A. The Role of Moral Externalities

How then might we frame the decision by federal courts to inject themselves into vertical politics? One way is using the lens of externalities. In the broader literature on federalism, many theories focus on economic externalities. These types of externalities—for example, how pollution in one state creates negative spillover effects in other states—are particularly salient in studies of fiscal federalism, which concern the provision of public goods across the national and state levels (Oates Reference Oates1972).

While certain types of judicial decisions—for instance, antitrust rulings—can create or mitigate economic externalities across states, decisions involving constitutional protections are more likely to implicate moral externalities (Janeba Reference Janeba2004; Cameron Reference Cameron2005). Some citizens may care not just about policy in their home state—that is, policy that directly affects them—but also policy in other states. This utility over out-of-state policy arises from moral concerns. A quintessential example in American history is slavery: Northern opposition was driven in part by disgust over the practice of slavery in Southern states, even though slavery was banned in the North (Graber Reference Graber2006). Moral externalities, of course, can run in both directions. A person opposed to abortion, for example, may suffer from abortion restrictions being minimal in other states, even if such restrictions exist in that person’s home state; conversely, a person who supports the right to an abortion and lives in a state with few restrictions may nevertheless perceive harm to women in other states with reduced access to abortions.

The presence of cross-state moral externalities means that the use of vertical judicial review may have a “unifying” function, if the distribution of preferences is such that national majorities may benefit from courts “reining in” outlier states. Whittington (Reference Whittington2007, 107) notes that “the Supreme Court has often used the power of judicial review to bring states into line with the nationally dominant constitutional vision.” Brown v. Board of Education, which struck down segregation statutes in multiple states, illustrates this function of federal courts. While there was substantial opposition to the decision, it was concentrated in the Southern states, which were directly affected; national majorities, on the other hand, favored the decision. Moreover, “northern voters were largely appalled” by many of the tactics used by officials in the Southern states to defy the Court’s decision (Klarman Reference Klarman2012, 187).

B. Courts and Status Quo Biases

A second way to frame the use of vertical judicial review is to consider the possibility that state policy may not match state majority opinion, in which case the decision by a court to alter policy may actually enhance representation. Such a disconnect may occur for several reasons. First, legislative majorities may understand the public’s desire for policy change but fear exercising such change for political reasons and thus may prefer to defer to courts and their use of judicial review (Graber Reference Graber1993; Lovell Reference Lovell2003; Lemieux and Lovell Reference Lemieux and Lovell2010). Second, the status quo biases inherent in the United States’ separation-of-powers system may actually lead to disconnect in the mapping from public opinion to policy, when the former moves and the latter does not (Lax and Phillips Reference Lax and Phillips2012; Kastellec Reference Kastellec2016). As a result, the status quo as established by statutes may lag behind public opinion, perhaps due in part to blocking by entrenched interests (Klarman Reference Klarman1997; Whittington Reference Whittington2005). As noted in Kastellec (Reference Kastellec2016, 5), “When evaluating judicial review with respect to the countermajoritarian difficulty, it is crucial to compare not just the correspondence between public opinion and judicial decisions but also how those decisions relate to the location of the legislative status quo.”

A. Preliminaries

I consider a federation of S states, where s denotes individual states. Each state has a continuum of voters, who are denoted by i; let i, s denote voter i in state s. It is useful to denote separate levels of aggregation within the government. Let N denote the set of all voters in all states, or the national district. Let C ₁, C ₂, …, C_k denote the set of circuits within the government, of number k. The name “circuit” is derived from the US courts of appeals, which are divided into geographic units. More generally, circuits can be thought of as aggregations of two or more states. Each circuit consists of a subset of the set of all states; these subsets are both exhaustive and mutually exclusive (i.e., every state appears in one and only one circuit). Let $a\in \left\{N,C,S\right\}$ generically devote a given level of aggregation.

Voters have single-peaked preferences over a one-dimensional policy space $x\in \left[0,\overline{x}\right]$ . Rather than a standard rendering of a one-dimensional spatial model, the policy space here is best characterized as the “amount of protection” for a specified activity by an individual. In the abortion context, for example, a policy of 0 would mean a complete ban on all abortions under any circumstances, whereas $\overline{x}$ would mean total protection of a woman’s right to obtain an abortion under any and all circumstances. Let x_s denote the implemented policy in a given state. The policy space can encompass a wide range of issues, including issues in which ideological conservatives would prefer greater protections. For an example with reverse ideological polarity, consider the issue of gun rights; a policy of 0 would mean a complete ban on individual ownership of any firearms, while a policy of $\overline{x}$ would mean unrestricted access to any and all firearms.

Denote the ideal point of voter i in state s as t_is ; voters prefer state policies that are closer to their ideal point. For now, assume no externalities or spillovers from policies in other states. Assuming linear loss, the utility function for a voter can be stated as

(1) $\begin{array}{}\text{(1)}& U_{\mathrm{is}}\left(x\right)=-\left|t_{\mathrm{is}}-x_s\right|.\end{array}$

It is useful to summarize the preferences of the median citizen at each level of aggregation in the government. Let m_s denote the median citizen’s ideal policy in state s, m_c denote the median citizen’s ideal policy in circuit c, and m_n denote the “national” median voter. Let m_a generically refer to the median voter at level of aggregation a.

