Assessing minority party influence on partisan issue attention in the US House of representatives,1989–2012

Majority party domination

The US House of Representatives is a rule-driven legislative arena. These institutional rules are the basis for Cox and McCubbins’ (2005, 2007) Cartel Theory, which hypothesizes the majority party can use House rules and strategic gatekeeping positions to move the party’s legislative priorities through the chamber. These agenda-setting powers allow the majority party to dominate all aspects of the legislative process in the House.

Perhaps the most important aspect of the majority party’s domination of the congressional agenda is the freedom to exercise negative agenda powers. Specifically, the majority party has no incentive to allow legislation proposed by the minority party to receive serious consideration, so only majority party legislative initiatives are considered within committees and on the floor. In addition to US congressional policy-making, the agenda-setting powers detailed by Cartel Theory have been tested across both US state and comparative legislative institutions (Cox, 2000; Cox and McCubbins, 2007; Cox et al., 2010; Clark, 2012; Döring, 1995, 2001; Neto et al., 2003). The majority party controls policy outputs and throughputs, but this does not exclude the possibility of minority party influence on the larger policy debate within a legislative institution. The nature of partisan legislative agendas provides clues as to how party influence plays out within policy-making institutions.

Budge and Farlie (1983) propose two potential outcomes regarding interparty influence on partisan agendas: divergent or convergent issue agendas. The identification of either divergent or convergent issue agendas is paramount, because divergent agendas preclude the ability of the minority party to influence the majority party. Divergent issue agendas result when the parties strategically focus on different policy issues. The main reason opposition parties choose to focus on different issues is the opposition’s freedom to focus on the most advantageous issues. On the other hand, the majority party is bound by both the election platform and the contemporaneous concerns of the public; the opposition is not bound by such mandates (Mortensen et al., 2011). There is evidence of similar issue agenda strategies in the UK Parliament where opposition parties attempt to differentiate themselves from the majority party (Budge and Farlie, 1983). Within the American context, Senate party leaders manipulate the congressional agenda to create and deepen party cleavages (Lee, 2009). ¹ While there is clearly some evidence for divergent issue agenda tactics by political parties, the empirical findings across a number of institutional contexts provide a solid foundation for the expectation of convergent partisan issue agendas.

Convergent issue agendas suggest a back and forth between the parties. Such dynamics imply influence from the minority party on the policy debate. The ability of either party to be the first mover on attention to an issue may fluctuate across time and policies, but the important aspect of convergent issue agendas is the active participation of both parties. For example, aggregate studies of European political parties indicate opposition parties can impact both policy debates and actual policy outcomes (Green-Pedersen and Mortensen, 2014; Seeberg, 2013). In these cases, the parties engage one another—rather than focus on different issues—because the opposition party is able to control the policy debate through the interaction of media coverage and/or public opinion. There is reason to predict similar dynamics in the contemporary US Congress, given individual legislators’ electoral fortunes are intrinsically tied to the party label (Cox and McCubbins, 2005, 2007; Lee, 2009). These shared risks often cause the parties to focus on similar issues in an effort to discredit the opposition. In a similar vein, American presidential campaigns display far more issue convergence than divergence due to the candidates’ focus on salient issues involving a broad voting audience (Damore, 2005; Holian, 2004; Simon, 2002; Sigelman and Buell, 2004).

Obviously, the issue agendas of opposing parties often converge on the same issues, and, in some cases, this convergence is the result of the minority party acting as the first mover on an issue. The presence of such dynamics in the House of Representatives is the focus of this study. As detailed above, the majority party dominates all legislative outputs and throughputs in the House. Therefore, the ability to uncover minority party influence in the House requires a move beyond standard definitions of legislative agendas to a broader conceptualization of partisan issue attention.

