Theory

Why might judges wish to avoid being reversed? Reference CaminkerCaminker (1994) points to two important reasons involving judges' self interest: “(1) fear that their professional audience, including colleagues, practitioners, and scholars, will disrespect their legal judgments or abilities; and (2) fear that a high reversal rate might reduce opportunities for professional recognition and advancement (including promotion to a higher court or appointment to judicial or other commissions)” (1994:77–8, footnotes omitted). The proposition that the desire to protect reputations, promotion prospects, and even current positions encourages compliance by lower court judges has been incorporated into empirical work as well (Reference ElderElder 1987; Reference Hansen, Johnson and UnahHansen, Johnson, & Unah 1995).

Concern for their careers and reputations is not the only reason judges might dislike being reversed. Another result of reversal—one that has received more attention in recent years—is that it impedes judges' efforts to shape policy. At best, it results in a specific holding contrary to their wishes. At worst, a doctrine the lower court dislikes becomes binding policy throughout the higher court's jurisdiction (the entire United States, for the Supreme Court). Accordingly, fear of reversal plays the central role in lower court compliance in models where judges are viewed as acting rationally to further their policy preferences (Reference CameronCameron 1993; Reference Songer, Segal and CameronSonger, Segal, & Cameron 1994; Reference McNollgastMcNollgast 1995).

Thus, the “core assumption” in Reference McNollgastMcNollgast (1995) is that all relevant actors, including lower court judges, “act rationally to bring policy as close as possible to their own preferred outcome” (1995:1636). Judges' “compliance with legal doctrine results not from a preference for stability or an adherence to a norm; rather, it results from the Supreme Court's threat to enforce the doctrine. When the Supreme Court punishes lower courts, the punishment meted out is not a loss of prestige or reputation; rather, the punishment consists of the loss in utility to the lower courts from the substitution of the lower court's decision with the Supreme Court's ideal point” (Reference McNollgastMcNollgast 1995:1681).

Although the specifics of Cameron's (1993) model are different, his ultimate position is essentially the same: “[t]he prospect of review, remote as it is, is sufficient to motivate most judges” (1993:30), even if reversal imposes no costs other than deviation from judges' ideal policy points. There are dissenting voices. Reference BeneshBenesh (2002) has argued that the prospect of reversal is too remote to produce substantial compliance. Reference CrossCross (n.d.) offers a full-scale theoretical challenge to Cameron and McNollgast. Yet in recent years it has become increasingly common for scholars to treat judges as strategic pursuers of policy preferences. As this view continues to gain adherents, invocation of the fear of reversal as an explanation for compliance is sure to become even more common. Before this explanation gains the status of accepted truth, it is essential that it be subjected to empirical tests.

Empirical Evidence

As of now, we have very little evidence on the extent to which judges' decisions are influenced by the desire to avoid reversal.Footnote ³ For one thing, it is unclear how much judges really mind it. Some informed observers—on or off the bench—have claimed that judges dislike being reversed (Reference SchickSchick 1970:143–7; Reference HowardHoward 1981:140, footnote j; Reference SatterSatter 1990:227–35; Reference PosnerPosner 1996:224). But the court of appeals judges surveyed by Reference HowardHoward (1981) claimed not to care much. Furthermore, they ranked the “anticipated response of the Supreme Court” well down in a list of influences on their decisions. The district judges interviewed by Reference KitchinKitchin (1978) responded similarly when asked about the anticipated responses of their courts of appeals.

Turning from judges' attitudes to their behavior, the story is much the same. Reference Songer, Segal and CameronSonger, Segal, and Cameron (1994) did find that circuit court decisions deviating from the Supreme Court's preferences were more likely than others to be appealed, “a necessary condition for monitoring [by the Supreme Court] to be effective” (1994:681). But as the authors recognized, rational behavior by litigants only allows the Supreme Court to monitor effectively; it does not ensure that it will. More fundamentally, their finding provides no grounds for concluding that circuit judges react to or even pay attention to the behavior of either the litigants or the Supreme Court.