B. The Structure of Federal Floors

In formal theories of federal mandates, state policy results from the interaction of preferences aggregated nationally as well as preferences in individual lower units (see, e.g., Crémer and Palfrey Reference Crémer and Palfrey2000). Judicial federalism and the power of judicial review simplifies matters a great deal, since courts can act unilaterally to change policy. Let FC denote a federal court, where SC denotes the Supreme Court; CC _k denotes a circuit court in circuit k, and DC _s denotes a district court in state s. With slight abuse of notation, denote the ideal policy of each court by this same notation. Thus, SC denotes the Supreme Court’s ideal federal floor.

In the absence of federal court action, states are unconstrained and free to choose to set policy anywhere in the policy space $x\in \left[0,\overline{x}\right]$ . A federal court then sets a federal floor, denoted F_a . With F_a in place, the set of allowable policies shifts to $\left[F_a,\overline{x}\right]$ .

If implemented, F_a applies to the jurisdiction of the court that implements it. For example, a floor set by a district court F _DC would apply only to the state in which the district court is located; a floor F _CC would apply to all the states in a given circuit; and a floor F _CC applies nationally. Under the norms of strict vertical stare decisis, federal floors set by higher courts supersede decisions by lower courts. The utility of a federal court is ${\sum }_s\left|t_a-\mathrm{FC}\right|$ —that is, the average distance between its ideal point and all state policies that fall under the court’s jurisdiction.

C. Policy Implementation in the Absence and Presence of Federal Floors

For now, and following the basic model in Crémer and Palfrey (Reference Crémer and Palfrey2000), assume that state policy is selected via referendum. Thus, in the absence of a federal floor, state policy would be set at m_s , the median voter in a state. This result follows straightforwardly from the assumptions of single-peaked preferences and a single policy dimension (and the lack of any agenda control).

Now assume that a federal court decides to implement a federal floor. Given its utility function, a federal courts’s best strategy is simply to set F_a at its ideal policy. This is because a court cannot “lower” policy in states with ideal points that are higher than FC, so it has no incentive to set the floor higher than its ideal point. Similarly, setting the floor lower makes the court weakly worse off, since states with median voters between F_a and FC will set their policies in this interval (see lemma 2 in Cameron Reference Cameron2005).

With a federal floor in place, policies below F are now off-limits, and hence state policy is set at $x_s^{*}=\max \left\{F_a,m_s\right\}$ . This result is illustrated in figure 1: all states where the median voter prefers $x<F_a$ see policy “shifted” to F_a . These are the states affected by the establishment of the federal floor. Conversely, states where the median voter prefers $x\ge F_a$ are unaffected by the federal floor.

Result 1: The implementation of a federal floor only affects policies in states where the median voter is located below the federal floor. In the absence of cross-state externalities, the implementation of a federal floor also only affects voters in states where the median voter is located below the federal floor.

Figure 1.

Effect of a federal floor on state policy making

While straightforward, this result is fundamental to understanding the role of judicial review by federal courts in a federal system. The establishment of a federal floor cuts off the ability of some states—but not all states—to implement their preferred policy. In particular, those with a “low demand” for protections in a given issue are affected, while “high demanders” are not. This, of course, has important consequences for representation: it means that the ability of state policy to match state opinion when a federal floor has been set will be asymmetric.

D. Judicial Floors and Voter Utility

I now consider how the establishment of a judicial floor affects aggregate voter utility. Continue to assume that state policy perfectly matches the preferences of the median voter within each state. It is first useful to consider how the implementation of a judicial floor affects the direct policy utility of a voter within an affected state—that is, ignoring the role of any externalities. Again, the median voter in a state is the relevant benchmark: if $F_a\le m_s$ , the implementation of the floor does not affect policy, and voters’ utilities are likewise unaffected. Conversely, if $F_a>m_s$ , many voters lose from such a decision (since the floor is to the right of the median voter). Some voters, however, will benefit. Specifically, as seen in figure 2, there is a cutpoint $\left(m_s+F_a\right)/2$ that lies halfway between the median voter’s ideal point and the judicial floor. Voters to the left of the cutpoint do worse under the floor, while voters to the right do better. Not surprisingly, voters with ideal points more extreme than the median and the location of the floor have more to lose and gain, while centrist voters see smaller utility shifts. The total shift in voters’ policy utility can be expressed simply as

(2) $\begin{array}{}\text{(2)}& {\gamma }_s=\sum _{i,s}\left(\left|t_{\mathrm{is}}-m_s\right|-\left|t_{\mathrm{is}}-\max \left(F_s,m_s\right)\right|\right).\end{array}$

That is, the total shift is given by the distance between the voter and the median state voter (i.e., the “old” policy), minus the distance between the voter and the new policy, which is determined by whether the voter’s state is affected by the federal floor. By construction, γ _s cannot be positive: setting the floor at or below the state median voter does not affect voters’ policy utility, while setting it above induces an overall decline in utility.

Figure 2.