Issue attention and legislative agendas

Examinations of agenda-setting focus on the relatively narrow set of issues that garner attention from political institutions and decision makers. According to Kingdon (1984: 3), agendas are “the list of subjects or problems to which government officials, and people outside of government closely associated with those officials are paying some serious attention.” Within this broad conceptualization of agendas, scholars also delineate between two different levels of agendas. The first level is the larger, “systematic agenda” characterizing policies garnering at least some attention from government officials (Cobb and Elder, 1983). The second, smaller level—referred to here as the “policy agenda”—explicitly states those aspects of the systematic agenda receiving serious consideration within policy-making institutions (Baumgartner and Jones, 1993, 2009; Binder, 2003; Cohen, 2012; Kingdon, 1984). The difference between agenda levels is one of tangible policy-making activities. An agenda item may receive some consideration on the systematic agenda, but the same issue may never be involved in policy-making activities within a decision-making institution, like Congress or the executive branch—for example, congressional hearings, bills, roll calls, presidential signing statements, or executive orders. This difference is the basis for demarcating policy agendas and issue attention in the current study.

The conceptualization of issue attention used for this study is more indicative of Cobb and Elder’s systematic agenda than of the more constrained policy agenda. Substantive policy issues are best defined as, “general themes that transcend a particular incident, structuring the content of political discourse and policymaking over a relatively long period of time” (Sulkin, 2005: 45). Therefore, issues refer to general policy domains—such as health care or education—and issue attention refers to the amount of consideration particular issues receive from decision makers. For the purposes of this article, issue attention is measured using a particular kind of speech given on the floor of the House of Representatives. Speeches given on the floor of the House are a form of legislative activity, but there is a qualitative difference between speeches and actual policy outputs and throughputs. Despite this difference, floor debate is an “up-front activity by party leaders for the explicit purpose of influencing the tenor and substance of the policy debate in Congress and the policy dialogue with the president” (Bader, 1996: 11). Moreover, floor speeches are the key to uncovering any influence the minority party may have on the policy process, because this medium allows for the concurrent tracking of issue attention for both parties.

One-minute speeches

Legislative outputs and throughputs only capture the policy priorities of the majority party and are therefore not appropriate for this study’s conceptualization of partisan issue attention. ² Political speeches are one type of legislative activity all parties in an institution can take part in—thereby overcoming the one-sided problem of other outputs and throughputs. ³ Speeches are effective tools for analyzing issue attention in comparative legislative institutions (Green-Pedersen and Mortensen, 2009; John and Jennings, 2010), but this is an underutilized unit of analysis in the study of American politics.

One-minute speeches are one type of speech that captures the issue attention of the two major parties in the House. The debate period takes place at the beginning of the legislative day and allows any member—regardless of party affiliation or seniority—to address the floor. The parties began utilizing the one-minute speech debate period in the 1980s, but party orchestration of speeches took on a more formal role when Dick Gephardt and Newt Gingrich ascended to their respective leadership positions within the House of Representatives in 1989 (Evans, 2001). Since this time, the communication groups of both parties—that is, the “Democratic Message Group” and the “Republican Theme Team”—routinely signal the parties’ policy stances through one-minute speeches.

A series of interviews conducted with both Republican and Democratic staff members further validate the use of one-min speeches. According to several staff members, the one-min speech debate period is ideal because it gives members valuable time—a scarce resource on the floor—to speak about salient issues and show off to their constituents (Staff Interviews, 2014). While most Americans do not watch Cable-Satellite Public Affairs Network (C-SPAN) or read the Congressional Record, one-minute speeches are used by individual members and the party leadership for publicity—often highlighted in constituent newsletters or posted to YouTube. The party caucuses also have a firm hand in the content of speeches. One Republican staff member noted, “the conference sends us weekly emails asking members to highlight specific issues during the [debate period]” (Staff Interviews, 2014). The Democratic caucus has gone as far as to draft stacks of one-minute speeches for members to use (Staff Interviews, 2014). Rank-and-file legislators understand the inherent link between electoral success and the party brand. Factoring in the valuable opportunity to address the House floor, members need very little incentivizing to take part in party messaging strategies during the one-minute speech debate period. Taken as a whole, it is clear the one-minute speeches provide a viable measure of partisan issue attention in the House of Representatives.