Reference Cross and TillerCross and Tiller (1998) and Reference Van WinkleVan Winkle (1997) approached the issue a bit more directly. Cross and Tiller examined circuit court cases reviewing agency interpretations of statutes in the wake of the Supreme Court's Chevron decision, which is generally viewed as directing lower courts to accord substantial deference to agency interpretations (Chevron v. Natural Resources Defense Council 1984). They uncovered a “whistle-blower” effect in these cases: judges were more likely to defer to interpretations that ran contrary to their own ideological leanings if there was an ideological split on the panel. At first glance, this finding seems attributable to fear of reversal. Judges in the majority appear to realize that if they were to follow their own preferences, the other judge might dissent, signaling a possible violation to the Supreme Court. Nevertheless, the authors concluded that this interpretation is untenable. The cases they studied were decided between 1991 and 1995, a period when the Supreme Court was conservative in its leanings. Since liberal decisions had a smaller chance of surviving Supreme Court review, liberal judges should have been more constrained than conservative judges by the presence of a potential whistle-blower. But the authors found the opposite.

Reference Van WinkleVan Winkle (1997) asked whether circuit judges sitting on panels seek to avoid being reversed by their own circuit en banc. He found that the likelihood that members of a circuit's ideological minority would vote consistently with their leanings increased with the number of ideological allies on the panel with them and the narrowness of the ideological divide on their circuit and that these judges were less likely than members of the circuit's ideological majority to dissent when they were outnumbered on a panel. Although these findings might be explained in other ways, it seems reasonable to interpret them as evidence that circuit judges try to keep from being reversed, in part by avoiding unwanted attention. However, it is not clear how strong the motivation or its effects are or whether they carry over to interactions with the Supreme Court.

Dependent Variable

To begin, we needed an educated guess as to whether the Supreme Court would uphold each challenged search if the case were before it. Fortunately, Songer, Segal, and Cameron have done most of the work for us. Drawing on Reference SegalSegal's earlier work (1984, Reference Segal1986), they constructed a model of Supreme Court search and seizure decisions based on fact patterns, e.g., where the search took place, how extensive it was, whether the searchers possessed a warrant. The model also took into account the growing conservatism of the Court from the beginning of the Burger years. Applying the coefficients from their model to the facts—including the decision year—of each circuit court case, we imagined that the search was before the Supreme Court and calculated the estimated probability that the Court would rule for the government, deeming the search valid. If the probability was greater than 0.5, we predicted that the Supreme Court would rule for the government; if it was less than 0.5, we predicted that it would rule for the defendant, invalidating the search. (The probability is not exactly 0.5 for any case.) The dependent variable took the value 1 if this predicted decision of the Supreme Court matched the actual decision of the court of appeals, 0 otherwise.

As previous findings would lead us to expect, the level of lower-court/higher-court consistency in our sample of cases was quite high. The actual decision of the court of appeals matched the decision that would be expected from the Supreme Court in 354 of 450 cases (78.7%).

Control Variables

In addition to the key explanatory variables to be described shortly, we included three control variables. The first two take account of circuit judges' ideology, an important influence on their decisionmaking (Reference GoldmanGoldman 1975; Reference HowardHoward 1981; Reference JohnsonJohnson 1987; Reference Songer and HaireSonger & Haire 1992; Reference Songer, Segal and CameronSonger, Segal, & Cameron 1994). Here its influence should be manifested through decisions that are less deferential, the less ideologically comfortable the judges are with the decision the Supreme Court is thought to prefer. We began by creating ideology scores for each panel, following the procedure set out by Reference Songer, Segal and CameronSonger, Segal, & Cameron (1994).Footnote ⁴ The panel scores ranged from about −0.3 to about 0.96, with higher scores indicating greater conservatism. We then created one ideological comfort variable for searches the Supreme Court would be expected to uphold and one for searches it would be expected to reject. The first, CA Ideology-Uphold, was equal to the panel ideology score if the Supreme Court would be expected to uphold the search and equal to zero if it would not. Because higher scores indicate greater conservatism—and presumably greater willingness to uphold a search—the effect of this variable should be positive. The second, CA Ideology-Reject, was equal to −1 times the original ideology score (so that now liberal judges score higher) where the Supreme Court would be expected to invalidate the search and equal to zero if it would be expected to approve it. Its effect, too, should be positive.Footnote ⁵

If circuit judges attempt to decide their cases as the Supreme Court would, they should be more likely to succeed where the Supreme Court's position is easier to identify. Whether or not they attempt to do so, their attitudes should be more likely to match those of the Supreme Court where the facts of the case are particularly one-sided. These two considerations provided the rationale for our third control variable, SC Predictability. We first subtracted 0.5 from the estimated probability that the Supreme Court would uphold the challenged search, then took the absolute value of this difference. The range of possible scores was 0 to 0.5, with higher values indicating a greater probability that circuit judges would correctly predict the Supreme Court's position or that, because of the details of the case, they would view it in the same way as the higher court. We expected this variable to have a positive effect on circuit court/Supreme Court consistency.