Winning and losing voters

When we move to federal decisions that affect multiple states (i.e., circuit courts or the Supreme Court), the effect of judicial floors on aggregate voter utility is subtler. The calculation of net shift in voters’ policy utility proceeds similarly, only now we consider all voters in level a:

(3) $\begin{array}{}\text{(3)}& {\gamma }_a=\sum _{i,s}\left(\left|t_{\mathrm{is}}-m_s\right|-\left|t_{\mathrm{is}}-\max \left(F_a,m_s\right)\right|\right),\forall s\in a.\end{array}$

The location of the state median voter is still the benchmark, but now the relevant comparison is a judicial floor that affects voters in multiple states. It is straightforward to see that γ _a also cannot be positive: the floor does not affect voters in states with median voters above the floor, but it induces a shift in policy away from the ideal point of voters who reside in states with median voters below the floor.

For illustration, consider the following numerical example. Suppose there are three states comprising a total of 99 voters, with 22 voters in state 1, 45 in state 2, and 32 in state 3. Let $\overline{x}=100$ . In each state, the ideal points of voters are drawn from a normal distribution; the mean ideal point of voters is increasing in each state. Specifically, the mean ideal point is 25, 50, and 75 in states 1, 2, and 3, respectively, with a standard deviation of 20 in each state (ideal points below 0 and above 100 are truncated at the limits). Figure 3A depicts these distributions, along with the location of the respective median voters in each state.

Figure 3.

Effect of a federal floor. A, Distribution of voters in a hypothetical three-state example. B–E, Aggregate shifts in utility as the floor is set sequentially from 0 to 100, under the scenarios described in each panel label. See text for further details.

Suppose these three states are in the same circuit—m_c depicts the overall circuit median—and a circuit court sets a floor F_c . Imagine the floor was set sequentially at every integer on x—figure 3B depicts the net utility shift from setting the floor at a given location on the horizontal axis. The dashed horizontal line indicates a net utility shift of zero—values above and below this line respectively indicate positive and negative shifts in voter utility, by comparing net voter utility before and after the introduction of a federal floor at each specific value. Figure 3B shows that there is no effect of establishing a federal floor until it hits just above the location of m ₁ (around 20), at which point the extreme low-demand voters in state 1 suffer a utility loss from the floor being set above m ₁. As the floor gets increasingly higher—particularly when it exceeds m ₂ and m ₃—more voters are increasingly affected. As the floor approaches $\overline{x}$ , only the most high-demand voters benefit, while all other voters suffer a utility decline.

E. Incorporating State Legislatures and Status Quo Policies

So far I have assumed that state policy is set via referendum. Thus, in the absence of judicial intervention, state policy would be set perfectly at the location of the state median voter. Now assume that state policy is decided by a representative legislature. One possibility is that legislatures perfectly reflect the will of state majorities. In Crémer and Palfrey (Reference Crémer and Palfrey2000), shifting from a referendum system to elected legislatures has a substantive impact on policy, as it results in the federal mandate being placed at the median of all median voters across districts, rather than the overall median of all voters. In the model here, however, it actually produces no meaningful differences, given that F_a is set by a court. If state legislatures perfectly represented the state median voter, then state legislatures would set policy at $x_s^{*}=\max \left\{F_a,m_s\right\}$ , and everything would be the same as above.

Now consider the scenario in which this is a potential mismatch between state policy and current state-level opinion. Let q_s denote the status quo policy of state s. Accordingly, $\left|m_s-q_s\right|$ represents the lag between what the current state median prefers and existing policy. Thus, there exists a range of federal floors that will improve upon the utility of the median voter; whether this is the case depends on the location of the state median relative to the status quo. Specifically, if $m_s>q$ , any floor that is to the right of q_s and to the left of $2m_s-q_s$ will benefit the median voter. Conversely, if $m_s<q_s$ , then the introduction or movement of a federal floor either has no effect or makes the state median voter worse off: the former occurs if $F_a\le q_s$ , the latter, if the floor is above q_s . Again, we see the asymmetric effect of federal floors, which is visualized in figure 4: they only improve the representation of voters who are “higher” on a given policy dimension.

Figure 4.

Effect of a federal floor on state median utility, depending on whether the median is to the right of the state status quo (top) or the left of the status quo (bottom).

Returning to the question of how federal floors affect aggregate voter utility, we can restate equation (3) but with q_s replacing m_s :

(4) $\begin{array}{}\text{(4)}& {\gamma }_a^{\mathrm{sq}}=\sum _{i,s}\left(\left|t_{\mathrm{is}}-q_s\right|-\left|t_{\mathrm{is}}-\max \left(F_a,q_s\right)\right|\right),\forall s\in a.\end{array}$

To see how the inclusion of lagging status quos may affect representation, I return to the numerical example in figure 3. Whereas previously state policy defaulted to the location of the state median voter, now assume that the status quo in states 1 and 2 lags “below” the location of the median: specifically, assume $q_1=15$ and $q_2=20$ ( ${\mathrm{ms}}_1=21$ and ${\mathrm{ms}}_2=49$ ). Figure 3C presents the shift in net voter utility, at every level of the federal floor. Because many more voters in states 1 and 2 benefit from even floors well above their ideal point (in contrast to what we saw in fig. 3B ), there are now regions (indicated via shading) in which the establishment of a federal floor increases voter utility—particularly in the region between $q_2=20$ and just above ms ₂; floors in this range benefit many voters in state 2 (and state 1 as well) while not affecting voters in state 3.Footnote ⁴

