In February 2024, a bipartisan group of senators released the text of a conservative-leaning border security bill. Republicans had demanded the legislation in exchange for military aid to Ukraine, and Democrats seemed willing to accept the deal. Yet, Republicans ultimately rejected it. Donald Trump voiced opposition, Minority Leader McConnell called it weak, and one House Republican said he refused to help Democrats in an election year. Facing obstruction, President Biden hit back. He blamed the opposition for inaction and promised, “the American people are going to know that the only reason the border is not secure is Donald Trump and his MAGA Republican friends” (Biden, Reference Biden2024). The president carried this combative posture into his State of the Union—traditionally, a marquee event for nationally oriented policy persuasion. Strategic disagreement and negative rhetoric are well understood in theories of Congress (Gilmour, Reference Gilmour1995; Lee, Reference Lee2016; Noble, Reference Noble2024). Yet Biden’s behavior runs counter to established theories of presidential leadership, which emphasize a nationally oriented and policy-focused public presentation (Kernell, Reference Kernell1997; Canes-Wrone, Reference Canes-Wrone2006). This perspective is so prevalent that current undergraduate textbooks state as fact: “The president often professes to be above partisanship to win bipartisan support in Congress” (Lowi et al., Reference Lowi, Ginsberg, Shepsle, Ansolabehere and Han2022, 324). But as this anecdote makes clear, presidents increasingly use partisan rhetoric—which is unlikely to appeal to the opposition or win bipartisan support in Congress. To what extent do presidents engage in negative partisan rhetoric, when are they most likely to deploy it, and what do they hope to achieve?
Presidents are long-term thinkers, motivated by their historical legacies (Howell and Moe, Reference Howell and Moe2020), which are built by enacting major policies. Yet presidents cannot enact legislation without congressional support.Footnote 1 Although that support is sometimes easier to come by (e.g., under unified government), these institutional factors are “largely beyond the president’s control, especially in the short run” (Edwards, Reference Edwards2003, 14). Presidents, then, “go public,” appealing to a broad, national audience to promote their policies and pressure Congress to act (Kernell, Reference Kernell1997). Lawmakers are responsive to constituents. By directly affecting public policy attitudes, presidents can indirectly influence congressional voting (Canes-Wrone, Reference Canes-Wrone2006). However, subsequent research has identified limits to this strategy (Edwards, Reference Edwards2003), especially as partisanship and polarization have increased (e.g.,Cameron, Reference Cameron2002; Bond, Fleisher and Wood, Reference Bond, Fleisher and Wood2003; Sinclair, Reference Sinclair2006; Cavari, Reference Cavari2017). Presidents have responded to these evolving conditions with new tactics. They travel more frequently, promoting their agendas in friendly locales (Cohen, Reference Cohen2009; Rottinghaus, Reference Rottinghaus2010; Heith, Reference Heith2013), and they appear in non-traditional, targeted media (Scacco and Coe, Reference Scacco and Coe2021; Pluta, Reference Pluta2023). Yet even these strategies are grounded in the traditional framework of persuasion and short-term policy success. Negative partisan appeals represent something different: they underlie both a shift in the intended audience for presidential appeals as well as a change in what presidents hope to achieve by speaking—one systematically related to the political environment in which they operate.
I argue that presidents speak as negative partisans when they anticipate low prospects for legislative success. In so doing, they eschew short-term, nationally oriented efforts to persuade the public and pressure sitting lawmakers. Instead, they mobilize co-partisan voters with negative appeals to change the makeup of the next Congress, improving future policymaking prospects. In short, presidents must achieve policy success to secure their legacy (Howell and Moe, Reference Howell and Moe2016; Reference Howell and Moe2020). However, presidents recognize that political factors—like divided government (Bond and Fleisher, Reference Bond and Fleisher1990), polarization (Sinclair, Reference Sinclair2006), and competition for congressional control (Lee, Reference Lee2016)—limit their ability to pass their agendas through broad-based, policy appeals (e.g., Heith, Reference Heith2013). Although presidents can do little to change these structural features in the short-term, they are not beyond a president’s longer-run influence. As the most salient party leader, presidents can affect perceptions of themselves and their parties (Jacobson, Reference Jacobson2019) and shape the future congressional environment. To do so, I argue they will use negative partisan rhetoric—referencing and attacking the opposition—rather than going public as policy-focused, non-partisan figures (Hinckley, Reference Hinckley1990; Coleman and Manna, Reference Coleman and Manna2007; Rhodes, Reference Rhodes2014). Although antithetical to lawmaking, this style of message politics accentuates party differences (Lee, Reference Lee2016), shifts blame (Hood, Reference Hood2010), and mobilizes voters (Abramowitz and Webster, Reference Abramowitz and Webster2016; Iyengar and Krupenkin, Reference Iyengar and Krupenkin2018). However, this rhetoric should be less prevalent when the legislative environment is more favorable. If presidents are likely to succeed, attacking the opposition is needlessly antagonistic, potentially repelling bipartisanship that could enhance the their legacies (Arnold, Reference Arnold1990; Wood, Reference Wood2009; Howell and Moe, Reference Howell and Moe2020). To support this theory, then, we should observe more presidential negative partisanship when the legislative environment is unfavorable: when congressional majorities are “insecure” (Lee, Reference Lee2016), when government is divided, and as elections approach. Behaviorally, this rhetoric should decrease co-partisan evaluations of the out-party.