Key Independent Variables

Supreme Court Oversight of a Particular Circuit

As with any supervisor, we might expect the Supreme Court to keep a particularly close watch on subordinates who have shown a tendency to deviate from its preferences. Evidence gathered by Reference Lindquist, Haire and SongerLindquist, Haire, and Songer (2000) is consistent with this expectation.

Measuring a particular circuit's vulnerability to review is complicated. The key element is the recent treatment of the circuit's decisions by the Supreme Court. But simply counting the number of times a circuit has been reversed would seem inadequate, since it does not take caseloads into account. A circuit that decides five times as many cases as another should not feel threatened by the fact that it has twice as many cases reversed as the other circuit. On the other hand, simply dividing the number of reversals by the number of cases decided by the circuit would not be correct, either. This is because all circuits' reversal rates must have declined dramatically over time as their caseloads have grown. That variation across time is already measured by our fourth independent variable.

To measure variation across circuits, our solution was to (1) divide the total number of cases reversed or vacated by the Supreme Court (not counting memorandum decisions such as grant-vacate-remands) in a particular period (the one, two, or three terms leading into the current year) by the number of reversals or vacates suffered by a given circuit in that period, yielding the circuit's proportional share of reversals and vacates for the period; (2) divide the total number of cases decided on the merits by all circuits in the appropriate period by the number of cases decided by that circuit,Footnote ⁹ yielding the circuit's proportional share of all reviewable cases for the period; and (3) calculate the ratio of the circuit's share of reversals and vacates to its share of all reviewable cases by dividing the result from Step (2) by the result from Step (1). A ratio above 1.0 indicates that the circuit suffered a disproportionate number of reversals in the period just past.

Again, we examined the Supreme Court's activity in the term leading into the year of the circuit court case, the two terms leading into it, and the three terms leading into it. The third measure predicted Supreme Court review most successfully, and it is included in the models discussed below. Descriptive information for this and the other independent variables is presented in Table 1.

Table 1. Summary of variables (and what they measure) with descriptive statistics. Expected effect is positive for all independent variables except CA cases. DC=district court

Before turning to our models, we want to be sure that our assumptions are clear. We do not assume that judges pay attention to all of the specific measures just described. In some cases, we assume only that they are aware of the broader phenomena reflected in these measures. For instance, we doubt that circuit judges usually notice whether or not a district court opinion has been published, but we suppose that they tend to notice the same aspects of the case that told the district judge whether it was important enough to publish. Further, we are sure that both our estimates and judges' estimates of the threat of review contain error. For any one pair of cases, reversal-averse judges could anticipate the Supreme Court in the case we thought was less likely to be reviewed and not do so in the case we thought was more likely to be reviewed. This could happen either because our predictions overlooked a special element of a case that made it more or less cert-worthy or because the judges deciding it made a mistake in their calculations. But across several hundred cases, if fear of reversal is really driving decisions, broad tendencies should emerge. Cases with a higher likelihood of review should more often be decided consistently with the Supreme Court's preferences.

First Model—The Likelihood of Review

Testing the first model required that we have—for each case, viewed from the perspective of circuit judges who had not yet decided it—an estimate of the probability that the Supreme Court would eventually choose to review the case. Deriving an estimate entailed several steps. First, for each of the 450 cases, we coded whether it was eventually reviewed by the Supreme Court (1) or not (0). We thus created the dependent variable for the second step, in which we used logit to estimate the influences of the seven measures of cert-worthiness just discussed. One control variable was added—whether the circuit court decision was consistent with the predicted Supreme Court decision.

Results are presented in Table 2. The control variable and four of the key independent variables—published district court opinion, en banc at the circuit court, conflict, and circuit court case volume—had highly significant effects. The Supreme Court's appetite for cases and recent tendency to reverse a particular circuit had no discernible influence on its current review decisions.