F. Adding Externalities

Now let us introduce the possibility of cross-state externalities in voters’ utility functions. To motivate this possibility, I modify the voters’ utility function as follows:

(5) $\begin{array}{}\text{(5)}& U_{\mathrm{is}}\left(x\right)=-{ɸ}_s\left|t_{\mathrm{is}}-x_s\right|-{\alpha }_{\mathrm{is}}\sum _{j\ne s}{ɸ}_j\left|t_{\mathrm{is}}-x_j\right|,\end{array}$

where $\alpha \ge 0$ denotes the sensitivity of voter (i, s) to policy in states $j\ne s$ , and ${ɸ}_s\in \left(0,1\right)$ denotes the relative size of state s, with ${\sum }_{s\in S}{ɸ}_s=1$ . Thus, externalities for an individual voter are weighted by the distance between her ideal and policy in all other states, as well as the relative size of each state.

Given this utility function, the total shift in voter’s utility after the introduction of a federal floor can be expressed as follows:

(6) $\begin{array}{}\text{(6)}& \begin{array}{cc}{\gamma }_a^e=& \sum _{i,s}(\left|-{ɸ}_s\left|t_{\mathrm{is}}-x_s\right|-{\alpha }_{\mathrm{is}}\sum _{j\ne s}{ɸ}_j\left|t_{\mathrm{is}}-x_j\right|\right|-|-{ɸ}_s\left|t_{\mathrm{is}}-\max \left(F_a,m_s\right)\right|\phantom{\sum _{j\ne s}}\\ -\sum _{j\ne s}{ɸ}_j\left|t_{\mathrm{is}}-\max \left(F_a,m_s\right)\right|\left|\right),\forall s\in a.\end{array}\end{array}$

While this expression is more complicated, the basic intuition is the same: voters compare their utility under the absence of a federal floor to that under the floor, which will now affect their utility gained or lost from policies being moved in states affected by the floor, even if their state is unaffected directly by the introduction of a floor.

How does the introduction of externalities affect the overall picture of voter gains and losses? Returning to the numerical example, figure 3D depicts the shift in aggregate voter utility as the federal floor is set increasingly higher. To highlight the importance of externalities, figure 3D returns to the assumption of perfect representation (i.e., state policy is set at the location of the median state voter), as assumed in figure 3B . For simplicity, I assume that α _is is uniformly distributed between 0 and 1 and that ɸ _j is based directly on the number of simulated voters in each state.

Figure 3D shows that the introduction of “moderate” floors—here in the range of roughly 25–70—induces positive net voter utility shifts (again indicated by the shaded region). This is because even though the setting of a floor has no in-state direct effect on many voters’ policy utility, high-demand voters now reap the gains from low-demand states being forced to set higher policies, thereby moving out-of-state policy closer to the high demanders’ ideal points. This shift, of course, induces negative externalities for low-demand voters (these externalities are on top of their in-state policy loss). However, in this example, only when the floor exceeds 70 or so (i.e., above m ₃) do the overall utility losses from the floor exceed the benefits. This example illustrates the classic case of courts bringing “outlying minorities” in line with national majorities, as voters who favor higher protection see federal courts move policy closer to their ideal points in states where the median voter prefers lower levels of protection.

What happens if we allow for both lagging status quos and the presence of cross-state externalities? Figure 3E combines the analyses in figures 3C and 3D . That is, the status quos lag behind the median voters in both state 1 and state 2, and the same distribution of externalities analyzed in figure 3D is assumed. Figure 3E reveals that, in this scenario, utility gains exist from the setting of federal floors at every level except very low levels (where they have no effect) and very high levels (beyond the ideal point of the median voter in state 3).

G. Defining a Countermajoritarian Decision

Taking the analyses of federal floors as a whole, the judicial federalism framework directly facilitates a definition of whether a decision is countermajoritarian. We can simply look at the effect of an introduction of a federal floor and evaluate its effect on overall voter utility. Recall that by construction, γ _s cannot be positive; this means that when only direct policy utility is considered, and if perfect representation of the median voter is assumed, a federal court decision to establish a floor must be countermajoritarian (as long as the floor affects at least one state). However, allowing for lagging status quos or externalities changes things, as both ${\gamma }_a^{\mathrm{sq}}$ and ${\gamma }_a^e$ can be positive. If they are positive, then more voters are benefiting from the court’s setting of a federal floor—such a decision is thus promajoritarian. If ${\gamma }_a^{\mathrm{sq}}$ or ${\gamma }_a^e$ is negative, however, then the decision is countermajoritarian. This leads directly to this result:

One upshot of this result is that it is not straightforward to evaluate whether a decision by a federal court is pro- or countermajoritarian. Such determinations must be made in light of both the location of the status quo in each state and one’s assumptions about how voters evaluate out-of-state policy.

