“I’m Not the President of Black America”: Rhetorical versus Policy Representation

Abstract

A key question in the study of minority representation is whether descriptive representatives provide superior substantive representation. Neglected in this literature is the distinction between two forms of substantive representation: rhetoric versus policy. We provide a systematic comparison of presidential minority representation along these two dimensions. Barack Obama was the first African American president, yet his substantive representation of African Americans has not been fully evaluated. Using speech and budget data, we find that relative to comparable presidents, Obama offered weaker rhetorical representation, but stronger policy representation, on race and poverty. While we cannot rule out non-racial explanations, Obama’s policy proposals are consistent with minority representation. His actions also suggest that descriptive representatives may provide relatively better policy representation but worse rhetorical representation, at least when the constituency is a numerical minority. We thus highlight an understudied tension between rhetoric and policy in theories of minority representation.

President Barack Obama’s election was heralded as a breakthrough that implied more equitable political representation for African Americans. Nearly 80% of African Americans surveyed at the time regarded Obama’s 2008 victory as a “dream come true” (Steinhauser 2008). How well was this expectation met? Did the first African American president act as a representative for African Americans?

There is reason to expect an affirmative answer. The concept of descriptive representation suggests that a disadvantaged constituency may be best represented by its own members. Normative theories of descriptive representation posit that representatives should resemble the people they represent, sharing similar characteristics, experiences, and perspectives (Mansbridge 1999; Williams 1998). These similarities are expected to result in improved substantive representation, defined as actions taken for the constituency (Pitkin 1967). The question of descriptive representation is particularly pressing when it comes to African Americans, as African American interests and preferences are far from equitably represented in policy outcomes (Griffin and Newman 2008; Hajnal and Horowitz 2014; Schaffner, Rhodes, and La Raja 2016). In line with theories of descriptive representation, African American officeholders often do attempt to change policy in the direction preferred by African Americans, at least under some conditions (Browning, Marshall, and Tabb 1984; Griffin and Keane 2006; Griffin and Newman 2008; Haynie 2001; Minta 2011; Preuhs 2006; Shah and Marschall 2012; Tate 2003; Whitby 2000). If these theories are correct, then African Americans should have seen an increase in presidential representation during the Obama administration.

On the other hand, many empirical studies in the literature on political incorporation in local or legislative representation point to considerable limitations constraining minority officeholders (Hajnal 2006; Haynie 2001; Hopkins and McCabe 2012; Pelissero, Holian, and Tomaka 2000). African Americans governing majority-white electorates have strong incentives to represent that majority, and are electorally accountable to it. These representatives may face pressures to avoid focusing on race—to deracialize their governance (Clayton 2010; Gillespie 2012; Nelson, Sanbonmatsu, and McClerking 2007). In trying to avoid the perception of racial favoritism, these representatives may end up representing minorities no better (or perhaps even worse) than a comparable white representative (Hajnal 2006; Harris 2012). As Obama put it, “I’m not the president of black America. I’m the president of the United States of America” (Dingle 2012).

However, this literature has not considered the possibility that in majority-white constituencies, African American representatives may provide worse rhetorical representation, but nonetheless offer better policy representation, than their white counterparts. We develop this conceptual distinction using the case of Barack Obama. While Obama’s rhetoric on race has been analyzed (Gillion 2016), there has been no systematic assessment of his policy actions on race and race-related issues. More generally, no study has systematically compared presidents’ policy action on issues of particular concern to African Americans or disadvantaged minorities (for partial exceptions, see: Gillion 2016; Hajnal and Horowitz 2014; Nteta, Rhodes, and Tarsi 2016). Most importantly for the concept of substantive representation, there has been no systematic effort to compare rhetorical versus policy representation of an identity group.

We attempt such an analysis using a specific set of metrics. First, unlike existing studies, we operationalize representation along two distinct dimensions: words and deeds. Specifically, we analyze speeches separately from the concrete amounts presidents propose for relevant programs. Presidential budget proposals are one of the ways that presidency scholars measure presidents’ efforts to implement their policy priorities (Berman 1979; Berry, Burden, and Howell 2010; Canes-Wrone 2001, 2006; Clarke 1998; Dearden and Husted 1990; Howell, Jackman, and Rogowski 2013; Kamlet and Mowery 1987; Kiewiet and Krehbiel 2000; Kiewiet and McCubbins 1988, 1991; Krause and Cook 2015; Rossiter 1960; Whittington and Carpenter 2003). We find that representation cannot be accurately evaluated unless speeches and policy proposals are conceptually and empirically disentangled. Second, the baseline against which scholars have evaluated the quality of Obama’s representation of African Americans is unclear (Rudalevige 2013, 1130). We offer a specific baseline: other presidents, including those governing under similar circumstances. Third, building on and extending the urban political incorporation literature, we systematically classify programs as serving the needs of a constituency based on the proportion of recipients they serve (Browning, Marshall, and Tabb 1984). We conducted a comprehensive assessment of all federal programs and agencies and collected information on the proportion of their benefits that go to lower-income or African American people. We are unaware of studies of federal representation that have used a precise measure to select programs based on their actual benefit to the descriptive group to generate a systematic list of the relevant programs. This quantifiable standard thus offers an additional advance. These constitute the central contributions of this article.

In keeping with previous studies of Obama’s rhetoric, we find that Obama gave no more attention to race in his major speeches than his predecessors (Gillion 2016). In addition, he gave substantially less attention than other presidents to poverty, which disproportionately affects African Americans. However, by the yardstick of presidential spending proposals, Obama exceeds every president since Nixon. These findings hold when controlling for economic and political conditions. They are also supported by a robustness check using presidential DW-Nominate scores related to poverty and civil rights. Obama provided weaker rhetorical representation but stronger policy action on behalf of African Americans.

These findings highlight a tension between rhetorical and policy representation, casting descriptive representation as a trade-off between the two. This conclusion is a departure from the theoretical and empirical literature on descriptive representation, which has not investigated when and why major speeches and policy proposals may diverge. Still, the findings are consistent with theories of minority incorporation: in order to enact policies that benefit a disadvantaged group, elected officials in majority-white jurisdictions may need to frame them as a universal benefit, to avoid a backlash (Browning, Marshall, and Tabb 1984; Nteta, Rhodes, and Tarsi 2016; Wilson 1987). The potential for backlash may be even stronger when descriptive representatives advocate for their own group, as their efforts are likely to be perceived as favoritism (Hajnal 2006; Nelson, Sanbonmatsu, and McClerking 2007; Tesler 2016). Thus, Obama may have eschewed racialized rhetoric to avoid activating the racial concerns of the majority-white electorate. Such considerations underscore the utility of disentangling rhetoric and policy efforts, revealing that descriptive representatives may provide representation that is weaker in one sense but stronger in another. Of course, Obama’s emphasis on policy proposals over rhetoric may be due to reasons other than the dilemmas of minority incorporation. It is impossible to definitively eliminate alternative explanations for Obama’s difference from other presidents, given the existence of only one African American president. Nevertheless, the Obama difference remains after accounting for the main non-racial explanations for presidential behavior. These findings suggest the need to sharpen a neglected theoretical contrast between rhetorical and policy representation in studies of representation more generally. This attention to rhetoric versus policy may provide a useful direction for future research concerned generally with executive behavior in majority-white jurisdictions.