Table 2. Logit model of Supreme Court review (1=case reviewed by Supreme Court; 0=case not reviewed, either because not appealed or because certiorari denied)

Next, we dropped the two insignificant variables from the logit and ran it again. Using the coefficients from this model, we calculated logit scores for each case based on the actual values of the four key independent variables, with the value for the consistency variable set at zero (inconsistent) for all cases. Our decision to include the consistency variable (our ultimate dependent variable) in this equation and then set its value at zero when calculating scores was necessary—it allowed us to obtain a predicted probability of review for each case based only on the direct effects of the key independent variables, with the effects of the lower court decision excluded. But since it may be the source of some confusion, we explain it more fully in the Appendix.

Once we had predicted logit scores for each case, the next step was to transform the scores into probabilities.Footnote ¹⁰ The result for each case was an estimate of the likelihood that the Supreme Court would agree to hear such a case if it went contrary to the decision that would be expected from the Supreme Court. This is the independent variable of interest in the first model. If fear of reversal was driving lower court deference to the Supreme Court, this variable should have had a positive effect on the likelihood that the actual court of appeals decision matched the predicted decision of the Supreme Court.

First Analysis—Effect of the Likelihood of Review

We turn to logit again to analyze this effect. The results, shown in Table 3, are quite dramatic. While all three control variables behaved just as expected, the coefficient for the review variable, expected to be positive, was in fact decidedly negative. Cases with characteristics making them more likely to be reviewed by the Supreme Court were less likely to be decided consistently with its apparent preferences.

Table 3. Logit model of effect of case's probability of being reviewed by Supreme Court on circuit court-Supreme Court consistency

Although the coefficient was very nearly statistically significant by the traditional standard, the effect uncovered was not large, as became apparent when we computed estimated probabilities of consistent decisions. For instance, if the other variables were held at their medians, moving from a 0.1 probability of review to a 0.5 probability decreased the probability of a consistent decision by only 0.06 (standard deviation=0.03).Footnote ¹¹ For this reason, we did not feel compelled to seek an explanation for the negative sign. The key point is that we did not find the effect we were looking for.

We recognize that judges, like other human beings, have “cognitive limitations that inhibit their ability to respond directly to stimuli whose complexity exceeds their computational capacity” (Reference Rowland and CarpRowland and Carp 1996:156). Perhaps our test requires too much of judges. Assessing the likelihood of review may simply be too hard, and circuit judges, while attempting to identify the cases where they need to proceed cautiously, may make mistakes. We can investigate this possibility by looking more closely at the probability of review.

Table 4 displays ranges for the predicted probability of Supreme Court review, given a noncompliant circuit court decision, along with the actual frequency of compliance in each range. It is most instructive to examine the extreme categories. Judges may well have had difficulty spotting the subtle distinctions between cases with a 0.35 likelihood of being reviewed and those with a 0.4 likelihood, but surely even cognitive misers would recognize that the kinds of cases falling in the 0.8–0.9 or 0.9–1.0 probability ranges (the last two rows) pose a far greater danger of reversal than those in the 0–0.1 range (first row). Did the disparity in danger affect judges' decisions in the predicted way? Obviously not. In the range of cases with the smallest threat of reversal, circuit judges were most likely to decide as the Supreme Court would be expected to—fifty out of fifty-two times. In the nine cases where the threat was greatest, they made four inconsistent decisions.

Table 4. Frequency of circuit court decisions consistent with and inconsistent with predicted decision of Supreme Court, by estimated probability of Supreme Court review for inconsistent decision

For a more rigorous version of this less demanding test, we can return to the full model. Instead of the original measure of the likelihood of review, we included a dichotomous measure, scoring cases at 0 if they had a predicted probability of review of 0.2 or less and at 1 if their predicted probability of review was 0.5 or higher. We omitted cases with predicted probabilities between 0.2 and 0.5. In other words, we had three categories of cases: one where the probability of review was quite low (trivial threat), one where it was even or better (serious threat, assuming judges are risk-averse), and one, excluded, where the threat was harder to judge. When we tested this model, the result for the threat of review was again fairly strongly in the wrong direction: coefficient=−0.64; standard error=0.35, p=0.065, N=286. Analyses using different cutoff scores and exclusions produced essentially the same results.

