The Effect of Partisan Representation at Different Levels of Government on Satisfaction with Democracy in the United States

Abstract In this paper, we analyze how variations in partisan representation across different levels of government influence Americans’ satisfaction with the democracy in the United States. We conduct two survey experiments and analyze data from the 2016 American National Election Study postelection survey. We find that Americans are the most satisfied with democracy when their most preferred party controls both the federal and their respective state governments. However, we also find that even if an individual’s least preferred party only controls one level of government, they are still more satisfied with democracy than if their most preferred party controls no levels of government. These findings suggest that competition in elections across both the national and state government, where winning and losing alternates between the two parties, may have positive outcomes for attitudes toward democracy.

However, in the United States, there are many opportunities for one's own party to have some governmental power. This can occur within the federal government, where parties have opportunities to control the Presidency, Senate, and House of Representatives, in state governments, where similar mixes of control can occur between the executive and the legislature, and even at the local (city or county) level with mayors, county commissions, and city councils. This American federalism provides partisans multiple opportunities for their preferred party to be in control of government. But, the above quote from Bill Bishop suggests that partisans crave increasing amounts of co-partisan government control-those in a Democratic city in a Republican state like Austin would prefer to move to Portland, a Democratic city in a Democratic state.
In American politics, scholars have observed increasingly tenuous support for American political institutions among the public (Levitsky and Ziblatt 2018). We argue that having one's preferred party in control of government should serve to increase satisfaction with democracy and support for democratic institutions in the United States. Just as individuals have more support for government institutions when their preferred party is in power, they should also have more generalized support for democracy. As voters become more partisan and begin to care more about partisan outcomes instead of the quality democratic processes (Levendusky 2009;Mason 2018), their support for democratic institutions may diminish overtime, especially if their party is out of power. Yet, federalism could mitigate this process. As long as a voter's preferred party wins elections at one level of government, they should still be somewhat satisfied with the democratic process and somewhat likely to support democratic institutions than if their most preferred party did not control any levels of government.
In this paper, we analyze how variations in partisan representation across different levels of government influence Americans' satisfaction with the democracy in the United States (Anderson and Guillory 1997;Anderson et al. 2005). We argue that Americans are the most satisfied with the democratic process when their most preferred party controls both the federal and their respective state governments. However, we also argue that even if an individual's least preferred party only controls one level of government, they would still be more satisfied with democracy than if their most preferred party controls no levels of government. That is, partisan control of government should influence not only how individuals feel about political outcomes, but it should influence their views of the democratic system as a whole.
This provides a unique insight into state politics in the United States. Given that state-level political decisions are often so consequential for citizens (Bowling and Pickerill 2013), it is perhaps unsurprising that citizens would care about state-level political outcomes. However, the considerable extant research on electoral outcomes and satisfaction with democracy focuses heavily on national level election outcomes. Given that support for political institutions and democracy in general is so vital to the health of democracy (Levitsky and Ziblatt 2018), state-level outcomes that give citizens more of a chance to "win" could benefit the health of democracy generally, if citizens are more supportive of democracy given that their preferred party controls at least one level of government.
To analyze the effect of federal and state partisan control on Americans' satisfaction with democracy, we use data from the two experimental studies and the 2008 and 2016 American National Election Study (ANES). The data from the experimental studies allow us to analyze how respondents react when different parties hypothetically win state and federal elections, and how those reactions influence their satisfaction with democracy. Since respondents in the experiments all had the equal chance of their preferred party hypothetically controlling their state and/or federal governments, we are able to account for differences in partisan type (Democratic or Republican), which is not possible using survey data, where partisan control of government is fixed at each point in time. Nevertheless, it is also important to determine how actual partisan representation influences voters' actual satisfaction with democracy. Hence, we also analyze real-world ANES data from 2008 and 2016.
The results of both experimental studies and the ANES analysis demonstrate that voters become more satisfied with democracy as the number of levels of government their most preferred party controls increases. Citizens are the most satisfied with democracy when their most preferred party controls both the state and federal government and the least satisfied with democracy when their least preferred party controls both the state and federal government. When their preferred party controls only one level of government, citizens remain more satisfied than if they control no levels, but less satisfied than citizens with dual control. We also find that these results are generally the strongest for stronger partisans. Note that these findings are in direct contrast to findings of previous studies that suggest while partisan representation at the federal level has a strong impact on satisfaction with democracy, partisan representation at the local level does not (Henderson 2008;Singh, Lago, and Blais 2011).
In the next section, we develop our theoretical argument while drawing upon several studies that analyze factors that influence satisfaction with democracy. We then analyze data from our two experimental studies and analyze data from the ANES. We conclude by discussing how the results of our analysis speak to the literature on support for democratic institutions in the United States.