Literature

Obama’s presidency has prompted numerous studies by political scientists (Crotty 2012; Liu 2010; Skocpol 2012). However, most of these studies do not examine Obama’s efforts to represent African Americans. Instead, they focus on Obama’s handling of the economic crisis, his use of presidential power, his struggles with partisan polarization, or his foreign policy efforts. His role as the nation’s first African American president has received surprisingly little attention given its historic nature.

The few existing studies of Obama’s descriptive representation have reached mostly negative conclusions. For example, The Obama Phenomena, a multi-author volume that includes contributions from more than twenty scholars, contains no overall positive assessments of Obama’s efforts on behalf of African Americans (Henry, Allen, and Chrisman 2011). Such works tend to characterize Obama as “constrained from promoting policies that would aid racial minorities” (Rudalevige 2013, 1129). Numerous scholars argue that Obama failed to lead a national conversation on race, abdicating his power as an agenda-setter for racial equality (Gillion 2016; Harris 2012; King and Smith 2011). They conclude that Obama’s silence on race and poverty reflected the low priority he placed on these groups (Gillion 2016; Harris 2012).

These negative verdicts treat rhetorical and policy representation as a package. They assume that if Obama was weak on rhetorical representation, his policy efforts on behalf of African Americans were also lacking. However, they do not systematically examine those policy efforts. In addition, the scholarship on Obama’s policy activities has not focused on policies relevant to his role as a descriptive representative. Thus, there is little evidence to support conclusions about the nature of Obama’s descriptive policy representation. More generally, there are no systematic studies of presidential policy representation of specific social groups.

Obama’s historic presidency presents the opportunity for a key test of minority representation. The literature on race and descriptive representation poses a simple question: do blacks represent blacks better than whites do (Whitby 2000)? More specifically, do descriptive representatives provide better substantive representation by making a greater effort to enact policy for their descriptive constituency? These important questions have yet to be applied to the most powerful office in the American political system. Answering them requires attending to the distinction between rhetorical and policy representation. To be sure, the analysis cannot rule out non-racial causes for Obama’s behavior. The best we can do is control for non-racial covariates and use placebo tests with other presidents. We hope our study is useful in doing so and in developing specific, quantitative, and systematic measures of representation that apply to every president in every year. In that sense, we are comparing presidents’ substantive representation of African Americans in a much more systematic way than has been done to date.

Substantive Representation: Saying versus Doing

There are at least two dimensions of substantive representation by which representatives can be evaluated: what they say and what they do. When a member of a disadvantaged social category gains office, one way they can raise the level of representation for their group is through their speech. As public awareness of an issue rises, elected officials may face increased pressure to address it (Gillion 2016; Hutchings, McClerking, and Charles 2004). A descriptive representative seeking to provide substantive representation may thus speak more often about the needs of their descriptive constituents.

Moreover, the bully pulpit seems to be a power that presidents in particular can use. A distinctive power of the presidency is access to the public. When presidents make public appeals on behalf of a social group or policy area, they set the agenda by raising public concern about these issues (Cohen 1997; Kernell 2006; but see Nteta, Rhodes, and Tarsi 2016). For instance, Cohen (1993) argues that presidential rhetoric on civil rights has been instrumental in establishing it as part of the public agenda. Presidents can thereby indirectly encourage Congress to enact policy that aligns with their own preferences. Given this, the frequency of public references to a particular issue can be taken as an indicator of the president’s issue priorities (Cohen 1997). It follows that a descriptive representative seeking to provide substantive representation would use the special rhetorical power of the presidency to raise issues of concern for their descriptive constituency. By this standard, Obama provided superior representation to African Americans if he talked about issues distinctively relevant to African Americans more often than his white predecessors did (Gillion 2016).

However, public utterances are not necessarily the most reliable indicator of efforts to provide substantive representation. If their goal is to enact policy, presidents should only speak out when the public supports their position or can plausibly be persuaded to do so (Canes-Wrone 2001). If the public is opposed to the president’s position, raising the salience of the issue could backfire by generating public resistance. When the public is unlikely to embrace the president’s preferences, it is more effective to use backdoor methods, such as private negotiation with members of Congress. In fact, presidents may strategically avoid public discourse to avoid mobilizing opposition (Covington 1987). Thus, the notion that public appeals are a reliable indicator of policy representation must be tempered with two caveats: First, presidential rhetoric may be more reflective of the chances of policy success from going public than of true policy priorities (Canes-Wrone 2001); second, public speech on a policy is not an accurate indicator of substantive policy representation.

In addition to the standard considerations faced by any president, a descriptive representative faces additional constraints. As Gillion (2016, 47) notes, “Herein lies the paradox of a black president . . . an unprecedented effort to be viewed as a president for all people.” A representative from a numerical minority may trigger significant opposition by advocating for their own group (Hajnal 2006; Nelson, Sanbonmatsu, and McClerking 2007). The less a descriptive president says, the more effective they could be in enacting policies that benefit minorities. As such, Obama’s silence on race and poverty cannot be regarded as sufficient evidence of a weak effort to provide substantive representation for African Americans.

Along with rhetoric, a second way that officials can provide substantive representation is by making policy proposals. This is a more accurate indicator of priorities, and arguably a more important form of substantive representation. First, policy proposals are a direct causal step in the presidential drive to enact policy, while speech is an indirect step (Canes-Wrone 2001). Second, while the intent behind rhetoric is unclear, policy proposals indicate a concrete commitment to act. While the clearest evidence for Obama’s success as a descriptive representative would be relatively greater attention to African Americans’ interests in both his rhetoric and policy proposals, policy proposals represent the most critical test.¹

Data and Methods

To test for substantive representation of African Americans, we first identified the distinctive preferences and interests of African Americans. African Americans and white Americans tend to express divergent preferences on civil rights and anti-poverty policy. A larger proportion of African Americans than whites believe that not enough progress has been made toward racial equality or civil rights (Hutchings 2009). African Americans are also much more likely to support anti-poverty efforts (Griffin and Newman 2008; Haynie 2001; Whitby 2000).