The next analysis, in which we consider the measures of cert-worthiness separately, may shed some light on this strange result.

Second Analysis—Individual Variables

Table 5 shows logit estimates for this model. Table 6 lists the predicted change in the probability of a consistent decision, moving from the minimum to the maximum value of the listed variable (from 0 to 1 for the dummy variables), with all others held at their medians. Leaving aside the control variables, one coefficient was both statistically significant and in the expected direction. Circuits that had been reversed particularly often in the preceding two years were more likely than others to defer to the Supreme Court. As Table 6 shows, other things being equal, a circuit with the least favorable two-year Supreme Court reversal score in the entire period of this study would be about 17% less likely to reach a decision inconsistent with the Court's preferences than the circuit receiving the gentlest treatment. Of course, this comparison is quite unrealistic, being drawn from the extreme ends of more than two hundred scores. To take a more realistic but still sizable interval, the difference in predicted probabilities for a court at the 20th percentile on this variable and one at the 80th percentile was about 0.04 (standard deviation=0.02).

Table 5. Logit model of circuit court-Supreme Court consistency, all cert-related variables included

Table 6. Estimated change in probability of consistent decision, moving from minimum to maximum value of listed variable, all other variables held at their medians. Estimates were generated through the CLARIFY program (Reference Tomz, Wittenberg and KingTomz, Wittenberg, & King 1998), using STATA 6.0

This finding does count as evidence that compliance stems from a desire to avoid reversal, but it is fairly weak evidence and it stands alone. The presence of conflict has about as strong an effect on consistency, but it is negative. The negative effect of an en banc hearing is far stronger still. As for the other three variables, one—the number of circuit cases reviewed by the Supreme Court in recent years—is incorrectly signed, though insignificant; the other two have no effect at all.Footnote ¹²

Interpretations

Because our findings run counter to a widely held view, it is important that we consider how they might be explained and what the explanations might suggest about the force of our conclusions. We can think of two possible ways to account for the weak or negative effects of en banc hearings, conflict, and district court publication. First, all three appear more often in cases involving unsettled issues of law than in routine cases. Judges may find it hard to predict how the Supreme Court would decide these cases. Second, district court publication, en bancs, and, to a lesser extent, conflicts, are more likely to be present in cases with substantial policy implications. In such cases, judges may be particularly reluctant to cede their policymaking power to the higher court.

Thus, it is logically possible that fear of reversal generally exerts a strong influence on lower court decisionmaking but is canceled out or even, in the case of conflict and en bancs, overwhelmed by judges' uncertainty about the Supreme Court's preferences or their desire to shape policy. On examination, though, this possibility does not appear credible.

We address the point about uncertainty first. One reason to doubt that uncertainty about the Supreme Court's preferences can adequately explain our findings is that many legal questions are unresolved only because they have not arisen previously; once they come up, it may be reasonably easy to predict how the Supreme Court would decide them. This can even be true of issues that engender conflict. The circuits may split because of ideological or other differences, even where the Supreme Court's position is discernible.

Second, even assuming that the Supreme Court's views on a particular doctrinal issue are difficult to read, we must avoid confounding uncertainty as to the Supreme Court's resolution of a particular legal issue and uncertainty as to its ruling on a search. As Reference SegalSegal (1984) wrote, in his original fact-pattern analysis of Supreme Court decisions.

The fact that Segal's model, incorporating only a handful of facts and ignoring the precise legal issues in the cases, was able to account for three-quarters of the Supreme Court decisions he analyzed demonstrates that the Supreme Court's decision whether to uphold a search is typically not very hard to predict, even in the tough cases that it chooses to hear.

Finally, recognizing that the fact-pattern model will not always yield a clear prediction for the Supreme Court, we included a variable measuring the strength of the prediction. We thus controlled for uncertainty about the Supreme Court's stance statistically.

Similarly, it is difficult to accept the notion that policy goals completely dominated an otherwise powerful desire to avoid reversal in the most cert-worthy cases. Note that in cases where the law is unsettled, circuit judges risk triggering an unfavorable precedent binding the whole nation, whereas in cases where the law is settled, the worst that can happen is an undesirable result in that case. Thus, while there is more of a chance to make law in the first type of case, there is also a much greater incentive to avoid being reversed.