To test these hypotheses, I measure how often, and how negatively, presidents evoke the out-party in public speeches. I collect a corpus of all public statements given by presidents between 1933 and 2024. First, I identify and count all opposition-party references. Then, I use a pre-trained BERT model to code the sentiment of each paragraph. Consistent with my theory, presidents reference the out-party more often during periods of increased congressional competition, divided government, and as elections approach. This rhetoric is also more negative during periods of competition and divided government. I provide further support with a quantitative case study of Democrats’ filibuster-proof Senate majority in 2009–2010. When Democrats unexpectedly lose their 60th seat and their legislative power declines, President Obama references Republicans more, and more negatively, consistent with the theory. Finally, I provide evidence of the behavioral micro-foundations with panel survey data from 2012 to 2017. More presidential negative partisanship decreases co-partisans’ approval of the opposition party.
These results contribute to our understanding of going public (Tulis, Reference Tulis1987; Kernell, Reference Kernell1997; Canes-Wrone, Reference Canes-Wrone2006) and message politics (Evans, Reference Evans2001; Lee, Reference Lee2016). My theory does not repudiate our traditional understanding of going public, which focuses on short-term policy persuasion. Rather, it highlights how environmental factors—namely, a difficult legislative environment—can prompt presidents to use a different rhetorical strategy. Here, negative partisanship is intended to mobilize supporters today to elect a more aligned Congress tomorrow, setting the president up for future legislative success. This research also provides an institutional logic for negative appeals as opposed to those grounded in personality (Milkis, Reference Milkis1993; Skinner, Reference Skinner2008), party coalitions (Jarvis, Reference Jarvis2004; Grossman and Hopkins, Reference Grossman and Hopkins2016), or antiquated norms. This work also has implications for presidential representation (Wood, Reference Wood2009; Kriner and Reeves, Reference Kriner and Reeves2015; Dearborn, Reference Dearborn2021), highlighting the goal-oriented nature of presidential self-presentation. Although this strategy may advantage presidents, it raises normative concerns. Given the current political environment, we should worry about the baleful effects of elite negative partisanship (Skytte, Reference Skytte2021; Reference Skytte2022; Bø ggild and Jensen, Reference Boggild and Jensen2025) promoted by the most salient American politician.
1. The offensive president
Like legislators, presidents are motivated by electoral and policy goals (Light, Reference Light1999). Unlike legislators, “fixated on the short term,” presidents take the long view (Howell and Moe, Reference Howell and Moe2020, 161). Their “overriding concern” about their historical legacy “drives them to seek durable policy solutions to pressing national problems” (Howell and Moe Reference Howell and Moe2020, 163, emphasis mine). Whether presidents succeed depends on lawmakers—whose goals, preferences, and constituencies differ. These differences may be small, and thus presidents are more likely to succeed legislatively, when their party controls Congress or ideological preferences overlap (Bond and Fleisher, Reference Bond and Fleisher1990). At other times, as when government is divided or ideological preferences are diffuse, presidents may struggle to pass their agendas. In either case, there is little a president can do to change these environmental factors in the short-term. Presidents must try other tactics to alter preferences and pass their agendas.
One tactic presidents employ is going public: appealing directly to Americans to raise the salience of issues, change public opinion, and implicitly threaten lawmakers’ electoral safety (Tulis, Reference Tulis1987; Kernell, Reference Kernell1997; Canes-Wrone, Reference Canes-Wrone2006). By changing constituents’ preferences, a president may pressure reluctant legislators to support his policies. Building on this theory, subsequent scholarship has focused on the issues presidents promote and the degree to which public attitudes change in response to presidential rhetoric (Cohen, Reference Cohen1995; Kernell, Reference Kernell1997; Canes-Wrone, Reference Canes-Wrone2006; Rottinghaus, Reference Rottinghaus2010). However, the consensus from this research is that going public rarely changes opinion (Edwards, Reference Edwards2003). Why, then, do presidents speak so often? I argue a resolution to this puzzle lies in reconsidering the underlying assumptions of the theory and presidents’ goals when speaking.
First, the implicit assumption of going public is that presidents target a broad, national audience (e.g, Kernell, Reference Kernell1997; Canes-Wrone, Reference Canes-Wrone2006).Footnote 2 Second, this strategy is focused on the short-term and is explicitly policy-motivated. Presidents are unlikely to durably change public opinion, but a well-timed shift in salience or attitudes may achieve major legislative change (Edwards, Reference Edwards2000; Cohen, Reference Cohen2009; Cavari, Reference Cavari2017). In support of these assumptions, literature on presidential self-presentation shows presidents generally avoid associating themselves with the parties in an effort to appeal to cross-party or disaffected voters (Hinckley Reference Hinckley1990; Coleman and Manna Reference Coleman and Manna2007; Rhodes Reference Rhodes2014; but see Azari Reference Azari2014). However, as the presidency has polarized (Cameron, Reference Cameron2002; Donovan et al., Reference Donovan, Kellstedt, Key and Lebo2020), out-partisans are especially resistant to presidential appeals (Cavari, Reference Cavari2017), and “For most members of Congress, following...the opinion of those constituents who regularly vote for them now means supporting a president of their own party and opposing a president of the other” (Sinclair Reference Sinclair2006, 242-3, emphasis original). Recognizing this change, some argue that presidents turn to a strategy of “going local.” Here, presidents travel to “‘naturally friendly’ localities...to build support for their policies” (Cohen, Reference Cohen2009, 31). Recognizing that their appeals are unlikely to persuade the opposition, presidents generate short-term policy success by solidifying support among their base. However, this strategy can only help presidents legislatively if their congressional coalition is large enough. What happens when the president’s party is in the congressional minority?