We also examined the policy literature to identify areas that tend to disproportionately impact African Americans. The same policy areas came up: poverty and civil rights. We take up poverty first. While Americans of all racial groups experience poverty, African Americans have historically experienced greater and more pervasive socioeconomic disadvantages (Harris and Lieberman 2013; Haynie 2001; Macartney, Bishaw, and Fontenot 2013). When Obama took office, nearly 26% of African Americans lived below the poverty line, compared with only 12% of white Americans. The poverty rate of African Americans has been approximately double that of whites since before 1970 (Gilens 1999). Accordingly, African Americans make up a disproportionately large number of beneficiaries of anti-poverty programs. In a recent survey, 31% of African Americans had received food stamps, nearly double the percentage of whites (Morin 2013). Anti-poverty programs have been essential in improving the life chances of African Americans. Thus, efforts to boost spending on anti-poverty programs are a form of substantive representation of African Americans.

The second policy area is civil rights (Haynie 2001). Agencies and programs intended to promote civil rights continue to handle a substantial number of complaints related specifically to race. For instance, in 2015, a plurality (35%) of complaints investigated by the Equal Employment Opportunity Commission were about racial discrimination.² As such, efforts to spend on these programs are an indicator of representation of African Americans.

Measuring Presidential Rhetoric

To quantify presidential rhetoric on these issues, we analyze the content of State of the Union speeches from 1965 to 2016.³ We start with 1965 because it marks the beginning of the modern era of federal anti-poverty and civil rights policy, including the Voting Rights Act and the War on Poverty. Presidents in office from 1965 on are thus all operating under a broadly similar set of policy conditions. While we could go back further, doing so would offer little additional insight into how Obama’s representation of African Americans compares to the relevant baseline—presidents operating in the same historical era. The State of the Union is particularly useful for our research question, because it is an institutionalized, routine activity of the presidency, rendering speeches in some sense comparable (Cohen 1997). This allows us to construct a time series of presidents’ rhetoric on poverty and race in their most public speeches. We developed a list of keywords and counted the number of times a president referenced them. Examples of poverty words are “hungry” and “poor.” Examples of race words are “African Americans” and “discrimination.” The full list is in section A of the online appendix.⁴

Each occurrence of a keyword was coded as being either neutral/positive or negative in tone. A word was coded as negative if it (1) called into question the utility of anti-poverty or civil rights programs; (2) called for the reduction or elimination of an anti-poverty or civil rights program; or (3) invoked negative stereotypes of the poor or African Americans. For instance, if a poverty word was used in the context of an exhortation for low-income Americans to work harder, it was coded as negative (refer to online appendix B and C for details and inter-coder reliability).

Presidential mentions of poverty and civil rights may be influenced by three confounding factors. First, Obama’s speeches could generally be less focused on social groups and domestic issues compared to other presidents, giving the false impression that he specifically neglects poverty and civil rights. To evaluate this possibility, we examine mentions of the middle class as a placebo test. If the alternative hypothesis is true, then Obama would also talk about the middle class less than other presidents. Given that there are a growing number of African Americans joining the middle class, rhetoric about this group may also have value for black citizens. However, because the middle class is primarily white, we treat it mostly as a placebo, with the appropriate caveat. Second, to account for the fact that some presidents are more talkative than others, we divide the number of keywords used by the total number of words in each speech. Finally, Obama’s rhetoric could be characteristic of any president facing similar economic and political conditions. We therefore control for those conditions in our analyses.

Measuring Presidential Policy Efforts

To examine each president’s policy representation of African Americans, we use their yearly federal budget proposals from 1970 to 2017.⁵ Specifically, we use their proposed spending totals for existing discretionary programs, and their proposed spending changes to discretionary programs, mandatory programs, and tax expenditures. These are the elements of the budget that presidents can directly attempt to alter (refer to online appendix D for details). Total non-discretionary spending estimates for entitlement programs are not included in the analysis. We begin with the 1970 budget proposal for two reasons. First, the contemporary American welfare state did not exist until the late 1960s. The programs that existed prior to that were decidedly different from those inherited by Obama. Second, prior to administrative reforms in the early 1970s, presidents did not include as much detail about the allocation of funds, making it difficult to isolate proposed spending for the specific programs of theoretical interest here (Tomkin 1998).

Following Canes-Wrone (2001), we note that budget proposals have important advantages over other measures of presidential policy preferences. The main alternative measures of presidential priorities are Congressional Quarterly’s rating of presidential positions on bills and DW-Nominate scores. These measures rely on publicly stated presidential positions on bills, making them less suitable than spending proposals for at least three reasons. First, the president must provide a budget every year, but they are not obligated to speak on every bill. In fact, some presidents speak on few bills, generating sparse data. Public positions on specific bills are likely less accurate for Obama, who took far fewer positions than previous presidents (refer to online appendix E) (Klein 2012). Additionally, the public positions that a president takes may not accurately reflect their policy priorities. The president can only take a formal public position after Congress has initiated a roll call vote. In some of the years in our dataset, the president had prohibitively limited opportunities to take a position on legislation related to poverty or civil rights. Finally, while standard DW-Nominate scores capture overall ideology, they are not necessarily an accurate indicator of presidential positions on minority-related policy.

Using budget proposals as our primary measure of policy effort circumvents each of these issues. Budget proposals correspond to the same types of spending categories across years, standardizing the measure of policy preference across presidents. Since our aim is to see how Obama’s representation of African Americans compares to his predecessors’, comparability of measures across time is a necessity. Additionally, whereas position statements do not reflect intensity of preference, proposed spending does. Dollars are a continuous scale and therefore vary more than simple endorsement or opposition. Finally, whereas position statements require public speech, the details of budget proposals are hammered out behind the scenes and receive relatively less publicity.

In addition, budget proposals avoid the problems associated with using other types of presidential action. Vetoes are less useful than budget proposals for measuring presidential efforts to affect poverty and civil rights. Obama only issued 12 vetoes. Of those, only one might be considered at all related to poverty and none were clearly related to civil rights (refer to online appendix E). Additionally, vetoes are subject to congressional action in that presidents cannot veto policies that never reach their desk. The president may want to enact or reject a particular policy but can only do so if Congress initiates. As a result, they may have relatively few opportunities to formally react to bills related to poverty or civil rights. Moreover, vetoes capture presidential preferences only in simple binary terms. They convey little information about how closely the bills that reach the president’s desk match presidential preferences.