Variations in Satisfaction with Democracy
There are a variety of factors that might influence individuals' satisfaction with the democratic process. One primary factor is whether a citizen's preferred party controls key policymaking institutions. Generally speaking, citizens who identify with parties or candidates that control key policymaking institutions in government are more satisfied with democracy than citizens who identify with opposition parties (Anderson and Guillory 1997;Anderson et al. 2005;Anderson and Tverdova 2001;Blais, Morin-Chassé, and Singh 2017;Howell and Justwan 2013;Singh, Karakoç, and Blais 2012). When a citizen's most preferred party or candidate controls key policymaking institutions, they likely develop policies the citizen supports, and when a citizen's least preferred party or candidate controls key policymaking institutions, they may develop policies that the voter dislikes. As such, it not surprising that voters are more satisfied with democratic outcomes when their most preferred party controls key policymaking institutions (Curini, Jou, and Memoli 2015;Ferland 2016;Reher 2014;Stecker and Tausendpfund 2016).
Beyond policymaking, citizens likely care about which parties control key policymaking institutions because of the psychological response citizens have to election outcomes. Citizens tend to have strong partisan attachments in the United States, and the parties are increasingly becoming sorted along ideological lines (Levendusky 2009;Mason 2018). As such, citizens have an increasing affective, or emotional, attachment to their parties (Mason 2018;Spoon and Kanthak 2019). Individuals may then feel more positive emotions when their preferred party wins an election, and more negative emotions when they lose. These feelings may impact satisfaction with democracy. Due to their positive emotional response to election outcomes, citizens may become more satisfied with democracy the more often their most preferred party wins. Likewise, due to their negative emotional response to election outcomes, citizens may become less satisfied with democracy the more often their preferred party loses (Anderson et al. 2005;Curini, Jou, and Memoli 2011;Singh 2014).
While citizens become less satisfied with the democratic process when their preferred party does not control key policymaking institutions, institutional arrangements can influence how dissatisfied they are with the democratic process. Under majoritarian institutional arrangements, majority interests tend to be well represented in the policymaking process while minority interests are not. In particular, single member district, plurality rule elections tend to repress the number of parties that can realistically win elections and control key policymaking institutions. Additionally, single member district, plurality rule elections can also lead to disproportionate election results, where one party gets a much larger seat share than their vote share. This could lead to the overrepresentation of the winning party and the underrepresentation of the losing party. Furthermore, unitary institutions only allow for representation at one level of government. In conjunction with single member district, plurality rule elections, this implies that only the interests of the winning majority party are represented in the policymaking process at one level of government. Thus, policy tends to follow the interests of the majority party in government and excludes the preferences of opposition party. As a result, it is not surprising that citizens who identify with opposition parties are very unsatisfied with democratic outcomes in these types of systems (Anderson et al. 2005;Berhard, Nordstrom, and Reenock 2001;Ezrow and Xezonakis 2011;McDonald, Mendes, and Budge 2004;Powell 2000;VanDusky-Allen 2017).
In contrast to majoritarian arrangements, under consensual arrangements, multiple interests are represented in the policymaking process. In particular, proportional representation elections may allow multiple parties to win elections and collectively control key policymaking institutions. Furthermore, federal arrangements allow multiple parties to compete for control over key policymaking institutions at several levels of government. Since no one single party or interest has a majority control over a key policymaking institution, multiple parties and interests must form coalitions and compromise in the policymaking process. As a result, policy outcomes tend to be more moderate (McDonald, Mendes, and Budge 2004;Norris 1999;Powell 2000).
Since multiple parties are involved in the policymaking process and policy outcomes are more moderate in consensus democracies than majoritarian democracies, citizens who identify with opposition parties are less unsatisfied with democracy in consensus democracies than in majoritarian democracies. For citizens in consensus democracies, even if their party does not win, another party with similar preferences may be involved in the policymaking process, closely representing their interests. Additionally, citizens who identify with opposition parties may prefer the more moderate policy outcomes under consensual institutions over the more extreme ones that could happen under majoritarian institutions (Anderson and Guillory 1997;Anderson et al. 2005;Bernauer and Vatter 2011;Ezrow and Xezonakis 2011;Lijphart 1984Lijphart , 1999Miller and Listhaug 1990;VanDusky-Allen 2017).
While majoritarian institutions enhance opposition party supporters' dissatisfaction with democracy while consensual institutions reduce opposition party supporters' dissatisfaction with democracy, most countries do not have purely majoritarian or consensual institutions. Instead, they have a mixture of both. Herewith in we focus on how a mixture of consensual and majoritarian institutions in the United States influences satisfaction with democracy. In particular, we focus on how the interaction between federalism, a characteristic of consensual democracies, and single member district, plurality rule elections with two major political parties, characteristics of majoritarian democracies, influence satisfaction with democracy.