Furthermore, judges can attempt to evade reversal while moving policy closer to their preferred points by ruling on a particular legal issue as they prefer to but, on the basis of some other issue, deciding the case as they think the Supreme Court would. This is not always possible, but cases often present enough legal questions to allow it. For these reasons, it is far from clear that the circuits' decisions should deviate from the Supreme Court's more often in cases involving unsettled issues than in others.

Finally, lack of certainty and policy goals can only help account for the findings as to conflict, en bancs, and district court publication. It is especially hard to see why, if fear of reversal were an important influence, we would find no effect for changes in circuit court caseloads. The circuits together decided more than seven times as many cases in 1990 (19,322) as in 1961 (2,681). Circuit judges could hardly have failed to notice the growth in caseloads, and they must have recognized the greater opportunity this gave them in later years to deviate from the Supreme Court's preferences while escaping review. Indeed, a positive relationship between caseloads and noncompliance is an explicit prediction of Reference McNollgastMcNollgast's (1995:1651) model. Yet our best estimate is that, in the average case, judges in 1990 were only 2% less likely than those in 1961 to anticipate the Supreme Court, and the standard error of that estimate is far too high to let us conclude confidently that the increasing caseloads affected judges at all.

In sum, while we believe that legal uncertainty and policy goals can help explain our findings, these explanations do not change our basic conclusion: fear of reversal was not a major influence on judges' decisions in the cases examined here. This is not to deny that the fear of reversal had any influence at all. Far less do we mean to suggest that it never affects judges' behavior. We strongly suspect that it does have an impact on some decisions and occasionally on other aspects of their behavior (such as the choice of language in their opinions). But the goal of this research was not to discover whether the desire to avoid reversal has any influence; rather, it was to determine how well that desire could account for the high rates of congruence between circuit court decisions and Supreme Court preferences. For search and seizure cases, at least, the answer seems clear: not well at all. Faithful agency on the part of circuit judges appears to flow from other sources.

Implications

There is nothing new about the proposition that fear of reversal plays a major, if not dominant role, in lower court decisions to defer to higher courts, but in recent years especially it has become common for scholars to invoke this proposition as an explicit premise (Reference Songer, Segal and CameronSonger, Segal, & Cameron 1994; Reference McNollgastMcNollgast 1995; Reference Cross and TillerCross & Tiller 1998; Reference BrentBrent 1999; Reference Spitzer and TalleySpitzer & Talley 2000; Reference Brace and LangerBrace & Langer 2001). Whether or not one accepts this proposition has crucial implications for one's understanding of judicial hierarchy. Because reversal is only a potential, not inevitable, consequence of a noncompliant decision, if the desire to avoid it were a major force behind compliance, then deference to a higher court would vary with the threat of reversal: other things remaining equal, compliance would be more likely where the probability that a noncompliant decision would be reversed was higher. In such a situation, a higher court's control over the development and application of the law in its jurisdiction would depend largely or entirely on its ability to maintain a credible threat of reversal.Footnote ¹³

Although we do not claim that our evidence conclusively disproves this proposition, we feel that it presents a serious challenge to it. Most immediately, the evidence suggests that higher courts can plausibly hope for substantial—though far from perfect—compliance and responsiveness even where the threat of reversal is weak. Since we are very unlikely to see either sustained decreases in lower court caseloads or increases in the work capacity of higher courts in coming years, this finding should be comforting both to higher court judges and to anyone who thinks it desirable for those courts to maintain some control over legal doctrine in their jurisdictions.

Our findings also have implications for how we understand judges' motivations and decisions. Here, our discussion is necessarily more speculative, but speculation is justified by the importance of the issues.

We will return to the main story of this article shortly. But first, it is important to emphasize two points that we have largely overlooked to this point: As in most studies, while we observed widespread compliance, we also found a substantial number of noncompliant decisions in our data set; and as in most studies, the measure of circuit judges' ideology was significantly related to their voting behavior. Our focus on compliance should not obscure what we take to be an incontrovertible finding of the empirical literature—that policy preferences exert a powerful influence on circuit judges' decisions. It is true that circuit judges are not Supreme Court justices; they encounter a different mix of cases, can more reasonably hope for advancement to higher positions, and must answer to a higher court. But the differences and their consequences should not be exaggerated. Advancement to a higher court is possible, but chances are very slim. Most cases will involve issues clearly covered by higher court precedents, but many will not. There is always the threat of reversal, but the threat will often be quite remote and, as we have argued here, may have only a limited effect. In other words, circuit judges will often operate in conditions similar to those of Supreme Court justices, and it would be odd if a factor that so greatly influences Supreme Court decisions had no effect in the circuits.