When the legislative environment is unfavorable, I argue presidents turn to a different strategy: negative partisanship, which electorally mobilizes co-partisans by attacking the opposition. Although presidents cannot change public opinion or the composition of Congress in the short run (Edwards, Reference Edwards2003), they can do so over the long-term (cf. Noble, Reference Noble2023). “For presidents to get things done, they have to be reelected, see that someone like them succeeds them, and give their co-partisans coattails to ride” (Lowande, Reference Lowande2024, 32)—and these motivations are especially salient when presidents lack political capital (Light, Reference Light1999). No actor is “as important as the president in defining the collective images of the parties” (Lee, Reference Lee2009, 77). Going public by avoiding party politics (or worse, praising the opposition) ratifies the existing power structure (Sundquist, Reference Sundquist1988) and is unlikely to persuade the opposition. Going local is also unlikely to help the president during divided government or when majorities are narrow. Thus, presidents use negative partisan rhetoric, blaming the opposition for gridlock (Hood, Reference Hood2010) and magnifying differences between the parties (Lee, Reference Lee2016). Doing so raises the stakes of the next election. Voters are increasingly mobilized by negative partisanship (Iyengar and Krupenkin, Reference Iyengar and Krupenkin2018), dislike of the opposition rather than love of their own party (Abramowitz and Webster, Reference Abramowitz and Webster2016), so these negative appeals can increase participation and vote share. Further, these messages are likely to be received by the intended audience. Co-partisan supporters tend to be more receptive to presidential messaging than opponents (Cavari, Reference Cavari2017) and the media is especially likely to transmit messages that feature the president and involve partisan conflict (Groeling, Reference Groeling2010). Unlike theories of going public, negative partisan presidents target their base. Unlike theories of going local, negative partisan presidents have electoral, rather than policy, aims. Facing legislative constraints, presidents try to change the balance of power in Congress to pass their agendas in the future.
If this strategy is effective, why don’t presidents always use it? When presidents are advantaged in Congress, this rhetorical posture is unnecessary and potentially damaging. Presidents are “held accountable...for embodying national values and national identities, pursuing the public interest, and addressing national problems” (Howell and Moe, Reference Howell and Moe2020, 163), which should motivate them to embody these principles when legislative success is probable. By acting above party and appealing across the aisle, presidents can promote a non-partisan image, playing into the public’s normative conceptions of the presidency (Wood, Reference Wood2009). They may also be able to solicit bipartisan support, which would legitimate and insulate their policy legacies (Azari Reference Azari2014; Westwood Reference Westwood2021; but see Case and Ommundsen Reference Case and Ommundsen2024). If a president is going to pass their agenda anyway, cross-pressured out-partisans may sign on. If presidents attack the opposition during these periods, they will come across as sore winners and make cross-party support unappealing.
To support my theory, presidents should behave as negative partisans (more, and more negative, out-party references) when they perceive lower prospects of legislative success and as electoral incentives increase. Below, I discuss three specific hypotheses:
Competition for congressional majorities
In the mid-twentieth century, Democrats dominated Congress. The party held huge majorities and felt secure in their power (Sinclair, Reference Sinclair2006). However, the party was not ideologically homogeneous. Southern Democrats served as a swing constituency, working with economically liberal Democrats and racially conservative Republicans. This situation, ironically, promoted cross-party cooperation. Congressional Republicans were willing to work with Democrats, believing bipartisanship was the only way to exercise policy influence (Lee, Reference Lee2016). Lower competition led to less congressional polarization, giving presidents opportunities to go public above party (Rhodes, Reference Rhodes2014).
In the 1970s and 1980s, Southern voters began electing conservative Republicans, resulting in a more sorted and polarized Congress (Sinclair, Reference Sinclair2006). Republican victories in the 1980 Senate elections and the 1994 House elections, led to “a politics of destruction, concerned less with legislation than with investigation and obstruction” (Hemmer, Reference Hemmer2022, 8). Renewed competition changed congressional incentives. Rather than work together, minority parties see the majority in sight and engage in messaging—withholding legislative support and drawing clear contrasts between the parties—to win back control (Lee, Reference Lee2016). “The 1994 elections destabilized the political environment” and forced Clinton and his successors to acknowledge Democrats were no longer the natural majority (Galvin, Reference Galvin2009, 255). Now, presidential leadership polarizes the parties (Lee, Reference Lee2009). These intra- and inter-branch conflicts are not without precedent however. Truman faced a similarly hostile Congress in 1940s when government was divided and majorities were narrower (Galvin, Reference Galvin2009; Lee, Reference Lee2016). Given that competitive congressional environments limit the prospects for lawmaking, I expect presidents to act more like negative partisans during these periods.
Divided government
Few factors affect presidents’ legislative prospects more than party control of Congress (Levinson and Pildes, Reference Levinson and Pildes2006). Co-partisans across branches share ideological and programmatic goals (Bond and Fleisher, Reference Bond and Fleisher1990), and presidential-party lawmakers have electoral incentives to ensure presidential success (Lebo and O’Geen, Reference Lebo and O’Geen2011). Out-partisans have symmetric incentives to block and damage the president, irrespective of their ideological preferences (Groseclose and McCarty, Reference Groseclose and McCarty2001; Lee, Reference Lee2009; Kriner and Schickler, Reference Kriner and Schickler2016; Christenson and Kriner, Reference Christenson and Kriner2017). Therefore, I expect presidents to act more like negative partisans when government is divided.