As to executive orders, scholars point to considerable limitations regarding their scope and impact (Chiou and Rothenberg 2013; Deering and Maltzman 1999; Peterson 1990). While presidents can use executive orders to influence the bureaucratic management of existing programs, they cannot commit additional resources to meaningfully expand benefits without securing the necessary funding from Congress. Consistent with this, Obama issued only four executive orders that could be reasonably described as related to poverty or civil rights, and these were minor in scope (refer to online appendix E). This is on par with other presidents. In sum, executive actions related to the policy areas of theoretical interest here are too scarce and inconsistent for systematic comparison across presidents.

In contrast, while Congress is under no obligation to act on the president’s budget proposals, it is one of the only means by which the president can formally exert meaningful, positive influence in the areas of poverty relief and civil rights. An abundance of research suggests that the executive branch exerts disproportionate influence over the budgetary process (Berman 1979; Berry, Burden, and Howell 2010; Cameron and McCarty 2004; Canes-Wrone 2006; Clarke 1998; Dearden and Husted 1990; Howell, Jackman, and Rogowski 2013; Kamlet and Mowery 1987; Kiewiet and Krehbiel 2000; Kiewiet and McCubbins 1988, 1991; Krause and Cook 2015; Pack 1987; Rossiter 1960; Whittington and Carpenter 2003). The president and executive Office of Management and Budget have substantially more information than Congress about the funding required to sustain existing bureaucratic programs. This informational asymmetry means that the president’s budget carries considerable weight during congressional deliberations (Berman 1979; Kiewiet and McCubbins 1991). It also sends a clear message to Congress about the sort of appropriations bills it may pass before being subject to presidential override. The president’s veto power means that presidents can exert negative influence over the final budget, incentivizing Congress to heed his proposals (Dearden and Husted 1990; Kiewiet and McCubbins 1988; Pack 1987). Thus, while presidents are unlikely to have their budget perfectly enacted, they have every reason to assume their proposed spending for each program will carry weight in the final legislation and they do little to tailor them to congressional preferences (Kiewiet and Krehbiel 2000; Kiewiet and McCubbins 1988; Krause and Cook 2015). Supporting this, Krause and Cook (2015, 243-244) conclude that “modern presidents’ formal prerogative to propose budgets not only serves as a critical element of executive authority . . . it also shapes budgetary outcomes . . . net of external political and policy considerations.” They further note (262) that “presidential influence over congressional appropriation decisions is comparatively stronger than the impact of executive acquiescence reflected as a result of the budgetary process.” This complements Kiewiet and Krehbiel’s (2000) finding that presidential characteristics, such as partisanship, exert a sizable and stable impact on appropriations, more so than partisan control of Congress. Most importantly, it strengthens and extends Kiewiet and McCubbins’ (1988) finding that presidential budget requests for agency funding have a significant impact on congressional appropriations decisions. As a whole, this literature strongly supports the notion that presidential budget proposals indeed represent an important, concrete step in policy-making.

While we do not claim that the president’s budget proposals represent a unilateral form of policy-making, they nonetheless provide a relatively good measure of presidential efforts to enact policy across time and in specific domains. Unlike alternative measures, the president’s budget offers an annual, precise quantity as a signal of their preferences for specific policy areas. Most importantly for the purposes of this paper, budget proposals allow us to systematically compare presidential policy efforts related to poverty relief and civil rights.

Still, it is reasonable to expect that alternative measures of minority policy efforts—despite severe limitations in the frequency and comparability of observations—should mirror presidential budget proposals. To this end, as a robustness check, we examine presidential DW-Nominate scores in the same policy areas (refer to online appendix F).⁶

We define anti-poverty programs as those that disproportionately benefit, either by law or in practice, low-income Americans. At least one of the following criteria had to be met for a program to be coded as anti-poverty: (1) at least 30% of beneficiaries or recipients had to belong to the bottom income quintile; or (2) at least 50% of beneficiaries or recipients had to belong to the bottom two income quintiles; or (3) it reduced the poverty rate by upwards of 5%, as demonstrated by previous scholarly research. The first two requirements are similar to Iversen and Soskice’s (2006) poverty threshold. Data on benefits by income level was acquired from the Congressional Budget Office and other relevant agencies. Where raw data was not available, we relied on eligibility requirements and scholarly analyses. The 103 programs identified are consistent with programs listed in the literature on anti-poverty policy (Ben-Shalom, Moffitt, and Scholz 2012; Irving and Loveless 2015; Palumbo 2010; Spar and Falk 2015). Examples include WIC (Women, Infants, and Children), Head Start, Section 8 Housing, and Pell Grants (refer to online appendix G for a complete list of programs; refer to online appendix H for details on how housing programs were standardized across years).

To identify civil rights programs, we examined the legislative history of civil rights laws and programs intended to ensure greater equity for African Americans and other racial minority groups (Donohue and Heckman 1991; King and Smith 2011; Laney 2013). We classified a program as minority and civil rights if it met the following criteria: (1) it offered cash or non-cash benefits specifically to racial minority groups; or (2) it offered cash or non-cash benefits specifically to racial minority-serving institutions or organizations; or (3) it promoted racial diversity, including the integration of African Americans, in workplaces and communities; or (4) its primary purpose was to promote and enforce civil rights; and, to avoid double counting, (5) it did not meet the criteria for anti-poverty spending. We identified twenty agencies, offices, or programs that met these criteria (refer to online appendix I). These include the Equal Employment Opportunity Commission and civil rights offices in the departments of Education and Justice. African Americans have been the largest—and among the likeliest—groups to use these programs.

The main outcome measure is total proposed constant dollars. In the online appendix, we examine three additional spending measures: (1) proposed dollars per poor person (for poverty) or African American person (for civil rights); (2) proposed dollars per unemployed person (for poverty); and (3) proposed discretionary dollars as a percentage of the proposed discretionary domestic budget (refer to online appendix J for details). These measures each have different strengths and weaknesses. However, we regard total proposed dollars as the most direct measure of presidential efforts to either increase benefits or expand eligibility. The other two measures—proposed spending per relevant person and as a percent of the budget—are shaped by exogenous social, political, and economic factors that are outside the president’s control and thus may not cleanly measure efforts at substantive representation. For instance, Obama faced a recession that demanded increases in overall poverty spending but may have actually resulted in decreased spending per relevant person given the need to cover many more affected people. No measure is flawless, but there is reason to think that total proposed spending is somewhat less susceptible to these problems because it does not build on exogenous factors that may vary irregularly or non-linearly. Given this, we focus on total proposed spending while controlling on factors that may influence this measure.

