskamp (KS-1)	Instead of being ”Master of insults” @senatorReid should come to the table and reopen the government #EndReidShutdown
Keith Rohfus (PA-12)	Glad to answer calls from #PA12 constituents today. #Shutdown was preventable, but @SenatorReid needs to work w/ us.

The 11,505 tweets were preprocessed by removing word stems, numbers, and rare words (those that appeared in fewer than 0.001% of tweets). The labeled examples were then used to train the Support Vector Machine (SVM) algorithm to predict shutdown-related tweets as well as the substance of those tweets (Joachims, 2002). More specifically, we trained on 75% of the labeled cases and set aside the other 25% for testing, iterating through all possible combinations of train/test sets. This process provides more robust information about performance compared with a single partition of the train/test data.

Figure 1 reports precision and recall performance for each of the dichotomous variables. The vertical lines in the figures indicate the baseline—what a random draw from the sample predict in each case.⁷ Precision indicates how many of the cases that are predicted to be about the shutdown are actually about it (according to the “gold standard” human annotators). Recall indicates how many of the true cases are correctly predicted to be true. Lower precision indicates more false positives. Lower recall indicates more false negatives. The density distributions indicate that the algorithm’s predictions exceed the baseline in almost every trial and that average performance is substantially better for every variable (Appendix A provides specific numerical information about the figures).

Figure 1.

Predicting four dichotomous variables (N-fold cross-validation).

The final step in the labeling process was to predict whether the remaining 10,505 tweets were about the shutdown and whether they mentioned policy, effectiveness, or blame. We use the 1,000 training examples to predict shutdown-related tweets, and only the shutdown-related training examples to predict tweet substance. Table 2 presents the final numbers.

Table 2.

Number of messages for each category.

All messages
11,505
(H = 9,439, S = 2,066)
No shutdown	Shutdown
3,847 (33%)	7,658 (67%)
(H = 3,154, S = 693)	(H = 6,285, S = 1,373)
	Policy	Effectiveness	Blame
	3,239	4,218	1,696
	(H = 2,544, S = 695)	(H = 3,610, S = 608)	(H = 1,482, S = 214)

Findings

If Twitter can provide insights into party messaging, we expect to observe differences in how lawmakers tweet about the shutdown. Supporters should be more likely to mention the policy differences driving it. Other Republicans should say less about policy and more about the party’s efforts to resolve the impasse. Figure 2 begins to test these expectations by comparing messaging emphasis among some prominent lawmakers. The main Republican leaders in the House and Senate mentioned party effectiveness 60% of the time and policy just 34% of the time.⁸ In contrast, leading Tea Party Republicans mentioned policy in 70% of their tweets and party effectiveness just 18% of the time.

Figure 2.

Policy versus effectiveness mentions in tweets (selected lawmakers).

House Republicans had more reason to support a branding strategy that emphasized the party’s efforts to resolve it because they were more accountable for the unpopular shutdown than Senate Republicans. Figure 3 confirms that House Republicans as a whole were more likely to discuss efforts to end the shutdown in their tweets than senators.

Figure 3.

Policy versus effectiveness mentions in the House and Senate.

A successful branding effort requires member support. We expect most House members to support their leaders’ emphasis on party effectiveness regardless of electoral circumstances. At the same time, House Republicans representing competitive districts where the shutdown was unpopular should be even less likely to mention it in their tweets and more likely to emphasize the party’s efforts to resolve it when they do. Members representing safer districts have more leeway and constituents who are more supportive of the shutdown. They should be more likely to tweet about the shutdown and the policy issues driving it.

Figure 4 groups House Republicans by district competitiveness (quantiles) using President Obama’s share of the two-party vote in 2012 (Canes-Wrone, Brady, & Cogan, 2002; Carson, Koger, Lego, & Young, 2010; Grimmer, 2013). As expected, lawmakers from all types of districts tweeted about party effectiveness than they tweeted about policy. At the same time, the safest lawmakers mentioned policy and blamed Democrats more often, whereas the most vulnerable lawmakers mentioned the party’s efforts to resolve the shutdown more often than other lawmakers.