Electoral timing
As elections approach, presidents have less time to pass policy (Light, Reference Light1999). Out-partisans should be especially resistant to presidential policymaking, as they want to avoid giving opposite party presidents “a win” right before an election (cf. Huber, Hill and Lenz, Reference Huber, Hill and Lenz2012). In their role as party leaders, presidents must also turn attention toward promoting their record and helping their co-partisans secure reelection. Partisan affect (Huddy, Mason and Aarøe, Reference Huddy, Mason and Aarøe2015; Iyengar and Krupenkin, Reference Iyengar and Krupenkin2018), negativity (Fridkin and Kenney, Reference Fridkin and Kenney2019), and anger (Valentino et al., Reference Valentino, Brader, Groenendyk, Gregorowicz and Hutchings2011; Phoenix, Reference Phoenix2019; Webster, Reference Webster2020) are key to turning out one’s base. Therefore, presidents should act more like negative partisans when midterms or their own re-election approaches.
2. Identifying negative partisanship in presidential rhetoric
To test these hypotheses, I collect a corpus that includes the text of all presidential speeches delivered between March 4, 1933 (Franklin D. Roosevelt’s first day) and March 29, 2024 (near the end of Joe Biden’s first term) from the American Presidency Project (APP, Woolley and Peters Reference Woolley and PetersN.d.), a total of 27,663 speeches.Footnote 3 These data include many types of presidential speeches from major national addresses, to minor statements, political rallies, exchanges with the press, and more. If a document contains multiple speakers (e.g., a joint appearance) or stage directions, I make every effort to automatically remove that text. This corpus extends the literature’s focus on a small number of major televised addresses, which can bias our understanding of presidents’ agendas (Russell and Eissler, Reference Russell and Eissler2022).
To determine when presidents invoke the opposition, I focus on three types of references. First, I look for presidents’ explicit use of party labels (i.e., “democrat” or “republican”).Footnote 4 Next, I look for presidents’ references to their two most recent out-party predecessors by last name (e.g., “Obama,” “Trump”).Footnote 5 Finally, I look for references to the surnames of opposition leaders in the House and Senate. Whenever a Republican president references the Democratic Party, one of the two most recent Democratic presidents, or a Democratic congressional leader, the instance is coded as an out-party reference (and vice-versa for a Democratic president). To put these references on a meaningful scale, and to account for presidents’ differential speaking rates, I specify my dependent variable as the number of out-party references per 1,000 words.
In Table 1, I provide descriptive statistics illustrating presidents’ use of opposition references. In the second column, Institutional Variation, I indicate whether the president experienced both unified and divided government during their tenure. This measure proxies the degree to which the congressional majority is in play in presidents’ and lawmakers’ minds. Next, I present several types of out-party references per 1,000 words. For example, President Trump referenced Bill Clinton and Barack Obama 0.50 times per 1,000 words, Nancy Pelosi and Chuck Schumer 0.17 times, and the Democratic Party 1.05 times: a total of 1.71 out-party references per 1,000 words. For context, President Trump spoke about 2,500 words a day. Thus, for every day he held office, President Trump referenced Democrats over four times on average. Compare that with Johnson, who referenced Republicans less than a third as often when Congress was less competitive. Table 1 makes clear that presidents deploy out-party references at differential rates. Presidents are also more likely to talk about parties rather than people. However, Trump and Biden represent a deviation from this pattern—increasingly invoking their predecessors and congressional leaders. Overall, there is a positive correlation between institutional variation and the use of opposition references: 0.43. As anticipated, presidents in more competitive congressional environments use more opposition references.
Table 1. Presidential Out-Party References per 1,000 Words, 1933–2024
Note: Institutional Variation is an indicator for whether a president experienced both unified and divided government.
What is not clear is whether these references are partisan attacks or bipartisan entreaties. Are presidents appealing to the opposition for support or are they going negative? Answering this question is difficult. Measuring sentiment in political rhetoric can be fraught given the common use of negation, irony, and sarcasm as well as political valence and semantic polarization. A second challenge is that the quantity of interest is sentiment about the opposition, not sentiment of the speech overall. Politicians engaged in partisan messaging frequently make contrasting statements about their party and the opposition. Paragraph-level sentiment may misclassify negative opposition references as neutral, or even positive, when aggregating over a paragraph.
To address these challenges, I take a two step approach to code the positive-vs-negative sentiment of each speech paragraph. First, I use OpenAI’s GPT-4o-mini model to isolate the contextually relevant portions of each speech paragraph. Through iterative refinement, I developed a few-shot prompt to instruct the GPT model to extract only the contextually relevant portions of text that would allow me to better measure out-party sentiment without including in-party contrasts or unrelated asides (see Supplementary Material A.3).Footnote 6 For example, consider the following statement made by President Clinton:
“But the Republicans in Congress have proposed a budget that will undermine the dignity and independence of our senior citizens. Here’s how: Medicaid’s the way our country helps families pay for nursing homes, home care, or other long-term care for elderly or disabled persons. Some people would have you think that Medicaid just helps poor children. Well, it does do that, and that is very important. Almost one in four American children are poor enough to need help from Medicaid.”
The bold text (extracted by GPT) explicitly criticizes congressional Republicans. After this attack, Clinton describes the benefits of Medicaid. The valence of this aside is positive and disconnected from his specific criticism of Republicans. The GPT step attempts to extract only the relevant portion of the paragraph, which results in substantial improvements when calculating sentiment.Footnote 7 The sentiment score (described below) for the entire paragraph is 0.27 while the sentiment of the bold snippet is lower, 0.08, better reflecting Clinton’s attack.