Classifying speeches

A total of 43,880 one-minute speeches were delivered on the House floor between 1989 and 2012. ⁴ Given the size of the population, an automated text classification technique is employed to categorize the speeches by policy topic. Unlike other automated text classification techniques, ⁵ this project uses a predetermined coding scheme. Specifically, Baumgartner and Jones’ Policy Agendas Project (PAP) major-topic coding scheme is utilized. The PAP coding scheme is ideal because of its hierarchical nature. Moreover, direct comparisons can be made between the measure of partisan issue attention developed in this article and other measures of attention from the PAP and Comparative Agendas Project (CAP). One drawback of this approach is the speeches are only coded for policy content—not for the policy frames or valence attached to those policies. However, identifying speeches by policy topic allows researchers to examine the language used within specific policy debates, which can then uncover the different rhetorical strategies employed by the parties. ⁶

The use of a predetermined coding scheme requires a “supervised learning” approach to text classification. While supervised learning is more costly than fully automated “topic model” approaches, a supervised learning process also incorporates the validity of human-coded text documents. Supervised learning processes classify text by first “training” on a sample of human-coded text and then categorizing the remaining uncoded documents using various statistical algorithms. In this way, supervised learning is a mix of human and automated text classification.

In order to build the training data set, a human coder categorized a total of 11,333 one-minute speeches. ⁷ Each speech was assigned a single category according to the PAP coding scheme. ⁸ The goal of the supervised learning process is to maximize the level of accuracy between the human-coded sample and the statistical algorithm’s attempts to classify the documents. ⁹ Three “low-memory” algorithms were used to classify one-minute speeches: support vector machine, elastic-net regularized generalized linear models, and multinomial logistic regression. ¹⁰ The training process uses information from all three algorithms to code the text documents. After several iterations of the training process, the amount of systematic error was reduced and the total population of policy-relevant speeches was classified. ¹¹

A broad look at the data

Table 1 summarizes the percentage of yearly speeches across the 20 major-topic codes ¹² and compares one-minute speeches to other measures of attention and agendas from Baumgartner and Jones’ PAP and Adler and Wilkerson’s Congressional Bills Project. ¹³ One takeaway from Table 1 is both the majority and minority parties talk about the same issues in the same relative proportions. While the speeches are not coded for tone, it is clear the two parties never talk about policies in the same way—the conversation is always adversarial. According to several staffers on both sides of the aisle, the party caucuses try to answer their counterparts and give the public a different product than their competitors. The goal is to reference the same policy issues but engage in different policy frames on these issues (Staff Interviews, 2014). It is also important to note speakers never deviate from the party message. One staff member noted, “you never give a speech on the floor and go against the caucus” (Staff Interviews, 2014). The information in Table 1 and the interviews with congressional staff demonstrate a convergence between the parties in terms of which policy issues garner attention on the House floor, which is highly suggestive of interparty influence.

OPEN IN VIEWER

Table 1.

Average percentage of issue attention captured across different political agendas, by policy types.