H. State Courts

In this section, I introduce state courts, which comprise the final piece of the judicial federalism framework. Under the supremacy clause of the US Constitution, state courts are obligated to respect floors set by the Supreme Court, but they can also interpret their own state constitutions as providing greater protection (Devins Reference Devins2010). Let SF _s denote a floor set by a state court in state s, where ${\mathrm{SF}}_s\ge F_{\mathrm{SC}}$ .Footnote ⁵ If a state court sets a floor, then policy is set at $x_s^{*}=\max \left\{{\mathrm{SF}}_s,m_s\right\}$ .

This aspect of the model speaks directly to the so-called new judicial federalism (see, e.g., Devins Reference Devins2010). As the Burger and Rehnquist courts turned rightward and limited the Supreme Court’s interpretations of individual rights, liberals turned to state courts for judicial expansion of such rights. In essence, they were asking state courts to set constitutional floors far above the level the US Supreme Court was willing to extend.

One implication of the fact that state floors can be above but not below the federal floor is that state courts should be more active in areas of the law with lower federal floors, since litigants are less likely to bring those cases to federal courts. Even when the Supreme Court has ruled a restriction constitutional, based on the justices’ understanding of the federal Constitution, state judges may still find it unconstitutional under their state constitution.

B. Initial Public Opinion and State Policy

Figure 5 depicts the estimates of public opinion in each state from 1993 to 2015; the shaded regions depict 95% confidence intervals. In each plot, the vertical axis depicts the estimated percentage of residents in each state who support legalized gay marriage. The states are sorted from lowest to highest support, on the basis of the 2015 estimates of opinion. In line with the trend in national support, the figure shows an increase in support in every state over this period. However, both the base rate of support (i.e., the intercept) and the slope of change differ significantly across states. At the low end, for example, support in Mississippi rose from 13% (95% confidence interval of [8%,18%]) in 1993 to 35% [30%,40%] in 2015. At the high end, support in Massachusetts started at 45% [38%,52%] and rose to 77% [75%,80%]. Early in this period, however, public opinion was decidedly against gay marriage, even among the states that would eventually come to support it. The earliest estimated support for gay marriage to cross 50% in any state was in 2002 (in Massachusetts); the median year across states for crossing the 50% threshold (including states still below it as of 2015) was 2012.

Figure 5.

State-level support for gay marriage, 1993–2015. States ordered by increasing support as of 2015. Dashed lines indicate when bans were passed (before 1993 depicted at 1993). Dot-dashed lines indicate when and how gay marriage was legalized. Leg/Ref = via legislation or referenda. SSC = state supreme courts. Dist. = federal district court. CoA = federal courts of appeals. USSC = US Supreme Court.

Next, the dashed lines in figure 5 depict the year in which a state first implemented a ban on gay marriage, through either statutory or constitutional means. (The dot-dashed lines depict the year and method by which gay marriage was legalized; I return to this below.) The earliest ban was enacted in 1973. (Eight states implemented bans before 1993; for convenience in fig. 5 I indicate their passage year at 1993.) The latest year of initial implementation was in 2006. Figure 5 shows that for every state that implemented a ban, public opinion at that time opposed legalized gay marriage, and thus these bans matched public policy with majority opinion—this can be seen by examining the intersection of the opinion estimates with the dashed lines, which always occurs below the 50% level of opinion.Footnote ¹² Thus, the only mismatches in representation occurred in the small number of states that did not pass bans on gay marriage. However, as discussed in footnote 11, even in these states same-sex couples did not enjoy equivalent marriage rights to opposite-sex couples, and thus policy in effect matched public opinion.

If public support for gay marriage had remained at the low levels seen in the 1990s, the implementation of a federal floor (or a state court striking down gay marriage) would have been highly countermajoritarian, given that a majority of voters in no states supported it. Indeed, this fact helps explain the extensive backlash to the Hawaii Supreme Court’s decision in 1993 and the fear that courts would impose same-sex marriage on the country. In terms of figure 1, such an imposition would mean that every state (in the relevant jurisdiction) would be affected by the federal floor. However, public opinion on gay marriage shifted rapidly in the 2 decades that followed, leading voters, legislators, and courts to respond in different ways.

C. Low Federal Floors and State Court Rulings

In this section, I evaluate the role of state courts in adjudicating challenges to existing bans on gay marriage. Recall that a consequence of a low (or nonexistent) federal floor is that litigants should be more likely to bring challenges in state courts. In line with the overall trajectory of the issue over the past 3 decades, the US Supreme Court initially established a very low level of protection for gay rights. In 1972, the Court dismissed an appeal of a decision by the Minnesota Supreme Court that upheld the denial of a marriage license to a same-sex couple (Baker v. Nelson, 409 U.S. 810). In 1986, the Court upheld the constitutionality of state bans on sodomy (Bowers v. Hardwick, 478 U.S. 186); this precedent led lower federal courts in subsequent years to reject challenges to other forms of discrimination against gays, such as employment discrimination (Klarman Reference Klarman2012, 38). Ten years after Bowers, the Court issued its first decision in support of gay rights, striking down a Colorado constitutional amendment that prohibited any type of government action intended to protect gay persons against discrimination (Romer v. Evans, 517 U.S. 620). Then, in 2003, the court overruled Bowers but did not address the constitutionality of same-sex-marriage bans (Lawrence v. Texas, 539 U.S. 558).