To produce sentiment scores, I use a pre-trained BERT model, fine-tuned for sentiment classification (twitter-roberta-base-sentiment). Unlike dictionary-based sentiment methods (e.g., AFINN) or static embeddings (e.g., word2vec), BERT is sensitive to the context in which a token (e.g., word) appears. Unlike a dictionary (and like static embedding methods), the BERT model assigns each token a dense vector, where tokens more similar to one another have more similar vectors. If a dictionary did not contain the word “wonderful,” it would not contribute to a paragraph’s sentiment score. The advantage of an embedding model is that knows “wonderful” is similar to other words like “amazing” and “awesome,” which are positively valenced. It accounts for any valenced word without a pre-built dictionary. Unlike static embeddings, the BERT embeddings for tokens change depending on context. For example, a static embedding would represent “taxes” using the same vector, whether it was preceded by the word “high” (negative valence) or “low” (positive valence). A naive model may interpret the word “high” as more positive than “low.” BERT will represent the token “taxes” differently depending on which of these two adjectives precedes it (as well as other relevant words). These contextual relationships mean BERT better understands sentiment and can provide a more nuanced label. Ultimately, each text snippet is assigned a score from 0 (most negative) to 1 (most positive) based on a weighted average of its predicted negative, neutral, and positive scores.Footnote 8 To facilitate comparisons between speeches that do and do not reference the opposition, I use the same model to code the sentiment of all words in paragraphs that do not contain a partisan reference and only the extracted snippet in paragraphs that reference the opposition.
To highlight face validity, I present five out-party-referencing paragraphs and their scores in Table 2. I present the entire paragraph and bold the contextually relevant portion extracted by the GPT model. The advantages of this GPT procedure are clear. Although imperfect and stochastic, it tends to isolate relevant text, especially in more negatively scored paragraphs. For example, in the fourth paragraph, the GPT model extracts the criticism of the Democrats while removing other positive in-party references. The BERT sentiment model also performs well, scoring paragraphs appropriately given their valence.
Table 2. Sentiment of paragraph excerpts referencing the out-party.
Note: Excerpts from more positive and negative paragraphs referencing the out-party. Bold portions indicate the GPT-extracted portions on which sentiment is computed.
To assess performance quantitatively, I hand-coded a small set of randomly sampled out-party referencing paragraphs and compared them to the machine generated labels (see Supplementary Material A.4).Footnote 9 I structured my validation as choosing between three categories (positive, negative, and neutral) and achieved 0.66 accuracy. Given that a random guess is accurate 33% of the time and guessing the most prevalent category would be accurate 44% of the time, this approach yields a substantial improvement over baseline.
I visualize the sentiment of presidential speech by administration in Figure 1. Each row represents the distribution of sentiment across paragraphs with out-party referencing paragraphs in dark gray and all other paragraphs in light gray. As expected, paragraphs containing opposition references are more negative. Truman is, again, an outlier, but his negativity is rivaled by recent presidents—especially Biden and Trump. Presidents are more positive toward the opposition in the mid-twentieth century, when Democrats dominated Congress. This trend is not a product of presidents getting more negative broadly. Presidents are getting more negative toward the out-party even as their other rhetoric becomes more positive. These descriptives comport with my expectations regarding presidential negative partisanship, and I test my hypotheses more formally in the following sections.
Figure 1. Sentiment of presidential speech paragraphs.
3. Empirical strategy
To test the argument formally, I conduct a series of correlational analyses using ordinary least squares regression. To get additional leverage on this question, I present evidence from a quantitative case study focusing on changes in President Obama’s rhetoric when Democrats gained, and then lost, a filibuster-proof Senate majority in the 111th (2009–2010) Congress.
First, I focus on how often presidents invoke the out-party. To do so, I define my dependent variable as the number of out-party references per 1,000 words at the speech level. To test the congressional competition hypothesis, I directly follow Lee (Reference Lee2016, 20-21) in defining periods of Majority Competition as the 80th–84th Congresses (1947–1956) and the 97th Congress and beyond (1981–2024). During these periods, the two parties held narrow presidential vote and congressional seat majorities, and control of Congress frequently alternated. The same could not be said of FDR’s tenure in office or of the period between the 85th and 96th Congresses when majorities were much more secure. To test the divided government hypothesis, I create a variable, Divided Government, that is 1 any time government is not fully unified, and 0 otherwise. To test the electoral timing hypothesis, I code a Major Election period as 1 every day between Labor Day and Election Day of a midterm or presidential re-election year, and 0 otherwise. The coefficients on these variables should be positive if presidents reference the out-party as I expect.
Second, I investigate the correlation between opposition references, the aforementioned independent variables, and the sentiment of those references. Here, the dependent variable is a paragraph-level measure of sentiment as previously described where more positive (negative) values indicate more positive (negative) sentiment. I interact each of the independent variables with the total number of references per 1,000 words at the paragraph level and include all constitutive terms. To account for correlation across speeches, I cluster standard errors at the speech-level. Here, the marginal effect of an additional out-party reference, conditional on each independent variable, should be negative.
My models include a series of controls: the president’s approval rating in the most recent Gallup survey,Footnote 10 whether a major war was occurring,Footnote 11 whether it is one of the president’s first 100 days in office,Footnote 12 the president’s term, and month fixed effects to account for seasonality. Models that include the Majority Competition variable include the president’s party. Otherwise, I include president fixed effects, allowing me to examine within-presidency changes.
4. Results
In Table 3, I test my core hypotheses. Column 1 presents the most basic test of the argument. In the first row, the coefficient on Majority Competition, is positive and statistically significant. On average, presidents in competitive contexts reference the opposition about 0.17 times more per 1,000 words, or once per every 6,000 words. Similarly, presidents in divided government (as compared to fully unified government) make about 0.07 additional opposition references per 1,000 words. Finally, during election season, presidents deliver over half an additional reference per 1,000 words. Substantively, these effect sizes probably understate opposition attention. A single reference likely suggests that an entire paragraph is allocated to out-party discussion. Together, all three of these coefficients are consistent with the theory of presidential negative partisanship: presidents facing legislative constraints increasingly reference the opposition party.