Figure 4.

District competitiveness and Twitter messaging emphasis.

Multivariate Analysis of Shutdown Communications

We next test three multivariate models for each chamber.⁹ We expect leaders to lead in terms of emphasizing effectiveness over policy. We also expect members representing competitive constituencies to be more likely to emphasize effectiveness over policy, and ideologically conservative members to be more likely than other members to mention policy. Finally, we expect these differences to be less pronounced among minority Republicans in the Senate in part not only because they were not to blame for the shutdown but also because a large proportion of them were not up for reelection in 2014.

Our list of party Leaders includes the House Speaker, Majority Leader, Majority Whip, and committee and subcommittee chairs. In the Senate, it includes the Minority leader, Minority Whip, and committee ranking members. We measure ideological extremism using a dichotomous Tea Party variable¹⁰ and first dimension DW-NOM INATEscores (where higher scores indicate greater conservatism). For constituency competitiveness, we include a dichotomous variable indicating whether a district or state is in the most competitive quantile (at least 45% share for Obama).¹¹ Finally, we test whether members who won by larger margins in their most recent General election or Primary election are more likely to mention policy. We include these latter variables as controls and do not have clear expectations regarding their effects (we consider constituency differences separately).¹²

The first House and Senate models (Figure 5) are Beta regressions (Ferrari & Cribari-Neto, 2004) predicting relative attention to policy versus effectiveness (with 1 indicating only policy mentions and 0 indicating only effectiveness mentions).¹³ The other models are ordinary least squares (OLS) regressions predicting the percentage of lawmakers’ tweets that are about the shutdown and that blame Democrats for it (Figure 6).

Figure 5.

Policy versus effectiveness emphasis in tweets.

Figure 6.

Tweets about the shutdown and blaming the democrats.

The results in Figure 5 are supportive. First, the patterns are more pronounced in the chamber responsible for the shutdown. House Party leaders were significantly more likely to emphasize effectiveness, as were members representing competitive constituencies. Translating the coefficients into estimated average effects, House Party leaders mentioned effectiveness 14% more than members with no leadership responsibility. House members from highly competitive districts mentioned effectiveness about 18% more than those from other districts. In contrast, Tea Party Republicans mentioned policy 10% more often than other members. In the Senate, the only significant difference is between senators representing the most competitive districts and everyone else. As discussed earlier, senators as a whole were much more likely to tweet about policy than House members.

The patterns for how often a member of Congress tweeted about the shutdown or blamed Democrats for it (Figure 6) are similar but the effects are not significant at the 95% confidence level. Vulnerable House Republicans were significantly less likely to publicly blame Democrats, whereas Tea Party Republicans were more likely to blame them.

Discussion

This article contributes to the literature on party branding by addressing a long-standing, and increasingly relevant, question of how party leaders balance the often competing electoral (and branding) needs of the party base and party moderates. The existing party branding literature argues that leaders keep issues that divide the party off the agenda (Cox & McCubbins, 2005; Sellers, 2010). However, this is not always the possible. How do leaders promote brands that are supported by and benefit most party members in such situations?

In this article, we put forward one answer to this question: Congressional leaders can promote a valence messaging frame to attract broader support among party members. We illustrate this argument by investigating an important branding event—the 2013 government shutdown. Party leaders did not have the option of exercising negative agenda control to keep the question of funding the government off the agenda. If they scheduled a vote on a clean funding bill, most in their party would oppose it, and the party base would be visibly “rolled” on the floor (Cox & McCubbins, 2005). On the contrary, preventing a vote meant shutting down the government, a move that would be very unpopular with most Americans. Either action was going to hurt the brand for some members of their coalition. Their response was to adopt a valence-branding strategy that diverted public attention away from the party’s responsibility for the shutdown.