Even if Obama proposed substantially greater spending, this is insufficient evidence of descriptive representation. An alternative explanation is that Obama’s spending proposals were driven by external pressures, incentives, or economic and political conditions, rather than a true preference to represent African American interests (Skocpol 2012). One possibility is that Obama’s greater spending proposals were part of an ongoing expansion of federal spending; that is, perhaps he simply kept pace with growth in GDP. Another is that Obama confronted a deep recession and thus faced strong demands to increase anti-poverty spending. A third alternative is that Obama proposed more anti-poverty spending prior to a precarious reelection year to rally his base. Fourth, Obama may have proposed more because he is a Democrat. Fifth, partisan pressures during years when Democrats controlled Congress may have led Obama to propose greater spending. A final possibility is that Democratic Congresses specifically affect Democratic presidents, and Obama proposed greater spending because he was a Democratic president partnered with a Democratic Congress. We account for these possibilities in the results section.⁷

Results for Rhetorical Representation

The results of the content analysis of State of the Union speeches from 1965 through 2016 are presented in figures 1 and 2. For ease of interpretation, we use the number of keywords per 10,000 words (the equivalent of approximately two speeches). The gray and black bars in figure 1 represent the average frequency of words about poverty in each presidential term. By almost all standards, Obama performed poorly as a rhetorical representative. Obama devoted approximately 10.6 per 10,000 words to poverty in each term, less than each prior presidential term. Obama’s average was significantly lower than both the average Democrat and Republican. Moreover, he mentioned poverty approximately half as often as each of his Democratic predecessors (refer to online appendix K for additional t-tests of each type of rhetoric).

Figure 1 Average poverty and middle class rhetoric by presidential term, 1965–2016

Figure 2 Average minority and civil rights rhetoric by presidential term, 1965–2016

One problem with this measure is that it may reflect negative mentions of the poor. If so, greater volume may not reflect better rhetorical representation of African American interests. Indeed, as shown by the black bars in figure 1, Ford, Reagan, and Clinton spoke about poverty often, but with a high percentage of negative mentions. Obama stands out as being one of only three presidents to never reference the poor in negative terms. Nevertheless, even setting aside negative references, the grey bars in figure 1 make clear that Obama still spoke about poverty less than others. His 10.6 per 10,000 word average for positive mentions is substantially less than the average positive poverty mentions of all his predecessors (10.6 vs. 16.8, p < .01), his Democratic predecessors (10.6 vs. 19.8, p < .05), and his Republican predecessors (10.6 vs. 15.1, p < .10).^8,9

Figure 2 shows that Obama also did not provide strong rhetorical representation on race and civil rights. In his first term Obama averaged 1.48 race mentions per 10,000 words, the lowest rate of any president’s term. In his second term Obama increased his average to 5.72 words per 10,000, which is similar to other Democratic presidents. His overall rhetorical representation of race and civil rights (an average of 3.60 words per 10,000) was similar to his predecessors. While he spoke on these issues less than Clinton, Carter, or Johnson, these differences do not reach statistical significance (online appendix K). And compared to his Republican predecessors, he mentioned civil rights at about the same rate. In sum, Obama’s rhetoric on race was not much worse, but also no better, than other presidents.¹⁰

One alternative explanation is that Obama’s speech differed from other presidents in its dearth of mentions of any social group, and was therefore not particularly silent on the concerns of African Americans. For instance, Obama may have focused less on race and poverty because he focused more on defense or international issues. In that case, Obama would have also talked about the middle class less than previous presidents. The per-term percentage of words about the middle class are indicated by the white bars in figure 1. Obama’s overall average was much higher than his predecessors’, approximately three times more than the average Democrat and the average Republican (refer to online appendix K for significance tests). Obama steered away specifically from the rhetorical representation of African Americans.

Another alternative is that Obama’s rhetoric was typical for presidents facing similar economic and political conditions. To account for this alternative explanation, we estimate OLS regressions of the number of keywords per 10,000, controlling on relevant conditions. Rhetoric is highly variable from year to year. For that reason, we use two versions of the dependent variable: the more volatile yearly number, and the smoother per-term average. Although using the term as the unit of analysis costs statistical power, it allows for more reliable estimation of the effect of theoretical interest: the difference that a descriptive representative makes. We present the per-term results in table 1, and the yearly results as a robustness check in online appendix L. The results are substantively similar.

Table 1 Regressions for per-term poverty, civil rights, and middle class rhetoric

* p < .05, p-values are derived from two-tailed tests

Robust clustered standard errors in parentheses

Note: Each type of rhetoric is measured as words per 10,000 in State of the Union addresses. The unit of analysis is the yearly average for each presidential term from 1965 to 2016. Robust standard errors are clustered by president. “Obama” is a binary variable indicating whether the president in each term was Obama. “Percent Unemployed” is the average per term unemployment rate. “Real GDP Growth” is the average growth of real GDP in each term. “Percent Poor” is the average per term poverty rate. “Democratic President” indicates whether the president was a Democrat for each term. “Democratic Congress” indicates whether both the House and the Senate were controlled by the Democrats for at least half the term. “Real GDP” is the average GDP (in billions of 2016 dollars) for each term. “Real Deficit” is the average federal deficit (in billions of 2016 dollars) for each term. “Re-Election” is a binary indicator for whether the president was up for re-election at the end of the term.

The predictor of interest is a binary variable indicating whether or not Obama was in office. Economic controls consist of the term average for real GDP (in billions of 2016 dollars); growth in real GDP; the federal deficit (in billions of 2016 dollars); the poverty rate; and the unemployment rate. Political controls consist of an indicator for whether the president was up for re-election at the end of the term; an indicator for whether Democrats controlled both the House and Senate for at least half the years in a term; and an indicator for whether the president was a Democrat.¹¹ Economic controls reflect conditions in the year prior to the speech; political variables reflect conditions at the time of the speech. Standard errors are clustered by president.

Models with the full set of controls require more statistical power than the small dataset provides. Instead, table 1 presents three approaches: a bivariate regression with only the Obama indicator (models 1A, 2A, and 3A); a regression with only significant controls (models 1B, 2B, and 3B); and a regression with the controls that produce the highest adjusted R² (models 1C, 2C, and 3C).¹² As noted above, yearly speech models with a full set of controls generate similar conclusions about Obama’s rhetoric (refer to online appendix L).