Table 3. Presidential out-party references during congressional competition, divided government, elections.
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Coefficients are from ordinary least squares models where the dependent variable is the number of references to the presidential out-party per 1,000 words in a speech.
In column 2, I add president fixed effects. Here, we can see that the divided government and major elections coefficients continue to be positive and statistically significant within presidencies. For example, an individual president who experiences both unified and divided government is expected to invoke the opposition an additional 0.21 times per 1,000 words in divided government as compared to unified government. In column 3, I re-rerun the model in column 1, disaggregating the competitive periods and comparing to them to the non-competitive baseline (73rd–79th Congresses, 1933–1946; 85th–96th Congresses, 1957–1980). Here, the results hold independently in each competitive period, speaking to the institutional, rather than temporal, nature of the theory.
However, my theory is not only about frequency. To determine whether these references are bipartisan entreaties or partisan attacks, I regress the sentiment of each paragraph on the number of references interacted with the key independent variables. These models are presented in Supplementary Material Table B.1. In Figure 2, I plot the marginal effect of an additional out-party reference (per 1,000 words) as I vary each independent variable. In the left-most panel, I find that a presidents in less competitive environments deliver rhetoric that is 0.003 points more negative when making an additional opposition reference. Presidents who experience competition more than double the negativity associated with each reference. In the second panel, I conduct a similar exercise for divided versus unified government (with fixed effects). Opposition references are associated with more negativity at the paragraph-level, irrespective of institutional control, but references are associated with even more negativity during divided government. In neither case do presidents engage in outreach when the legislative going gets tough. Finally, references are associated with no more or less negativity during or outside of major election periods, in contrast to the major elections hypothesis. Ultimately, increasing negativity at elections comes from additional references, but they are no more negative on average. Substantive effect size is difficult to interpret, but a 0.003 change in sentiment is equivalent to an increase in positive sentiment resulting from a 3 point approval gain. In today’s polarized political environment with stable approval ratings, three points is realistic but large. These effect sizes are substantive but not massive.
Figure 2. Marginal effects of out-party references on the sentiment of presidential speech paragraphs.
To summarize: presidents evoke the out-party more often when legislating is difficult. Presidents reference the out-party more when congressional majorities are insecure, when government is divided, and as elections approach. Those references become increasingly negative in the first two contexts as well. In no instance is referencing the opposition party associated with more positive presidential rhetoric, cutting against the idea that opposition references are designed to secure legislative support.
4.1. A quantitative case study: Obama’s 2009 Senate super-majority
To this point, I have focused on slow-moving environmental variables to test the relationship between legislative constraints and presidential negative partisanship. Here, I provide additional evidence in favor of the theory through a case study analyzing how President Obama’s rhetoric dynamically responded to the unexpected loss of the Democrat’s filibuster-proof Senate super-majority in the 111th (2009–2010) Congress. Here, I briefly summarize the context (full background in Supplementary Material B.3). Following Obama’s 2008 election victory, Democrats retained control of both chambers of Congress, increasing their Senate margin to 58 seats. Over the course of the spring and summer, Democrats increased their margin to 60 seats (a filibuster proof super-majority) due to an unexpected party switch and a victory in a contested election on July 1, 2009. They used this power to make progress on the Affordable Care Act, but they lost their 60th seat in a surprising special election upset on January 19, 2010. This loss forced Democrats to modify their approach and pass the final bill through reconciliation.
My theory of presidential negative partisanship would predict that the loss of the 60th seat, and its attendant legislative influence, would prompt President Obama to use more negative partisan rhetoric after January 20, 2010, than during the 60-seat period. To test this hypothesis, I run similar models to those in the previous section focused on the 111th Congress. I create a trichotomous indicator for whether the date is before July 1, 2009 (the first day after Franken was declared the winner), after January 19, 2010 (the date Scott Brown won the special election), or in between, when Democrats effectively held a filibuster proof majority. To support my hypotheses, the president should reference Republicans more, and more negatively, outside of the brief filibuster-proof window between July 1, 2009 and January 19, 2010.
I provide evidence of these effects in Figure 3. In the left panel, I present the predicted number of out-party references per 1,000 words at three key periods: before the 60-seat super-majority, during, and after. Both before and during that period, President Obama referenced Republicans about 0.30 per 1,000 words. After losing the 60th seat, the president nearly tripled the number of references made. As shown on the right, these references were not polite entreaties for cooperation. The sentiment associated with each additional reference per 1,000 words was also about three times more negative. Consistent with the underlying argument, Obama was more likely to reference Republicans and did so in an increasingly negative manner after losing influence in Congress. These results are consistent with the idea of presidential negative partisanship and at odds with the idea of bipartisan consensus-building.
Figure 3. Effects of out-party references on the number and sentiment of Obama opposition references, 111th Congress.
4.2. Presidential negative partisanship is electorally motivated
In this section, I provide additional descriptive evidence consistent with the electoral (rather than legislative) motivations underlying presidential negative partisanship.Footnote 13 If presidents evoke the out-party to mobilize their base, we should see these references cluster in campaign-oriented rhetoric, which is narrowly targeted and electorally motivated. By contrast, if presidents evoke the opposition to exert legislative pressure, references should cluster in major national addresses, which are broadly targeted and policy-focused. To determine when presidents deploy negative partisanship, I use metadata from the American Presidency Project to categorize speeches into three groups: major addresses (e.g., State of the Union, major televised speeches), rallies, and other types (e.g., minor remarks, press interactions). In Figure 4, I bin speeches into buckets based on their number of references per 1,000 words: 0, 
$(0,1]$, 
$(1,2]$, and 2+. I then determine how many speeches within each category fall into each bin. The overwhelming majority (72%) of major address and other speech types contain no opposition references, whereas most rallies (81%) contain references. Nearly half of all rallies contain more than 2 references per 1,000 words as compared to 13% of major address and 7% of other speech types. This pattern is consistent with an electoral, rather than legislative, orientation of presidential negative partisanship. These differences are statistically significant, as documented in Supplementary Material Table C1.