Federal institutional arrangements give voters multiple opportunities to elect representatives at different levels of government. In conjunction with single member district, plurality rule elections with two major political parties, this could lead to a variety of election outcomes for voters in the United States. At one extreme, citizens can have limited partisan representation in government, where both their state and the federal government are controlled by their least preferred party. On the other extreme, citizens can have full partisan representation at all levels of government, where both their state and federal government are controlled by their most preferred party. Another likely alternative is that citizens could have full partisan representation at one level of government but not the other (Anderson and Guillory 1997;Anderson et al. 2005;Blais and Gélineau 2007).
Additionally, there is evidence to suggest that individuals may, at some level, be interested in geographically sorting into areas where their political identity is in the majority. Bishop (2009) notes that geographic clustering of individuals in the United States has created politically homogenous communities. However, these claims are not without critique, as county-level party registration fails to show a similar pattern (Abrams and Fiorina 2012). Later work has demonstrated that partisanship, while a smaller factor than a host of others, plays a role in internal migration decisions in the United States, especially for Republicans (Tam Cho, Gimpel, and Hui 2013). Regardless of the true level of effects of partisan geographic sorting, it seems likely that this is a consideration for some people. This suggests that state-level electoral outcomes can influence a host of political attitudes, even satisfaction toward democracy.
The extent to which a citizen's most preferred party controls the state and/or local governments should affect the extent to which they are satisfied with democracy. If a citizen has limited partisan representation at the federal and state levels, they should be the least satisfied with democracy. In contrast, if a citizen has full partisan representation at the federal and state levels, they should be the most satisfied with democracy. Next, if a citizen has full partisan representation at one level but not another, they should be moderately satisfied with democracy. While they may not be fully represented in the policymaking process, they are represented at least at one level. As such, while they should be less satisfied with democracy than they would otherwise be if they had full partisan representation at all levels, they should be more satisfied with democracy than someone with limited partisan representation at all levels. This discussion leads to the following hypothesis: Hypothesis 1: As co-partisan control over government increases, satisfaction with democracy increases.
While increased co-partisan control over government should have a positive effect on satisfaction with democracy, citizens might care more about co-partisan control of one level of government over another. Yet, it is not entirely straightforward as to which level of government should matter more, the federal government or the state government. There are reasons why citizens could care about both.
Given the evolution of the federal government's policymaking powers throughout the 20th century, it is clear why citizens should care about co-partisan control over the federal government. Due to this expansion of its policymaking powers, the federal government adopts policies that affect a wide range of activities in states nationwide. In some cases, states may be willing to work with the federal government to pass and implement policies. In other cases, the federal government may have to coerce state governments to pass and implement policies through regulations and financial pressure (Bowling and Pickerill 2013). As such, citizens may care deeply about who controls the federal government as the federal government has a great deal of influence over the policymaking process at the state level.
On the other hand, states still play an important role in the policymaking process. State governments have taken the lead in developing state level policies in several different policy areas, such as in environmental policy, marriage and family policies, health care, education, marijuana policy, land management, and gun rights. Recently, it has been the states leading the fight against COVID-19 in the US states. It is also important to note that states often have a strong influence over the implementation of federal policy at the state and local level (Bowling and Pickerill 2013). As such, citizens may be acutely aware of the impact that state level governments have over public policymaking and therefore should prefer that their preferred party controls the state government.
Given that the federal government has become increasingly powerful over time, citizens may care more about co-partisan control over the federal government than their own state government. Alternatively, since states are willing to fight back against perceived federal overreach while at the same time engaging in their own policy innovations at the state level, citizens may care more about co-partisan control over the state government than the federal government. Hence, we have two different expectations as to how co-partisan control at different levels of government should influence satisfaction with democracy. Co-partisan control over the federal government could have a stronger positive effect on satisfaction with democracy than control over state level government. Or co-partisan control over state governments could have a stronger positive effect on satisfaction with democracy than control over the federal government. This discussion leads to the following two hypotheses: Hypothesis 2a: Compared to no co-partisan control, federal co-partisan control has a stronger positive effect on satisfaction with democracy than it does on state level government control.
Hypothesis 2b: Compared to no co-partisan control, state co-partisan control has a stronger positive effect on satisfaction with democracy than it does on federal level government control.