Policy	Majority attention	Minority attention	All congressional bills	Congressional bills reported out of committee	Congressional hearings	Public laws	SOTU speeches	News coverage	Public opinion
Macroeconomics	27.30%	26.19%	4.07%	1.40%	3.19%	1.17%	16.23%	2.35%	30.89%
International affairs	18.10%	15.49%	2.87%	3.42%	10.03%	4.52%	17.79%	16.68%	4.46%
Health care	12.56%	14.26%	11.04%	5.10%	8.28%	5.63%	8.26%	5.71%	9.50%
Government operations	5.83%	8.54%	9.83%	17.04%	12.19%	31.30%	4.75%	6.49%	5.18%
Energy	5.55%	5.15%	3.72%	3.77%	3.66%	1.67%	2.53%	1.31%	1.85%
Crime and family issues	4.99%	5.96%	5.99%	5.58%	5.83%	5.30%	6.69%	9.45%	12.07%
Transportation	4.99%	5.96%	4.22%	5.84%	4.26%	3.64%	0.69%	2.22%	0.00%
Defense	3.74%	3.47%	6.16%	7.47%	8.74%	6.82%	11.64%	5.48%	8.89%
Welfare	3.45%	3.54%	2.59%	1.18%	1.95%	1.83%	6.22%	0.64%	6.19%
Education	3.15%	3.89%	4.25%	2.32%	2.70%	2.76%	8.26%	3.25%	4.00%
Civil rights and liberties	3.09%	2.48%	2.03%	1.55%	2.15%	0.95%	2.10%	1.99%	3.18%
Banking and domestic commerce	2.16%	2.02%	6.76%	6.67%	9.49%	5.52%	2.45%	28.50%	0.50%
Environmental	1.92%	1.63%	5.14%	6.16%	5.17%	2.82%	1.65%	2.22%	1.51%
Labor	1.73%	2.12%	3.71%	1.42%	2.70%	1.20%	3.77%	1.27%	0.36%
Trade	1.64%	1.45%	10.08%	2.15%	2.53%	2.04%	2.71%	3.64%	0.67%
Immigration	0.76%	0.92%	1.77%	1.04%	1.04%	1.21%	1.14%	0.48%	1.78%
Agriculture	0.76%	0.65%	2.46%	1.27%	2.89%	2.39%	0.33%	0.81%	0.00%
Science, space, and technology	0.57%	0.39%	2.05%	3.20%	4.40%	1.78%	1.41%	4.20%	0.29%
Public lands	0.41%	0.46%	9.20%	22.11%	7.01%	16.57%	0.49%	1.13%	0.00%
Housing	0.36%	0.37%	2.04%	2.28%	1.79%	0.86%	0.88%	2.16%	0.00%

Note: Most measures of attention in the table range from 1989 to 2012. However, news coverage ranges from 1989 to 2008, congressional hearings range from 1989 to 2010, and public laws range from 1989 to 2011.

Table 1 further indicates one-minute speeches capture a different dynamic of attention in Congress than other measures. Speeches are dramatically different from hearings and bills reported out of committee in terms of the policy issues covered by each measure. For example, government operations and public lands are two of the largest categories of congressional hearings, bills reported out of committee, and public laws; the same policy areas make up a small percentage of one-minute speeches. The differences suggest the parties focus on more salient issues on which frames and party divisions can be easily explained to the general public. On the other hand, speeches are quite similar to both State of the Union speeches and public opinion. Broad issues with clear party divisions—such as macroeconomics, foreign affairs, and defense—dominate these measures of attention, but important differences still remain. ¹⁴ As a whole, Table 1 highlights how partisan issue attention adds to our understanding of macroinstitutional agendas by capturing aspects of attention not found in similar conceptualizations.

Figure 1 displays monthly time series for each party’s level of issue attention across the most discussed policy topics in the one-min speech debate period. ¹⁵ The information in Figure 1 points to two things that suggest the possibility of party convergence and responsiveness in partisan issue attention. As in Table 1, it is clear the parties discuss the same policy issues in the same relative proportions. This is noteworthy because it suggests the parties are trying to control the policy debate across a select number of issues. Figure 1 also suggests an apparent relationship between the parties’ time series across time. None of the time series in Figure 1 show a sharp divergence between the parties. Instead, the parties’ attention to particular issues looks very similar across time. There are points in time when one party clearly attempts to shift focus to a particular issue and the opposition responds accordingly. For example, the minority party displays a huge spike in attention to energy policy in 2008 and the majority party shows a corresponding increase in attention. Likewise, the majority party sees a large increase in attention to health policy in 2009 with the minority quickly following suit. This evidence is indicative of an interplay between the parties, which implies the minority party can affect the policy debate in the House. A proper assessment of temporal causality and responsiveness is needed before any solid conclusions can be reached.