As a result, until very recently, litigants and interest groups seeking to challenge gay marriage bans had little reason to believe that federal courts would be receptive to their claims. Thus, early efforts focused on state courts in more liberal states (such as Hawaii, Alaska, Vermont, and Massachusetts) where judges might be more receptive to striking down bans based on their state constitutions (Klarman Reference Klarman2012, 55, 89).

Public reaction to these decisions was not monolithic. For example, unlike in Hawaii and Alaska, there was little backlash against the Vermont Supreme Court’s 1999 decision mandating that the state provide same-sex couples the same rights and benefits as married couples (Hume Reference Hume2013, 75). Nevertheless, these early decisions (particularly the 2003 Massachusetts decision) in support of gay marriage rights, combined with the Supreme Court’s decision in Lawrence v. Texas in 2003, helped spur a movement among gay marriage opponents to push for amendments to state constitutions banning same-sex marriage. Such bans had been passed in 4 states before 2003; in 2004 alone, voters in 13 states approved referenda banning same-sex marriage, and by 2012, 31 states would amend their constitutions. These amendments effectively made state courts off-limits to challenges seeking legalization in those states. In the terms of the model, these amendments meant that state supreme courts could not interpret their constitutions as providing protections above the federal floor (which at the time was also 0), since the amendments explicitly ruled out such interpretations.

As a result, state supreme courts only heard challenges in a subset of the states that did not implement such constitutional bans. Unsurprisingly, public opinion was relatively liberal in these states. Figure 6 depicts the states in which state supreme courts ruled on gay marriage bans. The points (which are jittered vertically) depict the level of public opinion in that state in the year (indicated by the state-year label) in which a decision was issued; the horizontal lines indicate 95% confidence intervals. Decisions in which the court upheld (struck down) the ban are on the bottom (top) of the graph; the boldfaced labels indicate decisions in which the Court did not fully mandate the legalization of same-sex marriage, instead requiring the state to implement the functional equivalent (i.e., civil unions).

Figure 6.

State supreme court decisions on gay marriage bans. The points depict estimated state-level support, in the year a court issued a given decision, with 95% confidence intervals indicated by the horizontal lines. Decisions in which the court upheld (struck down) the ban are on the bottom (top) of the graph. Boldfaced labels indicate decisions in which courts did not fully mandate the legalization of same-sex marriage but required the state to implement the functional equivalent (i.e., civil unions).

How well did the decisions of state judges align with state-level public opinion? A total of 11 challenges were heard in 10 states (the New Jersey Supreme Court heard challenges in both 2006 and 2013). As figure 6 shows, in all but 3 states in which supreme court judges either struck down state bans on gay marriage or mandated the functional equivalent of marriage (i.e., civil unions), public support for legalization was either close to 50% or well above it. As noted earlier, the Hawaii Supreme Court’s 1993 decision was highly unpopular, with just 32% [21%,44%] favoring same-sex marriage at the time. While less countermajoritarian, the Vermont Supreme Court’s 1999 decision placed the court ahead of public opinion. Finally, only an estimated 42% [38%,45%] supported gay marriage in Iowa when its supreme court legalized gay marriage in 2009.Footnote ¹³ Conversely, state judges in Maryland, Washington, and New York rejected challenges. Thus, while state judges in only a relatively small number of states interpreted their constitutions as providing protections above the federal floor in this period, in these states the judges were usually aligned with public opinion.Footnote ¹⁴

D. The Introduction of Federal Floors: Changing Public Opinion and Lagging Status Quos

While the first federal challenge to a state ban on gay marriage was heard in 2005 (excluding the case of Baker v. Nelson in 1971), it was not until after the US Supreme Court decided Windsor in 2013 that challenges reached federal courts in large numbers. The 16 states in which legislators, voters, or state judges had implemented same-sex marriage would, of course, be unaffected by the introduction of a floor, since policy was already above it. All told, federal courts heard 44 challenges to policies in these 34 states; the overlap is due to challenges for a particular state being heard by both the district court and the court of appeals (and, in one case, also at the US Supreme Court). Thirty-six of these decisions resulted in invalidations of state bans, while eight decisions upheld them as constitutional.

How can we evaluate how federal decisions affected all other states in light of the theory of judicial federalism? First, setting aside for now any cross-state externalities, we can ask how countermajoritarian was each decision. As shown in figure 3B, if state policy perfectly matched the preferences of the state median voters, then the introduction of a federal floor would necessarily be countermajoritarian (even as voters “above” the median in states with bans would benefit). The existence of lagging status quos, however, means that federal floors can be welfare improving in some states.

How prevalent were such lagging status quos? Even as voters in many states approved constitutional bans on gay marriage in the middle of the 2000s, public opinion was swiftly moving in the direction of support for legalization. At the same time, the passing of such bans not only prevented state courts from interpreting their state constitutions, but they also helped to make this status quo more difficult to change in many states, as doing so would require a constitutional amendment and not regular legislation (i.e., a statutory change). Thus, in many states, the shift in public opinion created the “lagging status quo” scenario seen at the top of figure 4.