While co-partisan control over different levels of government might influence satisfaction with democracy for most citizens, we also suspect that strong partisans may care more about which parties control different levels of government than weak and leaning partisans do. These stronger affective partisan attachments among strong partisans (Mason 2018) should cause partisan control of government to matter more to these individuals. Strength of attitudes and attachments are correlated with perceived importance of these attachments (Petty and Krosnick 1995), suggesting that strong partisans may view partisan control of government as more important than weaker and leaning partisans. Partisan leaners also engage with politics differently than stronger partisans, the latter of who are more driven by ideological concerns (Klar 2014). As such, the effect of co-partisan control over government should exert a greater positive effect on strong partisans' satisfaction with democracy than it does over weaker partisans' satisfaction with democracy. This discussion leads to Hypothesis 3: Hypothesis 3: The positive effect that co-partisan control has on satisfaction with democracy becomes stronger as the strength of partisanship increases.
In order to provide support for our hypothesis, we conducted two survey experiments and analyzed data from the ANES in 2008 and 2016. We first chose an experimental design because we wanted to analyze how respondents reacted when different parties hypothetically controlled different levels of government. This allowed us to control out for other factors that might influence citizens' reactions to partisan representation, such as elected officials' characteristics. Additionally, in the experiments, respondents all had the equal chance of having partisan representation at both levels of government, no partisan representation at any level of government, or partisan representation at one level of government but not another. This is not possible with real world data as particular types of respondents would be more likely to fit within these different categories. Nevertheless, we also felt it was important to determine how actual partisan representation affected actual satisfaction with democracy. Hence, we also analyzed real-world ANES data from 2008 and 2016. We discuss the results of the experiments and ANES analysis below.

Survey Experiments
In our first survey experiment, we recruited participants from undergraduate courses at a regional state university in the Fall of 2018. Students were recruited from a variety of courses, though mostly were political science or criminal justice majors. Students participated in a multi-module omnibus study in exchange for extra credit in their courses. A total of 370 students participated in this study. As expected with a student sample, the sample skews young, with an average age of 22, ranging from 18 to 67. The sample was largely female, at just under 61%, and heavily white, at just over 78%. About 46% of the sample identified as Democrats or leaners, 41% Republicans or leaners, and 13% pure independents.
Our second study was conducted in March 2019 on Amazon's Mechanical Turk (mturk). Using mturk allows us to recruit a more diverse sample than is available using undergraduate students. In this study, we recruited participants 2 from mturk's platform to complete a short study, which averaged about 3 minutes to complete, in exchange for a payment of 50 cents. The sample's age ranged from 18 to 77, with an average of 37. The sample was about 55% male, and 45% female, and about 71% white. As is typical in mturk samples, the sample skewed Democratic-about 57% of the sample identified as Democrats, 31.5% Republican, and 11.5% pure independents.
In each study, participants were randomly assigned to one of four experimental groups, after answering a series of demographic questions. Participants read a brief text, asking them to imagine a state where partisan control varies at the state and federal level. The text reads as follows: This created a traditional 2 Â 2 experimental design, where partisan control of the state government varied between Republicans and Democrats, and partisan control of the federal government varied between Republicans and Democrats. Key for our predictions is the number of levels of government controlled by one's own party. In this case, the variable Hypothetical Co-Partisan Control is coded as 2 if one's own party 3 controls both the state and federal governments, 1 if one's own party controls either the state or federal government, and 0 if the out-party controls both the state and federal government. Of the 310 respondents retained for analysis in the student sample, 72 were experimentally assigned to no co-partisan control, 158 to mixed control, and 80 to complete partisan control. In the mturk sample, of the 886 respondents retained for analysis, 228 were experimentally assigned to no co-partisan control, 436 to mixed control, and 222 to complete partisan control.
After reading this text, participants were then asked, on a 0-10 scale, their hypothetical satisfaction with democracy in the United States. 4 Higher values indicated the respondent is more satisfied with democracy, while lower values mean less satisfied. Based on our expectations in Hypothesis 1, we expected Hypothetical Co-Partisan Control to have a positive effect on hypothetical satisfaction with democracy.
We also collected data on additional variables to control for demographic factors that could be correlated with both satisfaction with democracy and partisanship, since results are conditioned upon an individual's partisanship (see Kam and Trussler 2017). We collected data on participants' age, gender (0 for males, 1 for females), and race (0 for non-white, 1 for white). We additionally collected data on strength of partisanship and political interest.
We use an Ordinary Least Squares (OLS) regression models to analyze the effect of Hypothetical Co-Partisan Control on Hypothetical Satisfaction with Democracy. We present the results of the models in Table 1. The coefficients for Hypothetical Co-Partisan Control are positive and significant at the 0.01 level in both models. The substantive results suggest that as the number of levels of government controlled by a respondent's preferred party increases by 1, they are about one point more likely to express satisfaction with democracy. These results provide support for Hypothesis 1.