OPEN IN VIEWER

Figure 1.

Monthly count of one-minute speeches, by party and policy topic 1989–2012.

Error correction models

A series of error correction models (ECMs) are used to test for minority party influence on majority party issue attention. ECMs are ideal because this regression technique analyzes the level of responsiveness between time series. ¹⁶ Moreover, these effects are captured in a single coefficient—controlling for a number of other factors—making it possible to summarize results from 20 different equations in an effective manner. The analysis in this article uses raw monthly counts of one-minute speeches. ¹⁷ All variables used in the ECMs are aggregated on a monthly basis (T = 288), and separate ECMs are run for each policy topic in the data.

The ECMs used in this article follow a two-step process (Engle and Granger, 1987). The first step involves creating the error correction component by saving the residuals from bivariate ordinary least squares regression between the two variables of interest. ¹⁸ The error correction component—identified as S_t ,—represents the gap between two times series. S_t acts as an independent variable in the final ECM and is the basis for measuring responsiveness in the dependent variable.

The second step in the ECM is detailed in the following equation:

Δ M a j o r i t y S p e e c h e s t = α + S t − 1 + Δ M i n o r i t y S p e e c h e s t + K e y V o t e t + X B + ∊ t .

S_t−1 represents a single lag of the error correction component from step 1 and is used to test for minority party influence. After multiplying by negative 1, positive values of S_t−1 indicate gaps between the time series caused by the minority party placing more attention on a particular issue than the majority party. Therefore, a significant, positive effect signifies a contemporaneous increase to majority party attention following such a gap. The Key Vote variable measures whether or not a salient roll call vote was recorded in the contemporaneous month within the House on the respective policy issue. ¹⁹ Positive effects signal an increase in attention to issues on which there is a key vote in the contemporaneous month. XB represents a matrix of control variables included in the model. ²⁰ In addition to the dichotomous political and congressional activity variables, XB contains a control variable measuring changes to public opinion within each of the policy issues. ²¹

Finally, separate sets of equations are run switching the majority party and minority party variables in the equations—the dependent and independent variables are flipped for steps 1 and 2 of the ECMs. In doing so, the ability of the majority party to influence minority party issue attention is also assessed. Majority party influence is surely a reality, given the majority party’s control of other agenda-setting processes. However, both outcomes deserve analysis to determine if there is a back-and-forth between the parties or if one party dominates the other. The results of both sets of ECMs—those analyzing minority party influence and those analyzing majority party influence—are included to compare the magnitude and direction of each possible outcome.

Analysis

Figure 2 summarizes the coefficients and confidence intervals ²² for the first lag of the error correction component in the ECMs. The coefficients expressing both minority party influence and majority party influence are displayed.

OPEN IN VIEWER

Figure 2.

Regression coefficients of the error correction component, by party and policy issue 1989–2012.

Figure 2 demonstrates gaps between majority and minority party attention in the previous time period produce statistically significant increases to contemporaneous attention across all 20 policy issues. The effects for minority party influence are all in the expected direction—when minority party attention on a particular issue is above that of the majority party in the previous month, the majority party responds by increasing attention in the following month. It is worth noting minority party influence is significant across every policy issue in the data, regardless of saliency. Moreover, the coefficients across all 20 equations produce similar effects. In most cases, a gap of one speech in the previous time period produces a corresponding increase of a little less than one speech for the majority party in the contemporaneous time period. Even though the largest effect—housing and development—is nearly twice as large the smallest effect—immigration—the majority of effects fall within a range of 0.70–1.00. These significant, positive effects suggest the minority party can influence majority party attention by increasing attention to a specific policy issue.