Figure 4. Where presidential negative partisanship is most prominent.
The association between references and topics can also provide insight into the underlying motivations behind presidential negative partisanship. If this rhetoric is electorally motivated, references should be divorced from policy discussion. If presidents reference the opposition to gain legislative concessions, then references should be prominent in discussions of policy. To test these hypotheses, I fit a 100-topic LDA topic model to all paragraphs in my corpus. I hand-label each topic based on top-word clusters and representative paragraphs. Finally, I inductively categorize these 100 topics into 19 major categories.Footnote 14 In Figure 5, I plot the percentage of paragraphs referencing the opposition at least once within each category. A large plurality (28%) cluster in a single topic: “elections.” These paragraphs focus on mobilizing voters with rhetoric like: “a vote for the republicans in this election is a vote against your own interests,” and “vote the democratic ticket straight and the country will be in safe hands.” Other higher-reference categories include “functional” (e.g., quantitative explanations, oaths, toasts) and “government” (e.g., courts, media, separation of powers) topics, which, based on top-word clusters, are quite broad. The first specific policy topic, taxes, is ranked fourth and contains just 8% of all opposition-referencing paragraphs. These patterns are more consistent with an electoral, rather than legislative, motivation.
Figure 5. The topic distribution of presidential negative partisanship.
Issue ownership theory can also help us discriminate between these two potential motivations. If presidential negative partisanship is legislatively driven, presidents should deploy references on owned issues during unified government. Here, a president has more leverage to pressure the opposition on the issues where he is advantaged. If this behavior is electorally motivated, presidents should deploy these references on owned issues during divided government. There, a president is legislatively disadvantaged, but these issues should be more salient and mobilizing to his base. To test this expectation, I follow Egan (Reference Egan2013, Table 3.2) and assign each topic to its owning party (or to neither). In Table C2, I regress the number of opposition references per 1,000 words on the interaction between issue ownership and divided government. Consistent with the electoral story, presidents deploy more opposition references on owned issues during divided government (marginal effect: 0.11, [0.05, 0.17]). Together, these exploratory analyses provide additional support for the electoral motivations underlying presidential negative partisanship. They cut against the idea that presidents are going public in the traditional sense—with an intent to persuade for short-term legislative gain.
4.3. Presidents do not engage in positive partisanship
Definitionally, negative partisanship involves increasing out-party animus that is not offset by increasing love for one’s in-party (Abramowitz and Webster, Reference Abramowitz and Webster2016). An implication is that, if presidents are truly engaging in negative partisanship, their out-party appeals should not be offset by increasing, or more positive, own-party references. To test this expectation, I count in-party references following the same procedures used to measure out-party references. In Table C3, I find that the gap between out-party and own-party references is generally increasing alongside institutional constraints. That is, presidents facing institutional constraints reference the opposition much more than their own-party. These patterns are consistent with the behavioral underpinnings of presidential negative partisanship, where out-party affect drives voter behavior and own-party affect has little bearing.
5. Presidential negative appeals shape co-partisan attitudes
Presidents go public as negative partisans when they perceive lower odds of legislative success. But (how) do these appeals shape public attitudes? On the one hand, these appeals could serve as a pressure tactic. By attacking the opposition party, the president may persuade the out-party public to pressure their party’s lawmakers to support the president, consistent with theories of going public (Kernell, Reference Kernell1997). To support this hypothesis, we would need to see that increases in presidential negative partisanship were associated with decreases in own-party approval among out-partisans, given that lawmakers increasingly represent co-partisan constituencies (Jacobson, Reference Jacobson2015) and are increasingly responsive to their own party’s partisans (Sinclair, Reference Sinclair2006). By contrast, my theory expects that the president’s co-partisans, but not out-partisans, will decrease support for the opposition party. This decrease in support would be consistent with the literature on party cues (e.g., Arceneaux and Kolodny, Reference Arceneaux and Kolodny2009; Nicholson, Reference Nicholson2012; Noble, Reference Noble2024) as well as my mobilization-focused theory given that “it is partisans’ dislike and distrust of the opposing party that leads them to participate in political life” in the modern era (Iyengar and Krupenkin, Reference Iyengar and Krupenkin2018, 214).
To test these hypotheses, I leverage monthly survey data from The American Panel Survey (TAPS), fielded by the Weidenbaum Center on the Economy, Government, and Public Policy at Washington University in St. Louis.Footnote 15 For each month between January 2012 and November 2017,Footnote 16 respondents were asked “Do you approve or disapprove of the way the following are doing their jobs?” Democrats in Washington and Republicans in Washington are among the groups respondents were asked to evaluate on a four-point scale from “Strongly Approve” (4) to “Strongly Disapprove” (1). I used respondents’ entry wave party identification (Democrat or Republican, including leaners) to determine out-party and in-party approval.Footnote 17 Thus, for Democrats (Republicans), their monthly four-point rating of Republicans (Democrats) in Washington serves as their rating of the out-party. I exclude independents. The average out-party rating is 1.48 with a standard deviation of 0.67. The average own-party rating is 2.50 with a standard deviation of 0.81.
