A. Economic Incentives and Private Enforcement

Our conceptualization of how private enforcement regimes mobilize private litigation pivots on economic incentives for plaintiffs and their lawyers to file suit. To be sure, the choice of whether or not to sue may be influenced by forms of utility or disutility distinct from, and not reducible to, money. We focus on economic incentives because, from the standpoint of institutional design, they can be readily deployed by legislative drafters seeking to mobilize private enforcement, which is the focus of our analysis.

On the stylized law and economics account of the choice to litigate, a prospective plaintiff will sue when a case’s expected monetary value (EV) if tried is positive, where EV is a function of the plaintiff’s estimate of the expected monetary benefit of the case if she prevails (EB), the probability that she will prevail if the case goes to trial (p), and the expected costs of litigating the claim (EC). Thus, EV = EB(p) − EC, and the rational plaintiff will file suit if EV is positive (Posner Reference Posner1973; Priest and Klein Reference Priest and Klein1984; Polinsky and Shavell Reference Polinsky and Shavell1998; Cooter and Ulen Reference Cooter and Ulen2004).

Expected benefits, expected costs, and probability of success can all be influenced by the details of a statutory private enforcement regime (Farhang Reference Farhang2010, chap. 2). Expected benefits (EB) are determined to an important extent by rules governing how much monetary damages a plaintiff will be able to recover (Galanter and Luban Reference Galanter and Luban1993; Polinsky and Shavell Reference Polinsky and Shavell1998). Expected costs are the sum of the filing fee, attorney’s fees, and other costs of litigation (Shavell Reference Shavell1982; Zemans Reference Zemans1984; Kritzer Reference Kritzer2001). Probability of success is, of course, strongly influenced by case facts but also by rules governing burdens and standards of proof, discovery, evidence, and liability (Cooter and Rubinfeld Reference Cooter and Rubinfeld1994; Kornhauser and Revesz Reference Kornhauser and Revesz1994; Cooter and Ulen Reference Cooter and Ulen2004, 431–32).

In order to understand actual litigation activity, one must pay particular attention to the incentives faced by attorneys (Galanter Reference Galanter1974; Johnson Reference Johnson1980; Zemans Reference Zemans1984; Galanter and Luban Reference Galanter and Luban1993; Kritzer Reference Kritzer2004). Plaintiffs’ lawyers in the United States are regularly dependent on the profitable resolution of a case for some or all of their compensation due to the inability of plaintiffs to cover the often massive expenses of a lawsuit. This is commonplace in statutory regulation where private enforcement regimes are utilized. Lawyers representing plaintiffs alleging violations of statutes in the areas of civil rights, labor, securities, antitrust, consumer, and environmental regulation, for example, are often compensated, in full or in part, on the basis of statutory fee awards, a share of damages provided by statute, or some combination of the two (Derfner Reference Derfner1977; Zemans Reference Zemans1984; Greve Reference Greve1989; Davies Reference Davies1997; Mongoven Reference Mongoven2005; Goldfarb et al. Reference Goldfarb2006; Drahozal Reference Drahozal2009, 750; Farhang Reference Farhang2010). When for-profit lawyers’ compensation is based on the proceeds of successful litigation, their willingness to represent plaintiffs will typically center on both their assessment of the likelihood of establishing liability and their estimate of the extent of monetary recovery available (Johnson Reference Johnson1980; Galanter and Luban Reference Galanter and Luban1993; Davies Reference Davies1997; Kritzer Reference Kritzer2004).

We stress that one cannot presume that potential plaintiffs and counsel will evaluate expected value in the same way. In addition to noneconomic considerations, potential plaintiffs may estimate their probability of success as higher, or the magnitude of their compensable injuries as larger, than a potential attorney would. This may provide a partial explanation for the fact that some plaintiffs who seek and are unable to obtain counsel elect to proceed with litigation pro se.

B. The Importance of Attorney Representation to Private Enforcement

The efficacy of private enforcement regimes for policy implementation turns on Congress’s capacity to mobilize attorneys, not just plaintiffs. Litigation in courts is a complex process riddled with technicality, and unrepresented plaintiffs have great difficulty navigating it effectively without counsel. Empirical studies have found that litigants represented by counsel in court proceedings are far more effective than pro se plaintiffs in enforcing statutory rights. Across multiple policy areas, researchers have found that represented litigants are less likely to have their cases dismissed, more likely to have judgments entered in their favor, more likely to secure more of the relief sought, and more likely to achieve settlement (Fusco, Collins, and Birnbaum Reference Fusco, Collins and Birnbaum1979; Schwab and Eisenberg Reference Schwab and Eisenberg1988; Seron et al. Reference Seron, Frankel, Van Ryzin and Kovath2001; McDermott and Obar Reference McDermott and Obar2004; Pardo Reference Pardo2009; Nielsen, Nelson, and Lancaster Reference Nielsen, Nelson and Lancaster2010; Pattanayak, Greiner, and Hennessy Reference Pattanayak, Greiner and Hennessy2013).