Figure 7 illustrates the relationship between public opinion and the status quo. The horizontal axis depicts estimated state-level support. The vertical axis distinguishes states where voters or legislators legalized gay marriage (this occurred in 11 states) from states where bans were kept in place—until they were eventually struck down by a federal court. Beginning with the top of the graph, the points and labels give the level of opinion in the year in which each state implemented legal gay marriage. (For example, the Vermont legislature approved same-sex legislation in 2009, at which point an estimated 59% [53%,63%] of state residents favored it.) Recall from figure 5 that in every state a majority of residents did not support gay marriage early on; however, figure 7 shows that by the time of legislative or voter implementation, public support was well above 50%. Thus, in these states, policy was shifted to match the preferences of the state median voter as those preferences moved away from the status quo; in terms of figure 4, q_s and m_s became aligned again.Footnote ¹⁵

Figure 7.

State support for gay marriage versus decisions by voters/legislators to legalize, versus implementations by federal courts. See text for details.

Next consider the bottom of figure 7, which depicts states that maintained their bans (until federal courts struck them down); states in which state courts implemented same-sex marriage are not included. The state labels indicate the levels of estimated opinion in the year federal courts acted (the height of the labels are jittered to make them easier to read); the first federal judicial invalidation of a state ban came in 2010 (California), while the last came when the US Supreme Court decided Obergefell in 2015.Footnote ¹⁶ Comparing the top and bottom of the figure, it is clear that the states where federal courts eventually struck down bans are ones that tended to have lower support for gay marriage. However, in many states, existing policy lagged behind the movement in public opinion. At the time of federal judicial intervention, fully 21 states had opinion majorities in favor of same-sex marriage where bans nevertheless remained in place. While support in some of these states (e.g., Georgia and Missouri) was only just above 50%, in states like Colorado, Oregon, and Nevada, sizable majorities in favor of same-sex marriage existed. Thus, the judicial implementation of gay marriage in these states was promajoritarian, as these decisions removed the lag in the status quo and brought policy in line with public opinion.

E. The Role of Cross-State Externalities

To this point, the analysis has set aside the possibility of cross-state externalities and how they might affect the relationship between public opinion, state policy, and federal floors. However, as is the case with many hotly contested social issues, the issue of moral externalities has been front and center in the fight over gay marriage. This has been particularly true with respect to opponents of same-sex marriage, many of whose arguments have focused on the deleterious effect that legalization might have on broader societal norms (Hatzis Reference Hatzis2006). Allen and Price (Reference Allen and Price2015, 146) summarize these arguments (without necessarily endorsing them) as follows:

Such beliefs help explain the extent of the backlash to the Hawaii Supreme Court’s decision in the 1990s. On the one hand, the scope of the backlash is surprising, given that the strength of public opposition to gay marriage at the time made it unlikely that legislators or judges in other states would follow suit. On the other hand, if opponents of gay marriage suffered negative externalities from the potential implementation of gay marriage in even a single state, the reaction makes more sense.Footnote ¹⁷

In addition, in this period, the issue of gay marriage was likely less salient among some supporters of gay marriage, relative to opponents. Posner (Reference Posner2003, 36), for example, argued that “it is apparent from [the state of the law and public opinion at the time] that the opponents of homosexual marriage feel much more strongly about the issue than the supporters … [and] are passionate in their opposition because they feel deeply threatened by the proposed change in the concept of marriage.” Accordingly, the implementation of a federal floor in 2003 would have likely induced significant negative externalities for opponents of gay marriage that would have outweighed positive externalities for supporters. However, as opinion has evolved toward greater support for gay marriage, it seems likely that the sensitivity of supporters to out-of-state policies has grown concomitantly. One reason is that more Americans now report having family members or friends who are gay (Pew Research Center 2015).

How, then, should externalities factor into the analysis of the introduction of federal floors for gay marriage? First, as discussed in Section II.B, one way to place the issue of cross-state externalities within debates over the countermajoritarian dilemma is to think about the Supreme Court specifically (and federal courts more generally) “using the power of judicial review to bring states in line with the nationally dominant constitutional vision” (Whittington Reference Whittington2007, 107). While it would be overreaching to say that a “dominant constitutional vision” had emerged by the time federal courts started striking down state laws, a clear national majority in favor of same-sex marriage was in place.

Which states, then, were “pulled in” by the federal courts? Figure 8A plots the estimated support for gay marriage in each state, as of 2015, against the size of each state (based on the 2010 census). The dashed horizontal line depicts the 50% mark, while the solid line depicts national support (62% in 2015). The states in plain text are those in which federal courts implemented gay marriage by striking down state bans, while the states in italic text already had gay marriage in place (via either legislation/referenda or state court decisions).

Figure 8.