The analyses in Models 1 and 2 treat all levels of hypothetical control the same-as in, both state and federal hypothetical co-partisan control are treated as equal. As per out expectations from Hypotheses 2a and 2b, federal co-partisan control and state co-partisan control may have different impact on satisfaction with democracy, versus no control. To this end, we turn to another OLS regression models that allow for variance in the level of co-partisan control. Respondents are coded as controlling Both.
Levels if their party controls both the state and federal government, State Only if their party controls only the state and not the federal government, and Federal Only if their party controls only the federal and not the state government. Individuals for whom their party controls neither level of government are the reference category for these analyses. We present the results of the different levels of co-partisan control in Table 2 and the coefficient plots in Figures 1 and 2.
As shown in Figure 1, when compared to no co-partisan control, the effect of one's own party holding both offices remains substantively large in increasing satisfaction with democracy (β = 2.174, p < 0.01). Next, while there is little evidence to suggest that participants who had state only control were more satisfied with democracy that participants who had no co-partisan control, there is some evidence to suggest that participants who had federal only control were more satisfied with democracy than participants who had no co-partisan control (β = 1.193, p < 0.01).
As shown in Figure 2, the analysis from Study 2 present a slightly different (or, perhaps, less statistically ambiguous) story than in Study 1. For Hypothetical Satisfaction with Democracy, we see that one's preferred party hypothetically holding both offices exerts a large effect of nearly two scale points on democratic satisfaction, compared to holding no offices (β = 1.824, p < 0.01). Meanwhile, both federal (β = 1.109, p < 0.01) and state (β = 0.874, p < 0.01) co-partisan control predict increases of about a 1 scale point in satisfaction with democracy, compared to holding no co-partisan office.
These results suggest that partisan control of government clearly has an effect on satisfaction with democracy. However, it seems to not matter much which level of control one's co-partisans hold. While point estimates are slightly higher for federal control in both studies, compared to only state control, these results are not statistically distinguishable from one another. These results, then, are not clearly supportive of expectations from Hypothesis 2a, that suggested that citizens cared more about federal co-partisan control over government than state control. Here, we find that, while federal control exerts an effect with a magnitude larger than state control in both studies, we are unable to statistically distinguish the two estimates. 5 It is important to note that when we run separate analyzes of these models for Democrats and Republicans only, the effects of each level of control for each party are statistically indistinguishable from  Figure 1. Effect of co-partisan control on satisfaction with democracy (Study 1).
Next, as per Hypothesis 3, we suspect that the effects of Hypothetical Co-Partisan Control on.
Hypothetical Satisfaction with Democracy may be conditional upon the strength of one's partisanship. 6 Stronger partisans may care more than less strong partisans about hypothetical co-partisan control of government. To this end, we turn to an analysis of Hypothetical Co-Partisan Control conditional on strength of partisanship. Strength of partisanship is coded as 1 for leaning partisans, 2 for weak Democrats or Republicans, and 3 for strong partisans. We present the results of this analysis in Table 3 and in Figures 3 and 4. The results from Figures 3 and 4 suggest that as strength of partisanship increases, Hypothetical Co-Partisan Control's effect on Hypothetical Satisfaction with Democracy increases. In Study 1, for Leaning Partisans, an increase in co-partisan control did not seem to influence satisfaction with democracy. But for weak (β = 0.976, p < 0.01) and strong (β = 1.555, p < 0.01) partisans, as the number of levels of co-partisan control increased, satisfaction increased. And the effect was strongest for strong partisans.
In Study 2, as the strength of partisanship increases, the positive effect that co-partisan has on satisfaction with democracy increases. For leaning partisans, a one level increase in co-partisan control leads to a half point increases in satisfaction with democracy (β = 0.446, p < 0.10). For weak partisans, a one level increase in co-partisan control leads to a one-point increases in satisfaction with democracy To avoid issues related to conditioning on post-treatment variables, we measure strength of partisanship, and all other demographics included in all experimental models, prior to the experimental treatments.
(β = 0.832, p < 0.01). And for strong partisans, a one level increase in co-partisan control leads to a one-point increases in satisfaction with democracy (β = 1.218, p < 0.01). Taken together, these results provide substantial support for Hypothesis 3.
