The observation that presidents dominate US foreign policy is hardly novel. Around the world, executives generally have informational advantages in international affairs that translate into foreign policy autonomy (Baum and Potter 2015). Political institutions and practices make this doubly true in the United States. Indeed, presidential dominance in foreign policy is so widely accepted that there is shorthand for it: “two presidencies,” a weaker one for domestic policy and a stronger one for foreign affairs (Wildavsky 1966).
Most scholarship, however, glosses over how we got to these two presidencies, gauzily attributing it to presidential entrepreneurialism and the emergence of the United States as a superpower. In this telling, presidential foreign-policy power is the combined product of geopolitical change that thrust the US into the role of a global superpower and the savvy of presidents who availed themselves of that opportunity.
In a recent article published in Political Science Research and Methods, Thomas Gray, Philip Potter, and I make a different argument: the emergence of the modern presidency was partially the consequence of the shift to the direct election of US senators, which realigned the division of foreign policy power between the branches. By shifting senators’ electoral principals from state legislators to the median state voter, direct election disincentivized attentiveness to foreign policy—a domain that the average voter knows and cares little about. With senators no longer rewarded for muscularly engaging in foreign policy, presidents inserted themselves into the vacated political space.
We leverage the piecemeal phase-in of senatorial direct election to identify the impact on votes to either expand or limit presidential authority over tariffs or treaties between the 45th (1877-79) and 63rd (1913-15) Congresses. Our results indicate that senators subject to direct election were more inclined to delegate foreign policy autonomy to presidents. We confirm these results with a placebo test on similar votes in the US House, where no equivalent institutional change occurred.
Direct Election and Presidential Power
The shift to direct election of senators was driven by Progressive-era ambition to bring government closer to the people it ostensibly represented. There were more practical considerations at play as well—that direct election would limit corruption by making influence costlier and avoid the Senate vacancies produced by deadlocked state legislatures (Schiller and Stewart 2015).
As with any major institutional change, there were controversies that delayed adoption. Southern leaders, for example, resisted for fear that it would threaten the Jim Crow system. That said, in comparison to the fights over other constitutional amendments, the debate was muted. By the time the 17th Amendment was ratified in 1913, a majority of states, led by those in the West, had already moved toward direct election through the adoption of direct primaries (Figure 1).
While direct election may have moved the Senate closer to the ideals of plebiscitary democracy, we argue that it also shifted the incentives of senators in ways that directly empowered the executive in the area of foreign policy and contributed to the emergence of the modern presidency. Why?
The answer lies in the ultimate source of accountability. As elected agents, senators are held responsible for their actions by principals who must be both attentive and knowledgeable about their foreign policy performance. Information is required to ensure this accountability, but it hard for the average citizen to acquire this information and the individual rewards for doing so are minimal. Holmstrom (1979) notes that monitoring foreign policy is especially difficult due to the natural information asymmetries between leaders and citizens, which is compounded by low popular attentiveness to (and knowledge about) foreign policy.
The shift to direct election exacerbated these information and attentiveness gaps by shifting the principal from state legislators to citizens. Absent major conflict, foreign policy operates well beneath the public’s radar. Voters do not know much about or engage with foreign policy, and they tend to vote based on their immediate concerns rather than foreign policy considerations.
Compared to the mass public, state legislators were (and are) more informed about and concerned with matters of national significance including foreign policy (Gailmard and Jenkins 2009). Around the turn of the 20th century, as they remain today, state legislators were, on average, wealthier and better educated than their constituents, attributes that are consistently associated with better political information and foreign policy attentiveness.
Once citizens were given direct-election responsibility, senators had little choice but to change their behavior accordingly, with increased attention to local concerns at the expense of foreign policy. The takeaway is that the insulation that came with indirect election led to a Senate that was better positioned to constrain executive preferences over foreign policy. As that insulation evaporated as a consequence of direct election, the incentives to meaningfully constrain presidential foreign policy preferences went with it.
Finally, states in this period more directly bore the cost of assertive foreign policy and conflict than they do now. This meant that state legislatures and powerbrokers had substantial incentives to constrain presidential foreign policy initiatives that might impact their budget priorities. Peacetime defense outlays and deficit spending were minimal, so there was a tighter link between foreign adventures, state budgets, and taxation. Trade and treaties were also more immediately salient in this period because they dictated the bulk of the federal budget prior to the imposition of the income tax.
This leads us to the simple expectation that a shift from indirect to direct election of senators will be associated with an increased probability of voting to delegate foreign policy autonomy to presidents.
Research Design and Analysis
To assess this expectation, we leverage the phased adoption of direct primaries prior to the ratification of the 17th Amendment. This approach allows for better identification of the causal process by providing variation within Senate roll calls and within senators, with some senators facing popular election and others still answering to state legislatures.
To establish a valid set of votes for comparison, we rely on all roll calls on tariffs and treaties in the Senate from the 45th (1877-79) through 63rd (1913-15) Congresses—the period after Reconstruction until the ratification of the 17th Amendment. Among this set, we identify those votes that explicitly sought to expand or limit presidential discretion. Examples include authority to staff commissions and authorization to choose the details of policy implementation. These roll calls are mostly on amendments, motions, and resolutions that narrowly address presidential power, rather than final-passage votes on full multi-faceted bills. As required, we recoded such that a positive observation indicates a position in favor of presidential autonomy. The result is a dataset of 1,916 individual votes from 32 roll calls, 15 of which were on tariff issues and 17 on treaty-implementation issues. Our unit of analysis is the senator-vote.