Model 1A shows that without any controls, Obama mentioned poverty 10.4 fewer times per 10,000 words than his predecessors (p < .05). Adding controls in models 1B and 1C strengthens the negative Obama effect, suggesting that Obama about spoke about poverty 16 to 17 times less per 10,000 words (p < .001) than would be expected given relevant conditions. Model 2A shows that without controls, Obama mentioned civil rights at a similar rate as his predecessors. Models 2B and 2C show that accounting for relevant conditions, Obama mentioned civil rights two to three times less per 10,000 words, though this effect only reaches statistical significance in one of these two models. Finally, the placebo test in Model 3A shows that Obama spoke about the middle class 18.3 more times per 10,000 words than his predecessors (p < .001). When a control for GDP is included (model 3B), this effect decreases but remains significant. Model 3C includes controls for the best model fit, which raises the size of the Obama effect back to 18.1 (p < .01). These findings support the conclusion that Obama talked far less about poverty, less or the same about race, and far more about the middle class, than other presidents.

Overall, Obama’s rhetorical representation of African Americans was weaker than other presidents. There are caveats, to be sure: Obama avoided negative references to the poor and African Americans, and he increased mentions of race in his second term. However, on the whole, he spoke about race no more (and by some estimates, less) than his predecessors. He spoke about poverty, even in positive terms, far less than others. These results hold with economic and political controls. Finally, Obama’s frequent mentions of the middle class indicate that his low attention to poverty and race cannot be attributed to low attention to domestic policy or social groups.

Results for Policy Representation

Next, we turn to proposed spending as the second facet of representation. Figure 3 shows the average yearly proposed anti-poverty spending by president, for the primary measure (total dollars) and for three alternative measures: dollars per poor person, dollars per unemployed person, and dollars as a percent of the discretionary domestic budget. For total anti-poverty spending, Obama proposed an average of $172.9 billion per year. This roughly doubles each of his Democratic predecessors, and raises his immediate predecessor’s average proposal by 42%. With the exception of G.W. Bush’s spending per unemployed person, Obama exceeded each of his predecessors on every measure of proposed anti-poverty spending (for yearly anti-poverty and civil rights proposed spending, refer to online appendix M).

Figure 3 Yearly average proposed anti-poverty spending by president

Figure 4 shows analogous proposed spending for race and civil rights. Although the pattern is less clear, Obama did propose more than each predecessor on all three measures, the only exception being Carter’s spending per African American.

Figure 4 Yearly average proposed minority and civil rights spending by president

Alternative explanations for Obama’s large spending proposals derive from theories arguing that presidents respond to circumstances. To test for this, we estimate OLS regressions that include the yearly economic and political controls used in the speech analysis. These include real GDP (in billions of 2016 dollars); growth in real GDP; the federal deficit (in billions of 2016 dollars); the poverty rate; and the unemployment rate; an indicator for re-election years; an indicator for whether Democrats controlled both the House and Senate; and an indicator for whether the president was a Democrat. A dummy variable for Obama captures the effect of his presidency. Re-election is measured in the budget year. All other controls reflect conditions in the year the budget was proposed (one year prior to the budget year). Robust standard errors are clustered by president.^{13 14}

These OLS models account for gradual growth in the budget over time by controlling on GDP (it is .99 correlated with year; results are similar when replacing GDP with year). In our main analyses, we adopt this approach as an alternative to including a control for lagged spending for theoretical and statistical reasons. First, including a lagged dependent variable assumes that the budget is set incrementally such that proposals in the previous year determine proposals in subsequent years regardless of circumstances (Davis, Dempster, and Wildavsky 1966). However, this assumption is contested by budget scholars (Bailey and O’Connor 1975; Baumgartner and Jones 2009; LeLoup 1978; True 2000; Wanat 1974). It may be that the budget changes not in response to last year’s spending, but to relevant political, social, and economic factors, or epochs. To the extent that these factors change gradually from year to year, previous budgets will predict future budgets very well even though they have no true causal impact. Consistent with this is Baumgartner and Jones’ punctuated equilibrium theory: when relevant factors change in a dramatic way (as might be true with the election of the first African American president), so too will the budget. Second, lagged dependent variables are only appropriate when the data is stationary, meaning that it has a similar mean, variance, and auto-correlation over time. If the data is not stationary, then including a lagged dependent variable will bias the estimated effects of other variables (Keele and Kelly 2005). Augmented Dicky-Fuller Tests reveal that even with a control for real GDP (which, as noted earlier, is nearly perfectly correlated with year) the proposed spending data is non-stationary and therefore not well suited to models that include a lagged dependent variable. Stationarizing with first differences is problematic given our interest in Obama’s total proposed spending relative to his predecessor's, because the standard deviation of first differences necessarily reflects how each president distributed their overall change in spending across their years in office. Moreover, the first-differenced spending data has too much variation to allow for any meaningful analysis given our relatively few observations: regression models with first-differenced spending as the outcome variable have very poor fit (adjusted R² < .10) and none of the predictors are significant due to large standard errors.

Still, it is possible that excluding a lagged spending variable could introduce omitted variable bias such that the Obama effect is overstated in our main analyses. Therefore, after the main results provided later, we present additional analyses that rely on forecasting, in which omitted-variable bias is less problematic, to further evaluate Obama’s proposed spending relative to his predecessors’.

Table 2 presents OLS regression results from the main analyses of yearly anti-poverty and civil rights spending proposals. It includes four models: a bivariate regression with the Obama indicator (models 1A and 2A); a regression from which insignificant controls were successively removed (models 1B and 2B); a regression with the highest adjusted R² (models 1C and 2C); and a regression with a full set of controls (models 1D and 2C). Robustness checks with alternative spending measures are presented in online appendix N.¹⁵

Table 2 Regressions for proposed anti-poverty and Civil Rights spending

* p < .05, p-values are derived from two-tailed tests

Robust clustered standard errors in parentheses

Note: Spending proposals are measured in millions of 2016 dollars. The unit of analysis is the budget year from 1970 through 2017. Standard errors are clustered by president. “Obama” is a binary indicator for whether a budget was proposed by Obama. “Real GDP” is GDP in billions of 2016 dollars in the proposal year. “Re-Election Year” indicates if a president was up for re-election in the budget year. “Real GDP Growth” is the percent change in GDP in the proposal year. “Percent Poor” and “Percent Unemployed” are the poverty and unemployment rates in the proposal year. “Democratic Congress” indicates whether the Democrats controlled Congress in the proposal year. “Democratic President” indicates whether the president was a Democrat in the proposal year. “Real Deficit” is the federal deficit in billions of 2016 dollars in the proposal year.