My independent variable is the average sentiment of all presidential speech-paragraphs that reference the out-party at least once in a given month. To facilitate interpretation, I reverse code this variable such that higher values indicate negative sentiment. My key expectation is that more presidential negative partisanship should correlate with lower out-party approval ratings among presidential co-partisans. As surveys are fielded early in a month, and to limit the potential for presidential rhetoric and respondent approval ratings to be driven by contemporaneous omitted factors, I lag this variable by a single month.Footnote 18 Thus, party evaluations in e.g., February 2012 are regressed on presidential negative partisanship from January 2012. As shown in Supplementary Material Figure D1, these two variables have a correlation of −0.25 for presidential co-partisans and are not colinear with time.
Given the panel nature of my data, I include respondent fixed effects, allowing me to assess within-subject change in approval as presidential negative partisanship varies. I also control for lagged sentiment of all other paragraphs and similar covariates to the models in Table 3.Footnote 19 I cluster standard errors at the respondent and month-year level. I model this relationship using ordinary least squares, regressing out-party and in-party approval on the interaction between presidential co-partisanship and lagged presidential negative partisanship as well as the constitutive terms.
I present marginal effects in Figure 6 (see Supplementary Material Table D1). On the left, presidential negative partisanship is associated with declining approval of the out-party among presidential co-partisans. This result is consistent with my theory of negative appeals mobilizing co-partisans against the opposition party. No other results are statistically significant. These appeals need not increase same-party approval given that Iyengar and Krupenkin Reference Iyengar and Krupenkin(2018) find that out-party, but not in-party, affect drives modern turnout and political participation. More importantly, these appeals do not shape out-partisan attitudes toward their own party, cutting against the pressure tactic hypothesis. Were these appeals to pressure out-party lawmakers, we would expect that pressure to be driven by the out-party public.
Figure 6. Marginal effect of presidential negative partisanship on out-party approval.
This result is consistent with the behavioral microfoundations of my theory, yet it comes with limitations. First, this survey was fielded almost entirely during the Obama administration, limiting generalizability. However, there is theoretical reason to expect this relationship would hold symmetrically (Nicholson, Reference Nicholson2012). Second, the panel nature of the data and lagged independent variable allow for the assessment of within-subject change over time. However, I do not claim these results are caused by presidential appeals directly. Scholars should consider experimental approaches to gauge this relationship causally.
Overall, this result provides behavioral evidence consistent with the observational patterns of presidential negative partisanship. Presidential negative partisanship is correlated with changes in co-partisan attitudes—an outcome that should help presidents electorally, even if not in the legislative arena. They do not, however, correlate with out-party attitudes, cutting against the idea that these appeals are designed to solicit legislative cooperation from the opposition.
6. Conclusion
In this paper, I show that presidents facing legislative constraints go public as negative partisans. Presidents want to secure their historical legacies, which are driven by their policy success. However, that success depends on congressional support—which is harder to come by when government is divided and Congress is polarized. Unlike traditional theories of going public, where presidents secure support by making national policy appeals, I argue legislative constraints prompt presidential negative partisanship. Presidents attack the opposition to mobilize co-partisans and change the future composition of Congress. I support this theory through a set of consistent findings from presidential speeches (1933–2024) and behavioral survey data.
This research expands our understanding of presidential leadership, highlighting how presidents’ legislative prospects shape their rhetorical strategies. Where presidents used to go public to change public policy attitudes, today, their rhetoric is geared toward rallying the base for electoral gain. This strategic logic helps resolve the puzzle of why presidents speak so often even if their rhetoric does not shape policy attitudes (Edwards, Reference Edwards2003). Further, I contribute to our understanding of message politics and institutional negative partisanship (Groseclose and McCarty, Reference Groseclose and McCarty2001; Lee, Reference Lee2016; Noble, Reference Noble2024), showing how these congressional trends also apply to the modern presidency. My findings should not be taken as a criticism of extant literature. Like theories of congressional behavior developed during the “Textbook Congress” era, traditional theories of going public also came to prominence during an unusual ebb in polarization and congressional competition.
This study has considered the macro-factors that contribute to presidential negative partisanship, but future work could consider more dynamic measures of the legislative process. Subsequent studies could match the topics of presidential speeches with bills moving through Congress or the level of disagreement in congressional speech. We might expect presidential negative partisanship to increase when the bills a president champions face long odds of success or when negativity in congressional rhetoric increases. Researchers could also adopt an experimental framework to causally identify the behavioral effects of presidential negative partisanship. Finally, future work should consider how these messages reach the president’s co-partisans to shape their attitudes. Scholars could examine media, in the form of newspaper coverage or nightly news transcripts, to determine whether, and how, presidential negative partisanship is transmitted.
Even as presidents promise unity, they seem unable to resist negative partisan appeals. I show that this rhetorical style is not a personal failing of our leaders, but rather, a strategic response to institutional context. This shift likely advantages the president’s party in the electoral arena, but it may come at the cost of solving policy problems as well as decrease the reputation of the office and officeholder. When presidents act like negative partisans, they surely reinforce their role as a party leader (Kriner and Reeves, Reference Kriner and Reeves2015; Jacobson, Reference Jacobson2019), contribute to a focus on politics over policy, and deepen mass polarization and political disaffection.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2025.10041. To obtain replication material for this article, please visit https://doi.org/10.7910/DVN/FPPDKN.
Acknowledgements
I thank Carlos Algara, Pamela Ban, Meena Bose, Seth Hill, Samuel Kernell, Frances Lee, David Miller, Rachel Porter, Amna Salam, and Nora Schwaller for helpful conversations and feedback on this project. I also thank participants of the American Political Science Association and Southern California Political Institutions and Political Economy Conferences.
Open access