We introduce and demonstrate the value of Twitter as a medium for studying party messaging and political communications. Republican members sent more than 10,000 tweets during the budget crisis and most were about the shutdown (67%). Tea Party Republicans were more likely to bring up the policy issues and blame Democrats for the shutdown, whereas Republicans representing competitive districts were more likely to emphasize their party’s efforts to resolve it. More centrally, we find that the House leaders who instigated the shutdown did not focus on the policy positions that led to it. They took to Twitter (and probably other media sources) and emphasized the party’s effectiveness in working to end the shutdown. House Republicans as a whole (including members who favored the shutdown) supported their leaders by promoting this more inclusive effectiveness frame in their own communications. This valence-branding strategy was designed to mitigate the shutdown’s damage for the members most at risk. Although it is impossible to know whether this effort made a difference, we do know that in spite of the dire predictions based on the outcome of the 1995-1996 Republican-led shutdown, Republicans were able to hold onto their majority in 2014.

Footnotes

Appendix A

N-Fold Cross-Validation of SVM Algorithm for Four Classes.

Variable	Precision		Recall
Variable	M	95% CI	M	95% CI
Shutdown	78.40	[78.1, −78.6]	79.20	[79, −79.5]
Policy	90.50	[90.3, −90.8]	82.00	[81.7, −82.3]
Effectiveness	76.90	[76.5, −77.3]	79.30	[78.9, −79.6]
Blame	78.70	[78.1, −79.3]	62.90	[62.5, −63.3]

Note. SVM = Support Vector Machine; CI = confidence interval.

Appendix B

Comparing Mean Values and 95% CIs for the Most Competitive Quantile of Districts With Other Districts.

Variable	Very competitive	Others
Shutdown	66.8%	67.6%
(95% CI)	[55.7, −65.6]	[64.6, −69.6]
Policy	26.7%	40%
(95% CI)	[20.9, −32.6]	[37.0, −43.2]
Effectiveness	69.6%	67.1%
(95% CI)	[63.3, −75.9]	[64.6, −69.6]
Blame	15.9%	23.6%
(95% CI)	[12.0, −19.8]	[20.9, −26.2]

Note. CI = confidence interval.

Appendix C

Members of the Tea Party Caucus.

House	Senate
Bachmann (MN-6)	Palazzo (MS-4)	Blunt (MO)
Barton (TX-6)	Pearce (NM-2)	Cornyn (TX)
Bilirakis (FL-12)	Poe (TX-2)	Cruz (TX)
Black (TN-6)	Price (GA-6)	Enzi (WY)
Broun (GA-10)	Roe (TN-1)	Johnson (WI)
Carter (TX-31)	Ross (FL-15)	Lee (UT)
Cassidy (LA-6)	Royce (CA-39)	McCain (AZ)
Coble (NC-6)	Scalise (LA-1)	McConnell (KY)
Coffman (CO-6)	Schweikert (AZ-6)	Moran (KS)
Crenshaw (FL-4)	Sessions (TX-32)	Paul (KY)
Culberson (TX-7)	Smith (NE-3)	Risch (ID)
Duncan (SC-3)	Smith (TX-21)	Rubio (FL)
Farenthold (TX-27)	Stutzman (IN-3)	Scott (SC)
Fincher (TN-8)	Walberg (MI-7)	Sessions (AL)
Fleming (LA-4)	Westmoreland (GA-3)	Toomey (PA)
Franks (AZ-8)	Wilson (SC-2)
Gingrey (GA-11)
Gohmert (TX-1)
Hartzler (MO-4)
Huelskamp (KS-1)
Jenkins (KS-2)
King (IA-4)
Lamborn (CO-5)
Lummis (WY-0)
Marchant (TX-24)
McClintock (CA-4)
McKinley (WV-1)
Miller (CA-31)
Mulvaney (SC-5)
Neugebauer (TX-19)

Appendix D

Beta and OLS Regressions Predicting Twitter Messaging Emphasis.