Model 1A shows that Obama proposed significantly more than his predecessors to address poverty ($95 billion, p < .001). Model 1B shows that controlling for all statistically significant factors (GDP and re-election), the Obama effect decreases but remains substantial and statistically significant ($43 billion, p < .001). In Model 1C, with the best overall fit, the Obama effect also remains substantial and statistically significant ($47 billion, p < .001). Model 1D shows that with full controls, the Obama effect remains large and significant, exceeding the impact of unified Democratic control of government ($45 billion, p < .05). Obama proposed to fight poverty more vigorously than his economic and political circumstances would predict.

For civil rights spending, Model 2A shows that without controls, Obama proposed significantly more spending than his predecessors ($682 million, p < .001). Model 2B presents the results with significant controls, namely GDP. This causes the Obama effect to drop to approximately $210 million, but it remains statistically significant (p < .001). Model 2C includes full controls, and also produces the best model fit. Here, the Obama effect drops slightly, but remains significant and substantively large ($157 million, p < .05): the magnitude is similar to the effect of unified Democratic government.

These results are robust to a variety of alternative control variables. They hold when replacing GDP with year (with which it is highly collinear, ρ =.99) or with the number of African Americans (also highly collinear with GDP, ρ =.99); when using a binary variable for a recession rather than change in GDP to capture the state of the economy; when using the number instead of percentage of poor and unemployed; and when controlling for the conditions in the budget year rather than the proposal year.¹⁶

They are also robust to alternative model specifications. First, when controlling for GDP and one economic or political control at a time, the Obama effect retains its magnitude and significance in each model of anti-poverty spending and most models of civil rights spending (online appendix O). Second, the Obama dummy remains significant even when compared only to his Democratic predecessors; Obama proposed more than would be expected for a Democrat (online appendix P). Third, the relevant models in table 2 suggest that even after controlling for the partisan composition of Congress, Obama proposed more relative to his predecessors. However, we more thoroughly address the possibility that Obama’s proposals reflect political considerations rather than his policy priorities by showing that they were similar regardless of whether Congress was controlled by the Democrats or Republicans (online appendix Q). Finally, as an additional robustness check, we exclude Obama from the data and repeat the analyses with a dummy variable for G.W. Bush. If controlling for GDP and other time-varying factors is insufficient for addressing an auto-correlated time trend, then the Bush dummy should resemble the Obama dummy. However, while the Obama effects in table 2 are substantively and statistically significant in all specifications, the Bush effect is about half the magnitude and reaches statistical significance only in the most poorly-specified half of the models (online appendix R). This offers some reassurance that the results are not an artifact of our model specification.

In addition to the main analyses, we also estimated out-of-sample forecasting models to generate a synthetic counterfactual: what Obama would have proposed if he had behaved similarly to his predecessors given relevant conditions. We then compare these counterfactual predictions to Obama’s actual proposals to determine whether he exceeded expectations. This approach complements our previous analyses by further addressing concerns about how model specification may have affected the findings. The rigor of the synthetic counterfactual approach depends primarily on the predictive power of the forecasting model—not the theoretical relevance or explanatory power of the individual covariates. Coefficient bias due to model specification has no bearing on the overall predictive power of forecasting models. Thus, so long as we generate an accurate forecasting model and Obama still outperforms the synthetic counterfactual, the conclusion that he provided stronger policy representation remains.

The synthetic counterfactual approach required setting aside Obama’s data to estimate the forecast models, so we used data only from Obama’s predecessors. To ensure that the models generated accurate predictions, we successively removed insignificant controls to identify models with the highest predicted R² and lowest out-of-sample mean absolute prediction errors (MAPE) for Obama’s immediate predecessor, G.W. Bush (refer to online appendix S for these forecasting models).¹⁷ Again, for this type of analysis the choice of specific covariates matters much less than the overall predictive power of the model. For anti-poverty spending, all predictors except GDP growth, unemployment rate, and poverty rate were included. For civil rights spending, only GDP remained. In both models, the control for GDP accounts for over-time growth in proposed spending. These models generated reasonably accurate out-of-sample forecasts for Obama’s predecessor, G.W. Bush: his actual and predicted poverty and civil rights proposals were not significantly different (refer to online appendix T for additional details). This suggests that the forecasting models exhibit sufficiently high predictive power to determine whether Obama exceeded expectations. As such, if there exists some unobserved alternative explanation (including the incrementalist budget theory) that can account for Obama’s relatively greater spending, the forecasting models should generate predictions that are no different from his actual proposals. On the other hand, if the models do not accurately predict spending during the Obama years, the difference can be attributed to Obama’s distinctive policy efforts. The resulting forecasts for Obama, generated using political and economic data from his years in office, are shown in figure 5.

Figure 5 Obama’s predicted versus actual anti-poverty and civil rights spending

Figure 5 plots Obama’s predicted anti-poverty and civil rights proposals, bracketed by 95% confidence intervals, against his actual spending. For anti-poverty proposals, Obama exceeded the prediction in every year, often by a large amount. For civil rights proposals, Obama exceeded the predictions in five of the eight years, and marginally so in two additional years. Obama’s total predicted anti-poverty proposal was $812 billion, versus his actual total proposal of $1,383 billion (p < .05). Obama’s total predicted civil rights proposal was $12.17 billion, versus his actual total proposal of $13.80 billion (p < .05). Bush’s predicted proposals serve as a placebo test. Online appendix T shows that Bush’s proposals conform to predictions, in contrast to Obama’s proposals in figure 5. Bush’s actual poverty proposals fell within the predicted range in most years, and in the few years where they fell outside, it was just over the threshold. Obama’s actual proposals fell well above the predicted range in every year. Similarly, for civil rights, Bush’s actual proposals fall within the predicted range in all but one year. Obama’s are well above the predicted range in most years, and in the remaining years fall at the upper end of the predicted range.

Finally, to allay concern that budget proposals may not accurately reflect presidential priorities, we present a robustness check using presidential DW-Nominate scores derived from roll call votes related to poverty and civil rights. Online appendix F shows that on roll-call votes that disproportionately impacted African Americans, Obama was substantially more liberal than both his Republican and Democratic predecessors. By contrast, in line with our argument, DW-Nominate scores generated from presidential positions taken across all issue areas depict Obama as the most moderate president since World War II (see also Klein 2012). This discrepancy implies that although Obama was ideologically moderate overall, he made an exception for poverty and civil rights. These findings underscore the utility of distinguishing presidential positions in discrete policy areas from their overall ideology.