We also analyze the effects of different levels of co-partisan control (Both Only, Federal Only, and State Only) on satisfaction with democracy conditional on strength  of partisanship. We present the results of this analysis in Table 4 and in the coefficient plot in Figures 5 and 6. The results from Figure 5 demonstrate that leaning partisans show no effect of the treatment on their satisfaction with democracy. Yet, both weak (β = 1.941, p < 0.01) and strong (β = 3.063, p < 0.01) partisans show increased satisfaction with democracy when their preferred party controls both levels of government, compared to none. Additionally, weak (β = 1.134, p < 0.01) and strong (β = 1.443, p < 0.05) partisans show increased satisfaction with democracy when their preferred party controls the federal government only, compared to none. We repeat this analysis for Study 2, shown in Figure 6. The results suggest that treatment effects are once again mostly pronounced for weak and strong partisan respondents, not leaning partisans. Control over both levels of government predicts about a 2.4 point increase in satisfaction for strong partisans (β = 2.432, p < 0.01), about a 1.7 point increase in satisfaction for weak partisans (β = 1.665, p < 0.01), and nearly a 1 point increase in satisfaction for leaning partisans (β = 0.899, p < 0.10). These were the strongest observed effects. In contrast, control over the federal government only predicts nearly a 2-point increase in satisfaction for strong partisans (β = 1.755, p < 0.001) and nearly a 1-point increase in satisfaction for weak partisans (β = 0.857, p < 0.001). And control over the state government only predicts about a 1-point increase in satisfaction for strong partisans (β = 1.145, p < 0.001) and nearly a 1-point increase in satisfaction for weak partisans (β = 0.797, p < 0.01). These results provide support for Hypothesis 3. The positive effect that co-partisan control has on satisfaction with democracy becomes stronger as the strength of partisanship increases.  When we run the models separately for Democrats and Republicans, the effects of each level of control for each party are statistically indistinguishable from each other. However, in some circumstances, the effects of different levels of control have a statistically significant impact on Satisfaction with Democracy for Leaning and Weak Democrats but not Leaning and Weak Republicans.
Taken together, the results of both experimental studies provide robust support for our hypothesis-when individuals are exposed to some levels of co-partisan governmental control, they are more likely to express satisfaction with democracy, than when their preferred party has no control of government.
These effects are heightened by strength of partisanship and control of multiple levels of government, however we do not see significant differences emerge based on state only, or federal only, partisan control.
While these results are compelling, they come from only two convenience samples. While mturk and student samples have both been long used in academic research, they provide some unique barriers compared to a nationally representative sample-namely, that they are, on average, quite younger, and a bit more educated (Berinsky et al. 2012). We are less concerned about the age differences between these samples and nationally representative samples, as age has long been shown to be a predictor of stronger partisan attachment (see Hobbs 2019). However, in contrast, higher levels of education, which are exhibited in our samples, is associated with stronger levels of partisan strength (Clifford 2017). As such, our younger sample should see lower effects of partisanship on attitudes, given that these attitudes are less crystalized (Hobbs 2019), while our more educated sample might overestimate effects. Since the samples deviate much more strongly on age, rather than education, from nationally representative samples, we are less concerned that our effects are overestimated. However, demographic differences are not the only concern with online samples such as mturk. There has been an increasing rise in automated or fraudulent responses on mturk (Kennedy et al. 2020). While generally concerning, this pattern is less concerning for our purposes. Since these automated responses generally increase error in experimental research (Kennedy et al. 2020), they would lead to larger standard deviations in our effects, making it harder to detect statistical differences. Additionally, we replicate the study on a student sample, which should contain more attentive respondents, and find similar sized effects, suggesting that we do not have particular issues with our mturk sample's quality. However, we realize our experimental findings remain conducted only on convenience samples, and may be detached from real world effect. Because of this, we turn to our survey data to analyze the real-world impact of co-partisan control on satisfaction with democracy.

and 2016 American National Election Studies
To provide additional support for our hypotheses, we use data from the ANES postelection studies in 2008 and 2016. To match the design of our experimental study, we only use data for participants that lived in states where one party had unified control over the state government. 8 To analyze the effect of variations in Co-Partisan Control on Satisfaction with Democracy, we use data from the following question to measure respondents' expressed satisfaction with democracy: "On the whole, are you very satisfied, fairly satisfied, not very satisfied, or not at all satisfied with the way democracy works in the United States?" We re-code the answers to the question such that higher values indicated a respondent was more satisfied with democracy and lower values indicated a respondent was less satisfied with democracy. In the analysis for 2008, the main independent variable, Co-Partisan Control, is coded as 0 for participants that identified as Republicans and lived in states with governments controlled by Democrats (i.e., no co-partisan control), 1 for Democrats and Republicans that lived in states with governments controlled by Republicans (i.e., one level of co-partisan control), and 2 for participants that identified as Democrats and lived in states with governments controlled by Democrats (i.e., full co-partisan control). 9 To determine respondents' partisan identification, we use questions from the survey that asked respondents if they identified with or were closer to a Democrat, Republican, Independent, or Other. We only include respondents that identified as Democrats and Republicans, or felt they were closer to Democrats or Republicans. 10 8 While we are not especially concerned with differential levels of attrition in the pre-and post-election waves, this concern merits some discussion. However, the rates of re-interviews are relatively high for the ANES (between 80% and 90%), and while research has shown that demographic variables may predict attrition in longer time horizon panel surveys, there is little evidence that political variables predict attrition (Frankel and Hillygus 2014). As such, we are not especially concerned that differential rates of attrition in the pre-and post-election surveys are driving these results. 9 Technically, a Republican controlled the presidency during the 2008 post-election wave and a Democrat controlled the presidency during the 2016 post-election wave. However, it was clear that Democrats would soon have unified control over the federal government in 2008 and Republicans would soon have unified control over the federal government in 2016. We expect that individuals based their opinions on the incoming Presidents and Congresses, rather than the outgoing ones. 10 All of the demographic variables come from the pre-election survey.