We define the “treatment,” which we call Direct Election, as whether the senator’s state had, by that Congress, introduced some form of direct primary for the election of senators. We also include a dichotomous measure of co-partisanship (President Co-partisan) on the expectation that this could lead to more willingness to support an expansion of presidential authority.
In Table 1, we present results from simple linear probability models as well as conditional logistic regressions that better fit the dichotomous dependent variable. Within each model type, we estimate models with (1) senator and roll-call fixed effects and (2) state and roll-call fixed effects. We also include a specific time trend for each state in all models and cluster our standard errors at the senator or state level, depending on the fixed effects. State-specific time trends address the possibility that states’ preferences on these issues changed over the period, and that this change could vary by state. One potential threat to our design is if important features of states’ preferences for foreign-policy oversight were also important for switching to popular election. Our state-specific time trends help partially mitigate this by allowing states to change in different directions and in different rates.
The two fixed-effects approaches allow us to test slightly different things. The senator-fixed-effects models assess changes to the individual senator’s behavior rather than change brought about by the election of new senators. This measures the impact of the change-in-principal on individual senators. In total, we observe 26 senators with 396 total votes cast before and after a change to direct primaries. In the state-fixed-effects models, information is drawn from changes within states’ combined voting, which combines the effects of individual change with replacement, the changes brought about by new senators with different tendencies getting elected under a different electoral system.
The models indicate that senators subject to direct election were more likely to vote in favor of presidential autonomy. Senators subject to a popular vote were about 25 percentage points more likely to support presidential discretion than earlier in their careers under legislative electoral systems (Model 1). We recover a similar relationship (22 percentage points) in the state-fixed-effects model (Model 3).
Our theoretical arguments do not predict whether this transition is driven by changing the behavior of existing senators or replacing senators with new ones. However, a comparison of the models allows us to speak fairly conclusively to that question, and the evidence weighs in favor of the conversion of existing senators. Specifically, the similarity of these two effect sizes between Model 1 and Model 3 indicates that the impact of direct election arises primarily from the changes in the behavior of sitting senators rather than changes in who was in the Senate. This does not, however, foreclose the possibility that, had existing senators been less responsive, replacement could also have been a powerful mechanism.
Co-partisanship is only significant in models with state fixed effects, most likely due to the limited variation in presidential partisanship in these phase-in years (Democrats only held the presidency during the split administration of Grover Cleveland) meaning that many senators had little or no variation in presidential party during their time in office. As a consequence, the state-fixed-effects models are better equipped to reveal the role of co-partisanship—senators sought foreign policy autonomy for their own party’s presidents more than rival-party presidents by about 10 percentage points.
To better understand the source of the shift to deference on foreign policy matters and the extent to which it might be concentrated among the president’s co-partisans or rivals, we interact Direct Election and President Co-partisan. Again, we present four models (Table 2): two OLS models (1 and 3) and two conditional logistic regressions (Models 2 and 4), varying whether we specify senator or state fixed effects.
The increase in deference among partisan rivals as we move from indirect to direct election likely owes to the low partisan salience of foreign policy at the turn of the 20th century. In this context, the president’s partisan opponents in the Senate were more available to shift once overseen by less watchful eyes, in large part because co-partisans were naturally more likely to delegate even before the institutional shift to direct election. It was primarily party power brokers who enforced state legislative oversight of senators when it came to constraining presidential foreign policy. As a result, co-partisans were under less pressure than rivals to constrain under indirect election, allowing for the greater swing to occur among the rival partisans with the transition to direct election. For example, during the McKinley, Roosevelt, and Taft Administrations, attempts to provide the president with authority to negotiate trade agreements (to promote trade reciprocity) and to create a Tariff Commission found some Democratic support. This pattern is borne out in Figure 2.
Additionally, we consider the possibility that we are capturing larger Congressional trends that happened to coincide with the transition to direct elections. This would undermine our thesis as the shift to direct election only occurred in the Senate. To address this, we conduct placebo tests on the US House, following identical specifications to those found in Tables 1 and 2. We find no comparable relationship with switching senatorial election type on House votes. (As an example, Figure 3 below is derived from Table 1 Model 3 and the identical model for the House.) This implies that the changes in senatorial allowance of presidential discretion are senate-specific and timed to the piecemeal adoption of popular election.
We find that the shift to the direct election of senators played an underappreciated role in the emergence of the modern presidency. Prior to direct election, comparatively more sophisticated and informed state legislators were the principals selecting and overseeing senators. Afterward, amateurs (regular citizens who were less sophisticated and informed) took on this role. As a consequence, the shift to direct election broke the careful equilibrium required for the system of checks and balances to function as the Framers intended and fundamentally altered both the demographics and policy preferences of the relevant median voter, with predictable consequences. With their political futures now determined by direct election, senators no longer had sufficient incentives to engage with foreign policy, thus opening the door for the emergence of modern presidents.
 Even when not particularly versed in or concerned with foreign affairs in their own right, state legislators were pushed in ways that led to the election of senators better positioned to constrain presidential foreign policy preferences. State legislators in this period were highly reliant on “power brokers”—party bosses, party machines, and wealthy patrons—for political survival. These power brokers, who had wide-ranging business and financial interests, pushed state legislators to select senators who were attentive to defending the institutional prerogatives of the chamber, including in the realm of foreign policy.
 The approach also differentiates between our argument and the rival possibility that the federal income tax empowered the modern presidency by increasing financial autonomy. While the 16th and 17th Amendments were ratified in close succession, the income tax was not adopted piecemeal by the states.