The information from the monthly time series in Figure 1 provides clues about the substantive impacts of the coefficients in Figure 2. For example, macroeconomics is by far the most talked about issue during the one-minute speech debate period, but this is one of the smaller effects in Figure 2. On average, macroeconomics is mentioned 19 times per month in one-minute speeches, ²³ with a maximum of 194 speeches given on macroeconomics in a single month. Conversely, housing and development—the largest coefficient from Figure 2—is also the least talked about policy issue during the one-minute speech debate period, with a monthly average of less than one speech and a maximum of 15 speeches. Even though macroeconomics has a smaller coefficient than housing and development, a change in attention to macroeconomics has a much larger substantive effect. Given the relatively small amount of variation between the coefficients, the largest substantive effects are observed within the most salient policy issues. The results suggest the minority party can influence majority party issue attention on any policy issue, regardless of saliency. However, only high saliency issues produce a noticeable change in majority party attention.

Figure 2 also contains the coefficients for the error correction components analyzing majority party influence on minority party issue attention. Just as with the coefficients representing minority party influence, the majority party coefficients are all significant and positive. The sign and significance of these coefficients suggest the majority party also successfully influences the policy debate during the one-min speech debate period—when the majority party’s attention to an issue is above that of the minority party in the previous month, the minority party responds accordingly. Again, the coefficients for majority party influence are all similar in size. The coefficients indicate a gap of one speech in the previous time period produces a contemporaneous increase of slightly less than one minority party speech.

The only difference between the coefficients for the two parties is the magnitude across different policy issues. Most of the coefficients in Figure 2 are between 0.50 and 1.00. On some policy issues, the effect of majority party influence is larger; on others, minority party influence is larger. Regardless, the most important aspect of Figure 2 is significant, positive effects across all policy issues. These findings suggest a reciprocal relationship between the parties.

It should be no surprise the majority party influences minority party issue attention. However, the very presence of a give-and-take between the parties implies influence from the minority party. Figure 2 suggests both parties are able to affect one another, which means the minority party plays an important part in this particular aspect of congressional policy-making. This finding does not diminish the role of the majority party. Instead, it is clear both parties engage in a struggle over the control of the policy debate in Congress. Rather than one party dominating partisan issue attention—or the two parties focusing on separate sets of issues—the parties influence each other.

Finally, Figure 3 contains the coefficients for the Key Votes variable, ²⁴ which assesses whether or not the presence of salient roll call votes in the House affects the number of majority party one-minute speeches delivered on a particular policy issue. ²⁵ The expectation is months with at least one salient vote within an issue area will cause an increase to one-minute speeches given on the topic. ²⁶ The results in Figure 3 are mixed. Half of the policy topics covered in the data display statistically significant increases to majority party attention on specific issues when a salient floor vote is taken on the same issue. However, the other 10 policy issues display insignificant results or significant effects in the opposite direction. All but two policy issues display positive coefficients, despite not meeting the threshold for statistical significance.

OPEN IN VIEWER

Figure 3.

Regression coefficients for contemporaneous key votes, by policy issue, 1989–2012.

The substantive interpretation of the results in Figure 3 shed light on the impact of salient floor votes. For example, social welfare sees an average increase of nine one-minute speeches during the same month as a salient floor vote on the issue. ²⁷ In this case, the increase in attention is nontrivial, because social welfare garners relatively low amounts of attention during the one-minute speech debate period—an increase of nine speeches in a single month is substantial. ²⁸ The other substantive effects in Figure 3 are far less impressive. For the most part, the presence of a salient floor vote causes an increase of anywhere between one and five speeches, even on high salience issues—for example, macroeconomics or foreign affairs. Taken together, the results in Figure 3 demonstrate majority party issue attention is at least partially a function of the highly visible business of the House. The results also indicate salient floor votes are not the driving factor behind majority party issue attention. In some cases, salient floor votes cause a significant increase in majority party attention. However, this finding is not consistent across all 20 policy issues in the data; the effect of minority party influence is more consistent and powerful.