In the analysis for 2016, the main independent variable, Co-Partisan Control, is coded as 0 for participants that identified as Democrats and lived in states with governments controlled by Republicans (i.e. no co-partisan control), 1 for participants that lived in states with governments controlled by Democrats (i.e. one level of co-partisan control), and 2 for participants that identified as Republican and lived in states with governments controlled by Republicans (i.e. full co-partisan control). To determine respondents' partisan identification, we once again use questions from the survey that asked respondents if they identified with or were closer to a Democrat, Republican, Independent, or Other. Additionally, we only include respondents that identified as Democrats and Republicans, or felt they were closer to Democrats or Republicans.
In all our models, we account for various demographic characteristics of the respondents. We include variables that accounted for respondents' age, gender, race (white versus non-white), income, and education. We additionally account for strength of partisan identification (0 for independent leaning partisans, 1 for "Weak" partisans, and 2 for "Strong" partisans) and political interest, as in the previous studies. We present the results of both models in Table 5.
The results in Table 5 provide support for Hypothesis 1. The coefficients for Co-Partisan Control are positive and significant in both models. And the values of the coefficients are quite similar in both models. The results suggest that in both 2008 and 2016, as the number of levels of government a respondent's preferred party increased by 1, they were about 0.2 units more satisfied with democracy.
Next, as in the previous analyses, we analyze the effect of State Level Only control, Federal Level Only control, and Both Levels of control against No Control (the excluded category) on Satisfaction with Democracy. We present the results in  The results re-iterate that co-partisan control influences satisfaction with democracy, although the results were different for 2008 and 2016. In the 2008 model, experiencing at least one level of co-partisan control exerts a positive effect on Satisfaction with Democracy. Individuals who experience at least one level of co-partisan control were between 0.5 and 0.6 units more satisfied with democracy versus individuals who did not experience any level of co-partisan control. In contrast, in the 2016 model, only full co-partisan control had a positive effect on Satisfaction with Democracy. Individuals who experience full co-partisan control (in this case, Republican respondents living in states with Republican control) are about 0.4 units more satisfied with democracy, compared to those with no control. State only control (Democrats living in states with Democratic control) and federal control only (Republicans living in states with Democratic control) have no effect on satisfaction with democracy. The differences between the results for the 2016 models are striking. It appears that, based on the results from Models 11 and 12 and Figures 7 and 8, Republicans in 2008 were at least somewhat more satisfied with democracy if Republicans controlled their respective state governments versus no Republican control. But in 2016, they were only more satisfied with democracy versus no co-partisan control if they had full co-partisan control, meaning they lived in states controlled by Republicans. If they  The results for Democrats tell a slightly different story. In 2008, controlling at least one level of government led Democrats to be more satisfied with democracy versus no co-partisan control. Even if Democrats lived in states controlled by Republicans, they were more satisfied with democracy than individuals with no co-partisan control. But in 2016, Democrats living in states controlled by Democrats were equally satisfied with democracy as Democrats living in states controlled by Republicans. It is possible that at least in 2016, Democrats were generally dissatisfied with democracy due to the tumultuous nature of the 2016 presidential election. Even if they had representation at the state level, that did not increase their satisfaction with democracy.
Next, to ascertain whether strength of partisanship can alter the effect of Co-Partisan Control on Satisfaction with Democracy, we interact levels of co-partisan control with partisanship strength, as in the previous studies and present these results in Table 7 and graphically in a marginal effects graph in Figure 9 for the model with the 2016 ANES results.
The results from the 2016 ANES model provide support for Hypothesis 3. The results in Figure 9 suggest that as strength of partisanship increases, the effect of co-partisan control on satisfaction with democracy increases. While co-partisan control does not seem to influence satisfaction for leaning partisans, it does appear to have a positive effect on satisfaction for weak and strong partisans.