	Dependent variable
	Effectiveness (%)—policy (%)		Shutdown (%)		Blame (%)
	β			OLS
	(1)	(2)	(3)	(4)	(5)	(6)
	House	Senate	House	Senate	House	Senate
Obama (>45.2%)	−1.014*** (0.196)	−1.209*** (0.405)	−0.059* (0.034)	−0.068 (0.063)	−0.101*** (0.032)	0.061 (0.053)
Tea party	0.458** (0.202)	0.297 (0.334)	0.052 (0.035)	0.036 (0.053)	0.056* (0.033)	0.115** (0.045)
General share	−1.861** (0.824)	−0.434 (1.700)	−0.097 (0.142)	0.111 (0.274)	0.008 (0.138)	−0.171 (0.228)
Primary share	1.005*** (0.369)	−0.205 (0.818)	0.041 (0.063)	0.095 (0.131)	0.024 (0.061)	0.002 (0.109)
DW-NOM	0.010 (0.269)	0.258 (0.476)	−0.019 (0.046)	0.040 (0.076)	−0.057 (0.045)	−0.116* (0.064)
Leader	−0.686*** (0.159)	−0.069 (0.358)	−0.033 (0.027)	0.049 (0.057)	0.027 (0.026)	0.034 (0.048)
Constant	−0.062 (0.581)	0.476 (1.070)	−0.289*** (0.100)	−0.495*** (0.173)	0.241** (0.097)	0.231 (0.144)
Observations	216	43	218	43	216	43
R ²	.103	.139	.036	.137	.077	.271
Adjusted R²			.009	−.007	.050	.150
Log likelihood	142.232	7.505

Note. OLS = ordinary least squares.

p < .1. **p < .05. ***p < .01.

Appendix E

The following figure illustrates the robustness of our findings across different model specifications. The three top subfigures illustrate this for the main model (beta regression) predicting effectiveness–policy branding in the House. Each of the subfigures shows the robustness of the coefficients for the key covariates Leader, District Competitiveness (Obama > 45%), and Tea Party membership. When we exclude from the analysis messages sent by secondary accounts (Check 1), or members who tweeted very little about the Shutdown (Check 2, 3, and 4), the direction, significance, and magnitude of the effect of the key covariates remain the same. In the robustness checks for the Senate model, we observe almost the same. In this case, we observe some covariates (Leader in Check 1 and Tea Party in Checks 2-4) to have on average a contrary effect but the confidence interval around those is always big and crosses zero.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was partially supported by the National Science Foundation under Grant No. 1243917. Andreu Casas gratefully acknowledges support from the “La Caixa” Fellowship Program.

Notes

Author Biographies

Andreu Casas is a political science PhD student at the University of Washington.

John Wilkerson is a professor of political science at the University of Washington.

References

Barbera

(2015). Birds of the same feather tweet together: Bayesian ideal point estimation using Twitter data. Political Analysis, 1-16.

Barbera

Rivero

(2015). Understanding the political representativeness of Twitter users. Social Science Computer Review, 33, 712-729.

Butler

D. M.

Powell

E. N.

(2014). Understanding the party brand: Experimental evidence on the role of valence. Journal of Politics, 76, 492-505.

Campbell

Converse

P. E.

Miller

W. E.

Stokes

D. E.

(1960). The American voter. New York: John Wiley & Sons, Inc.

Canes-Wrone

Brady

D. W.

Cogan

J. F.

(2002). Out of step, out of office: Electoral accountability and House members’ voting. American Political Science Review, 96, 127-140.

Carson

J. L.

Koger

Lego

M. J.

Young

(2010). The electoral costs of party loyalty in Congress. American Journal of Political Science, 54, 598-616.