Ideology—more precisely, shared policy preferences among judges—could even help explain the observed congruence between actual circuit court decisions and predicted Supreme Court decisions. Circuit judges might decide as the Supreme Court would because their ideological leanings are similar. Ideological agreement must surely account for some decisional congruence. Yet considering the ideological divisiveness of search and seizure issues, the wide variation in ideology evident among judges, and the high levels of congruence found here and in other studies, we do not believe ideological agreement can explain all of the congruence. This doubt is reinforced by the fact that the findings of decisional congruence and responsiveness hold even when researchers control—admittedly imperfectly—for judges' ideology. (For the cases examined here, see Reference Songer, Segal and CameronSonger, Segal, & Cameron 1994).

Thus, while we believe that our study demonstrates the influence of ideology, we think it provides equally strong evidence that something other than ideology is at play in judges' decisions. One might reasonably ask, “What else?” A number of answers are possible, but we would offer two.

The first will probably seem less familiar but still deserves to be taken seriously. Over time, as the caseloads of circuit courts have become increasingly burdensome, judges have adopted a number of formal and informal shortcuts to cope with them (Reference PosnerPosner 1996; Reference Richman and ReynoldsRichman & Reynolds 1996). Instead of investing the time needed to weigh every detail of each case before them and consider the legal merits of every issue anew, judges might limit their attention to key facts or patterns and simply ask themselves, based on what they have observed in the past, how the case would normally be decided in the federal courts. In fact, it would be very surprising if judges did not do this often; the ability to simplify a judge's work is one of the arguments in favor of stare decisis. Since what has been done in the past is surely a good predictor of what the Supreme Court would do in the future, the desire to save time may promote decisional congruence even if judges do not care about congruence itself.

Another possibility strikes us as especially likely—that congruence flows from lower court judges' attempts to reach legally sound decisions. This is almost certainly the explanation most judges would offer, and there is reason for scholars to take it seriously. Trained as lawyers and facing an audience trained in the same way, judges might place a high value on making decisions in accordance with accepted principles of reasoning, whether they derive satisfaction from doing so, think it ethically right for them to do so, or desire the good opinion of their colleagues and audience.Footnote ¹⁴ Since faithful adherence to a higher court's precedents is a central element of legally sound decisionmaking, the desire to make sound decisions should tend to produce congruence in cases involving settled issues of law—that is, in typical cases. Less obvious, the desire to make legally sound decisions should tend to have the same effect even in less routine cases, at least where the Supreme Court does not deviate sharply from its own precedents. It will sometimes be reasonably clear how existing doctrine should be extended to new issues or circumstances. And even where existing precedents are not sure guides, if judges and justices make sincere attempts to weigh the legal strength of opposing positions, the fact that they received similar legal training and interpret the same sources of law should more often than not lead them to come down on the same side.

We repeat that the foregoing is speculation. Our data do not allow us to draw any firm inferences about the forces that do foster compliance. Nor is it clear to us how one could identify these forces.Footnote ¹⁵

It is also important to be cautious in generalizing from our findings. They might not hold for other time periods or other areas of law. In the period studied here, search and seizure law underwent a good deal of doctrinal change. Circuit judges might have had trouble charting these changes or predicting outcomes. This possibility strikes us as unlikely, given Reference SegalSegal's (1984) finding of fact-pattern coherence across Supreme Court decisions and the nearly 80% congruence rate in the cases we examined. Further, it is easier to picture variation in rates of compliance across time and issues than in causes of compliance. Still, the possibility cannot be denied.

A more likely possibility is that results would be different for different courts. The threat of reversal may carry little weight with circuit judges only because it is so weak. Judges whose decisions are more often reviewed, such as appellate judges in some states and, especially, trial judges in virtually every court system, might take the threat more seriously.

These caveats aside, we believe that our results entitle us to draw two important conclusions: Lower court compliance can flow from sources other than the fear of reversal; consequently, higher courts may be able to enjoy substantially faithful, responsive behavior from their subordinates even without maintaining a credible threat of reversal.