The results from the 2008 ANES do not suggest that strength of partisanship alters the impact Co-Partisan Control has on Satisfaction with Democracy. These results in conjunction with the Models 9 and 10 suggest that the specific effect that Co-Partisan Control had on Satisfaction with Democracy between these two elections changed. In the post-2008 election period, Leaning Partisans, Weak Partisan, and Strong Partisans were all equally more satisfied with increased co-partisan control over government. But in the post-2016 election period, as strength of partisanship increased, the positive impact that Co-Partisan Control had on Satisfaction with Democracy increased. As in the previous studies, we also interact different levels of co-partisan control with strength of partisanship. We present the results in Figures 10 and 11 and Table 8. In the 2008 study, weak partisans are slightly more satisfied with democracy than strong and leaning partisans if their preferred party controls at least one level of government versus none. Substantively these results suggest that in comparison to Republicans who live in states controlled by Democrats, Weak Democrats are more satisfied with democracy if Democrats control both levels of government or the federal level only. Additionally, Weak Republicans are more satisfied with democracy if they live in states controlled by Republicans versus Republicans who live in states controlled by democrats. Last, for Strong Democrats, they are more satisfied with democracy if they experience control over both levels of government.
In contrast, in the 2016 study, having some level of control at the state level seems to matter a bit more for stronger partisans (i.e., Democrats that live in states controlled by Democrats), although the effect just passes the threshold from being statistically indistinguishable from zero. Substantively these results suggest that Weak and Strong Republicans who live in states with Republican governments are more satisfied with democracy than Democrats who live in states controlled by Republicans. Additionally, Strong Democrats who live in states controlled by Democrats may be slightly more satisfied with democracy than Democrats who live in states controlled by Republicans.
Taken together, these results suggest that in 2008, regardless of their partisan attachments, respondents were generally more satisfied with democracy as the

Conclusions
We presented evidence from two survey experiments and observational data collected soon after the 2008 and 2016 elections to demonstrate how the levels of co-partisan control of different levels of government influences Americans' satisfaction with democracy. As one's preferred party controls more levels of government, general satisfaction with democracy in the United States increases. This effect is heightened, especially, among stronger partisans. This work has important implications for scholars of both American and comparative politics. While these findings are interesting in the current American political context, future work could examine whether the findings are time bound or consistent over time. Additionally, our analysis could be extended to countries outside the United States where there are variations in the strength of partisan attachments and variations in the types of democratic political institutions. Such an analysis would provide further insight into how variations in the strength of partisan attachments combined with variations in the types of democratic institutions affect how election outcomes influence voters' satisfaction with democracy.
This research has both theoretical and real-world implications. Political competition at the state level is declining (Basham and Polhill 2005;Forgette, Garner, and Winkle 2009), suggesting that more states are becoming "solidly" Democratic or Republican. Our results suggest this may have consequences for attitudes toward democracy in general, where political minorities in these states may only achieve winning at the federal level, which may damage their long-term overall satisfaction with democracy. This suggests that competition at the state level where both parties are equally capable of winning office is essential for citizens' satisfaction with democracy.
These results additionally shed greater light on the effect of winning and losing elections. While considerable work has been done on national winners and losers, and the conditions under which this effect on satisfaction with democracy are heightened or muted (see Singh 2014), comparatively little work has been done at the subnational level. However, some of this work does suggest winning local elections in both Canada (Blais and Gelineau 2007) and Germany (Singh, Karakoc, and Blais 2012) does increase satisfaction with democracy, albeit at lower levels than federal victories. Given the amount of power state governments have in the United States, we believe we may observe larger effects of controlling state-level governments on satisfaction with democracy in this context because of that additional power in local political institutions.
Beyond examining election outcomes, future studies could also examine how winning and losing influences perceptions of government performance, especially among strong partisans (Christmann 2018;Quaranta and Martini 2016;Sirovátka, Guzi, and Saxonberg 2018). It is likely that full winners would be the least critical of elected officials while full losers would be the most critical. Nevertheless, based on the results of our analysis, it is also possible that individuals who are winners at one level of government, either state or federal, might be more critical than full winners but less critical than full losers.
In an increasingly polarized and ideologically sorted political environment, satisfaction with democratic norms, regardless of outcomes, is imperative for the future of American democracy. We find that election outcomes, quite troublingly, can exert large and important effects on how citizens view the democratic process, especially among the most ardent partisans. These findings suggest that robust competition in elections across both the national and state government, where winning and losing alternates between the two parties, may have positive outcomes for attitudes toward democracy, compared to long periods of single party rule.