Casas

Morar

(2015, September 3-6). Different channel, same strategy? Filling empirical gaps in Congress literature. Paper Presented at the Annual Meeting of the American Political Science Association, Chicago, IL.

Cribari-Neto

Zeileis

(2010). Beta Regression in R. Journal of Statistical Software, 34(2), 1-24.

Cox

G. W.

McCubbins

M. D.

(1993). Legislative Leviathan: Party government in the House. Berkeley: The University of California Press.

10.

Cox

G. W.

McCubbins

M. D.

(2005). Setting the agenda: Responsible party government in the U.S. House of Representatives. New York, NY: Cambridge University Press.

11.

Cummings

Wang

(2010). Who needs polls? Gauging public opinion from Twitter data. Unpublished Manuscript. Retrieved from: http://nlp.stanford.edu/courses/cs224n/2011/reports/nwang6-davidjc-harukioh.pdf

12.

Downs

(1957). An economic theory of democracy. New York, NY: Harper & Row.

13.

Dutton

De Pinto

Salvanto

Backus

(2013, September 29). Government shutdown threat unacceptable, poll says. CBS News. Retrieved from http://www.cbsnews.com/news/government-shutdown-threat-unacceptable-poll-says/

14.

Evans

C. L.

Oleszek

W. J.

(2001). Message politics and Senate procedure. In Campbell

C. C.

Rae

N. C.

(Eds.), The contentious Senate: Partisanship, ideology, and the myth of cool judgment. Lanham, MD: Rowman & Littlefield.

15.

Executive Office of the President of the United States. (2013). Economic activity during the government shutdown and debt limit brinksmanship. Council of Economic Advisers. Retrieved from https://www.whitehouse.gov/sites/default/files/docs/weekly_indicators_report_final_0.pdf

16.

Ferrari

Cribari-Neto

(2004). Beta regression for modelling rates and proportions. Journal of Applied Statistics, 31, 799-815.

17.

Gayo-Avello

(2012). I wanted to predict elections with Twitter and all i got was this Lousy Paper—A balanced survey on election prediction using Twitter data. Arxiv Preprint arXiv:1204.6441.

18.

Golbeck

Grimes

J. M.

Rogers

(2010). Twitter use by the U.S. Congress. Journal of the American Society for Information Science and Technology, 61, 1612-1621.

19.

Grimmer

(2013). Representational style in Congress: What legislators say and why it matters. Cambridge, UK: Cambridge University Press.

20.

Grimmer

King

(2011). General purpose computer-assisted clustering and conceptualization. Proceedings of the National Academy of Sciences of the United States of America, 108, 2643-2650.

21.

Groeling

(2010). When politicians attack: Party cohesion in the media. New York, NY: Cambridge University Press.

22.

Groseclose

(2001). A model of candidate location when one candidate has a valence advantage. American Journal of Political Science, 45, 862-886.

23.

Grynaviski

J. D.

(2010). Partisan bonds: Political reputations and legislative accountability. New York, NY: Cambridge University Press.

24.

Hemphill

Otterbacher

Shapiro

(2013). What’s Congress doing on Twitter? In Proceedings of the 2013 Conference on Computer Supported Cooperative Work (pp. 877-886). New York, NY: ACM.

25.

Jacobs

L. R.

Shapiro

R. Y.

(2000). Politicians don’t pander: Political manipulation and the loss of democratic responsiveness. Chicago, IL: University of Chicago Press.

26.

Jacobson

G. C.

(2012). The electoral origins of polarized politics: Evidence from the 2010 Cooperative Congressional Election Study. American Behavioral Scientist, 56, 1612-1630.

27.

Joachims

(2002). Learning to classify text using support vector machines. Boston, MA: Kluwer Academic.

28.

Keller

Lehmann

(2006). Brands and branding: Research findings and future priorities. Marketing Science, 25, 740-759.

29.

Krehbiel

(1998). Pivotal politics: A theory of U.S. lawmaking. Chicago, IL: The University of Chicago Press.