In sum, Obama proposed spending substantially more than his predecessors on both poverty relief and civil rights. His large spending proposals cannot be accounted for by economic or political factors. For instance, the forecasting models suggest that he proposed more than expected regardless of whether Congress was controlled by Democrats or Republicans. Additionally, the magnitude of the Obama effect is substantial, equaling or exceeding the effect of Democratic party control of both Congress and the presidency. These findings, corroborated by presidential DW-Nominate scores for poverty and civil rights, are consistent with the hypothesis that Obama provided better policy representation for African Americans than other presidents.

Conclusion

The Obama presidency is an opportunity to test theories of descriptive representation in the highest office in the American political system. As an African American president, Obama was expected to provide stronger substantive representation to racial minorities. Did he do so? Despite a growing number of studies of his presidency, this question has received little attention. What attention it has received has yielded a uniformly negative verdict. As one eminent political historian put it, “in several areas . . . there was great continuity, either by choice or by political necessity, between the administrations of Obama and President George W. Bush . . . Nowhere was this more frustrating and disappointing than in race relations” (Zelizer 2018, 5-6). However, these studies have not systematically analyzed Obama’s policy efforts as distinct from his rhetorical efforts. More generally, very few studies compare the representation of any identity group on the same metric across presidents (although see Gillion 2016). We aimed to do so, with new measures of policy substantive representation distinct from rhetorical substantive representation. We also control for circumstances, which existing studies of presidential substantive representation had yet to do. The results show that Obama provided stronger policy representation for African Americans, even as he provided weaker rhetorical representation on the same issues, with the exception of his abstention from negative mentions of poverty and the potential symbolic resonance to many African Americans of his frequent talk about the middle class.

These findings speak to several areas of scholarship. First, theories of descriptive representation suggest that a constituency is best represented by its own members. Our results provide some evidence that this expectation holds for the highest and most consequential of all political offices. However, there are important caveats. Descriptive representatives may provide considerable substantive representation in policy efforts, but not necessarily in rhetoric. We thus highlight a significant potential trade-off between these two facets of substantive representation. While the normative literature has discussed various problems with descriptive representation, this trade-off has not been attended to and further complicates the concept of descriptive representation (Mansbridge 1999). Neither has the empirical literature on descriptive representation and minority incorporation offered a distinction between these two facets, though it has long recognized the constraints that can lead to de-racialized rhetoric. We advance both literatures, suggesting the utility of unpacking two distinct facets of substantive representation.

The results also engage with the literature on the politics of social welfare policy. In a classic book, William Julius Wilson (1987) argued that the best way to provide policy that aids African Americans is with a “hidden agenda.” To allay the concerns of many white voters, Wilson recommended a political strategy that avoids speech about race but implements economic policies that stand to help African Americans. He urged leaders to pursue policies that alleviate societal needs, thereby indirectly disproportionately benefitting African Americans (Haynie 2001). Though the evidence is merely suggestive, Obama’s behavior was consistent with this type of strategy. He avoided talk of poverty or race, while attempting to increase spending in these policy areas.¹⁸ His spending proposals offered disproportionate assistance to African Americans, but also benefitted many white Americans (Gilens 1999).

Our findings also suggest that because descriptive representatives of majority-white jurisdictions may not be well-positioned to improve the rhetorical representation of their descriptive constituency, interest groups and social movements play an important role in rhetorical representation specifically. Different institutions may be better suited to advancing different types of efforts toward equal representation. Descriptive representatives may be able to advance concrete policy proposals, while non-governmental groups can give voice to concerns and raise the salience of their issues (Harris 2012; Henry, Allen, and Chrisman 2011).

In addition, the findings speak to scholarship on presidential rhetoric and policy-making. While presidential studies sometimes emphasize the power that accrues to presidents from “going public” or “the bully pulpit,” the president may adopt the opposite tactic when the policy is unpopular. This is consistent with other scholars who find that presidents do not seek to raise the salience of a policy proposal if it is likely to be unpopular (Canes-Wrone 2001; Covington 1987). As a member of a marginalized minority, Obama may have faced additional constraints. Obama may have had less capacity to publicly advocate for black interests due to the potential backlash in public opinion (Hajnal 2006; Nelson, Sanbonmatsu, and McClerking 2007; Tesler 2016). In other words, the public may be less inclined to support a policy designed to improve the lives of African Americans if its advocate is an African American representative. This could partly explain Obama’s relative silence on race and poverty.

Methodologically, our study bolsters the utility of using concrete, specific policy proposals (Canes-Wrone 2001). Measures of policy priorities that rely on officials’ major public utterances tend to miss important policy actions, especially when going public carries costs. Relying on Obama’s public statements as a measure of his intent or effort leads to the incorrect conclusion that he provided weaker substantive representation than other presidents. More generally, it is more accurate to unpack officials’ public speeches from their policy proposals, and examine the conditions under which these two types of representation diverge or move in tandem.

To date, conclusions about Obama’s representation have been quite negative because they focus on his rhetoric. Yet the results here show that Obama’s concrete proposals—the actual dollars he sought to spend—constitute a clear increase in substantive representation for minorities. In that respect, relative to comparable predecessors, Obama’s presidency was advantageous to African Americans. Whether the cause is descriptive representation is a question that can only be more definitely answered with the election of future minority presidents.

Supplementary Materials

Appendix A. Keywords in State of the Union Addresses

Appendix B. Content Analysis Methodology

Appendix C: Inter-Coder Reliability

Appendix D. Details for Budget Data

Appendix E. Obama’s Public Positions, Vetoes, and Executive Orders

Appendix F. Robustness Check Using Presidential DW-Nominate Scores

Appendix G. Anti-Poverty Programs in the Analyses

Appendix H. Standardizing Project-Based Housing Assistance Proposals

Appendix I. Civil Rights Programs in the Analysis

Appendix J. Details for Alternative Spending Measures

Appendix K. T-Tests for Obama’s Poverty, Civil Rights, and Middle Class Rhetoric

Appendix L. Yearly Rhetoric Analyses

Appendix M: Yearly Proposed Anti-Poverty and Civil Rights Spending

Appendix N. Regressions for Alternative Spending Measures

Appendix O. Robustness Checks for Proposed Spending

Appendix P. Regressions Comparing Obama to Democrats and Republicans

Appendix Q. T-Tests for Obama’s Proposals Under a Republican vs. Democratic Congress

Appendix R. Placebo Tests for Spending by G.W. Bush

Appendix S. Regressions Used to Predict Obama’s Spending

Appendix T. Out-of-Sample Predictions for G.W. Bush

To view supplementary material for this article, please visit https://doi.org/10.1017/S1537592719000963