Role of externalities in federal courts’ implementation of gay marriage. A, Estimated support for gay marriage in each state versus the size of each state. The dashed line depicts the 50% mark, while the solid line depicts national support (62% in 2015). The states in plain text are those in which federal courts implemented gay marriage by striking down state bans, while the states in italic text are states that already had gay marriage in place (via either legislation/referenda or state court decisions). B, Simulations of the effect of voter sensitivity to out-of-state policy on overall voter welfare. C, Simulated utility over time. See text for details.

Thus, using figure 8A , we can separate the “in state” versus “out of state” effects of the introduction of federal floors. With respect to the former, as discussed earlier, the striking down of bans in states where policy lagged behind changes in public opinion (i.e., the plain-text states above the 50% line) was promajoritarian. Conversely, the implementation of gay marriage in states where opinion majorities opposed gay marriage—all of which had bans in place—was countermajoritarian. With respect to externalities, even in states where gay marriage was already in place, the introduction of federal floors induced positive externalities for supporters in these states, even though the actions of federal courts had no effect on their own state policies. Conversely, the implementation of gay marriage by federal courts created negative externalities for opponents in every state, on top of any in-state policy loss suffered where bans had been in place. Finally, while there is only a weak relationship between state size and state-level public opinion, most of the states where opinion majorities opposed same-sex marriage are below the median state in terms of population; furthermore, many of the states are in the South.

How can we weigh these positive versus negative externalities? Recall that, in the framework, α denotes the sensitivity of a voter to policy in other states. It seems unlikely that a direct measure for this quantity can be found—at least one that adequately places the sensitivity of both supporters and opponents on the same scale (as well as their joint distribution). Accordingly, I use a simulation approach to illustrate how the introduction of federal floors benefited and harmed supporters and opponents of gay marriage, and to examine the aggregate impact of these decisions.

I perform two simulation analyses. First, I create a simulated “voter-level data set,” with the number of voters in each state proportional to the actual state population (as of 2010). Then, for each “voter,” I simulate support for gay marriage, using a binomial distribution based on the estimated level of support in 2015.Footnote ¹⁸ Next, I define κ as the ratio of the sensitivity of supporters (s) of gay marriage to the sensitivity of opponents (o) (i.e., ${\alpha }_s/{\alpha }_o$ ); thus, if $\kappa =.5$ , supporters and opponents weigh out-of-state policy equally.Footnote ¹⁹ For every value of κ or each voter, I then compare the “utility” before and after the introduction of a federal floor (based on eq. [5] above), taking into account the status quo in the respective states. I then calculate the overall total utility for each level of κ; I assume that if policy matches their preferences, they receive a benefit of 1 and vice versa.

The results are presented in figure 8B . The horizontal axis depicts κ. The vertical axis depicts the simulated net utility shift from the introduction of a federal floor—similar to figure 3. The solid line indicates no net shift in utility from the introduction of the floor. Figure 8B shows that the degree to which the introduction of the federal floor is promajoritarian is directly increasing in κ. As supporters care more about out-of-state policy than opponents, they gain more from the introduction of a federal floor than opponents lose. Due to the assumptions of the simulation, and the distribution of overall opinion in 2015, the “crossover” point at which the decision turns from counter- to promajoritarian occurs when $\kappa ~.4$ .

Next, I perform a simulation analysis over time. For each year from 1993 to 2015, I again create a simulated voter-level data set, with both the proportion of voters across states and the distribution of opinion based on the actual levels in each year. Then, in every year, I calculate the overall utility shift from the introduction of a federal floor, again taking into account the status quo in each state in each year (thus, the simulations incorporate both externalities and lagging status quos). I do these calculations under three assumptions: first, voters have no cross-state externalities ( $\alpha =0$ ); second, supporters and opponents have positive but constant sensitivities (I assume $\alpha =.5$ ); third, the ratio of sensitivity for supporters versus opponents is increasing over time as a function of the distribution of national opinion in a given year.

The results are presented in Figure 8C . Each version of α shows the same substantive pattern: the introduction of a federal floor would have been countermajoritarian until a national majority favored same-sex marriage, which occurred in 2011.Footnote ²⁰ Of greater interest, however, is that the existence of externalities has a “multiplier” effect. Under the assumption of both constant externalities and (especially) correlated externalities over time, the introduction of a federal floor early on would have been drastically more countermajoritarian compared to the assumption of no externalities, because most voters would have suffered losses from the introduction of same-sex marriage in their states as well as in all other states. Symmetrically, the introduction of a federal floor after 2011 means that many voters are doubly benefiting from the implementation of same-sex marriage both in their states and nationwide. Thus, the distribution of cross-state externalities has significant implications for how we assess the countermajoritarian nature of judicial review.

More broadly, the results in figure 8 allow us to place the implementation of gay marriage by federal courts in the larger context of federalism and the countermajoritarian difficulty. In his dissent in Obergefell, Justice Scalia argued as follows:

It is certainly true that federal courts cut off the prospect of gay marriage being legalized at the state level with the introduction of a federal floor for the right to marry. Yet, the results above show that in many cases state opinion majorities favored legalization and benefited from federal court intervention into the policy arena. In addition, the existence of cross-state externalities, combined with the national distribution of opinion, mitigates the force of Scalia’s countermajoritarian critique.Footnote ²¹