30.

Lupu

(2013). Party brands and partisanship: Theory with evidence from a survey experiment in Argentina. American Journal of Political Science, 57, 49-64.

31.

Memoli

M. A.

(2014, January 24). Boehner on “Tonight Show”: Shutdown was a “Predictable Disaster.” Los Angeles Times. Retrieved from http://articles.latimes.com/2014/jan/24/news/la-pn-john-boehner-tonight-show-20140123

32.

Neiheisel

J. R.

Niebler

(2013). The use of party brand labels in congressional election campaigns. Legislative Studies Quarterly, 38, 377-404.

33.

O’Connor

Balasubramanyan

Routledge

Smith

(2010). From tweets to polls: Linking text sentiment to public opinion time series. Proceedings of the International AAAI Conference on Weblogs and Social Media, 122-129.

34.

O’Keefe

(2014, January). Boehner appearing on Leno: GOP is to blame for shutdown. The Washington Post. Retrieved from: https://www.washingtonpost.com/news/post-politics/wp/2014/01/24/john-boehner-appears-on-the-tonight-show-with-jay-leno/?utm_term=.63030dcb426a

35.

O’Keefe

Helderman

(2013, September 28). On cusp of shutdown, house conservatives excited, say they are doing the right thing. The Washington Post.Retrieved from https://www.washingtonpost.com/politics/on-cusp-of-shutdown-house-conservatives-excited-say-they-are-doing-the-right-thing/2013/09/28/2a5ab618-285e-11e3-97e6-2e07cad1b77e_story.html?utm_term=.b8d6f044be79

36.

Oriella PR Network. (2012). The influence game: How news is sourced and managed today (Global Digital Journalism Study). Author.

37.

Schwartz

(2013, October 8). $10 million gift to help head start through shutdown. The New York Times.Retrieved from http://www.nytimes.com/2013/10/09/us/10-million-gift-to-help-head-start-through-shutdown.html

38.

Sellers

(2010). Cycles of spin: Strategic communication in the U.S. Congress. New York, NY: Cambridge University Press.

39.

Siddiqui

(2013, August 15). Adam Kinzinger: Defunding Obamacare will cost republicans house majority (VIDEO). The Huffington Post. Retrieved from http://www.huffingtonpost.com/2013/08/15/adam-kinzinger-obamacare_n_3762843.html

40.

Smithson

Jay

(2006). A better lemon squeezer? Maximum-likelihood regression with beta-distributed dependent variables. Psychological Methods, 11, 54-71.

41.

Snyder

J. M.

Ting

M. M.

(2002). An informational rationale for political parties. American Journal of Political Science, 46, 90-110.

42.

Stokes

D. E.

(1963). Spatial models of party competition. American Political Science Review, 57, 368-377.

43.

Stone

W. J.

Simas

E. N.

(2010). Candidate valence and ideological positions in U.S. House elections. American Journal of Political Science, 54, 371-388.

44.

Tumasjan

Sprenger

Sandner

Welpe

(2010). Predicting elections with Twitter: What 140 characters reveal about political sentiment. In Proceeding of the Fourth International AAAI Conference on Weblogs and Social Media (pp. 178-185). Menlo Park, CA: The AAAI Press.

45.

Weisman

(2013, August 22). Clock is ticking for recess, and for a deficit deal. The New York Times.Retrieved from http://www.nytimes.com/2013/08/23/us/politics/chances-of-a-deficit-deal-are-rapidly-fading.html

46.

Woon

Pope

J. C.

(2008). Made in Congress? Testing the electoral implications of party ideological brand names. Journal of Politics, 70, 823-836.

47.

Zwillich

(2013, July 25). Sen Richard Burr, R-NC calls conservative effort linking government shutdown fight to repealing Obamacare—the dumbest idea i’ve ever heard. SoundCloud.Retrieved from https://soundcloud.com/toddzwillich-1/burr1-071325-wav