The question naturally arises whether these differential effects are driven by the attorney screening process: perhaps litigants with better cases are more likely to secure representation, and thus represented plaintiffs do better because of the quality of their cases, not because they are represented. As noted above, however, the attorney screening process involves an assessment of not just the merits of a case but also, given the merits, available economic recoveries. Further, we note that the studies by Seron et al. (Reference Seron, Frankel, Van Ryzin and Kovath2001) and Pattanayak et al. (Reference Pattanayak, Greiner and Hennessy2013) deployed powerful research designs to guard against this selection bias issue. Lawyers were randomly assigned to a group of unrepresented litigants, and those with lawyers did far better than those without.Footnote ¹

The foregoing discussion of economic incentives and private enforcement regimes connects directly to an important line of research concerning the significance of certain forms of legal infrastructures to the elaboration and protection of rights through courts. Marc Galanter (Reference Galanter1974) famously argued that “repeat players” (who use the courts frequently and are typically “haves”) possess significant advantages over “one-shotters” (who use the courts infrequently and are typically “have-nots”). The advantages importantly include access to greater legal resources, skill, and expertise in the litigation process, and Galanter speculated that one way to counterbalance the inequality between repeat players and one-shotters could be through legal rules and policies that have the effect of increasing the supply of quality legal services for have-nots. The logic linking economic incentives to private enforcement regimes highlights that the existence of an important “support structure” (Epp Reference Epp1998) to enforce statutes—a bar of ready, willing, and able lawyers—may be significantly a function of legislative design.

A. Empirical Research on Changes in Expected Value

Two studies have evaluated whether Congress increased the assertion of claims by increasing their economic value, though they did so without regard to whether claimants were represented by counsel. Schwab and Eisenberg (Reference Schwab and Eisenberg1988) evaluated the effects of the Civil Rights Attorneys Fees Awards Act of 1976, which created a plaintiffs’ fee-shifting provision to govern numerous already-existing civil rights statutes, including section 1983 actions for violations of federal constitutional rights by state officers. They found, at best, “scant evidence” that the law affected the volume of filings (Reference Schwab and Eisenberg1988, 760–61). Farhang (Reference Farhang2009) evaluated the effects of the Civil Rights Act of 1991—which markedly increased monetary damages available under Title VII of the CRA of 1964—on the volume of Title VII charges filed with the Equal Employment Opportunity Commission (EEOC). He finds that the law produced a 58% increase in Title VII charges before the EEOC (27). While filing an EEOC charge is a formal legal precondition to suit, claimants in a large majority of EEOC charges are unrepresented by counsel and do not file litigation following that administrative process (LeRoy and Feuille Reference LeRoy and Feuille2003, 290; Taylor Reference Taylor2009, 147; Hickox Reference Hickox2010, 160). Neither of the above studies sheds light on the issue of representation.

In this article, we investigate the effects of the CRA of 1991 on the probability that Title VII plaintiffs were represented by counsel, and we provide the first empirical study to test the effectiveness of a self-conscious congressional effort to bolster attorney representation, in the face of extensive pro se litigation, via increasing the economic value of claims. We do so in an area in which Congress perceived a problem that victims of unlawful conduct were unable to secure counsel and therefore were unable to meaningfully enforce the public policies embodied in a statute. Given the importance of representation to meaningful and effective litigation, evaluating whether Congress succeeded in its stated goal of increasing representation, and if so, to what extent, has important implications for the design of regulatory implementation.

B. Theorizing Linkages between Changes in Expected Value and Representation

Predicting how an increase in expected value will influence the probability of representation is not straightforward. Since an increase in expected benefits can enlarge the population of claims that potential counsel and/or plaintiffs regard as having positive expected value, it can affect the overall mix of characteristics in the pool of potential claims in ways associated with plaintiffs’ ability to secure counsel. This is complicated further by the fact that changes in expected value may mobilize pro se litigation as well as represented litigation because potential plaintiffs may assess expected value as positive where counsel judges it to be negative. Thus, predicting how an increase in expected value will influence the probability of representation hinges on speculative assumptions. A few simple examples serve to illustrate the nature of these assumptions and how variation in them can lead to different outcomes for the probability of representation.

Recall the equation EV = EB(p) − EC, where the expected value of a claim is a function of expected benefits, probability of success, and expected costs. Many factors influence probability of success, including case facts. Consider case facts as they affect probability of success to be fp, where higher values of fp are associated with stronger case facts. Clearly, an increase in recoverable damages (EB) via a legislative intervention, other things equal, may induce the filing of claims with lower values of fp.

In panel A of figure 1, the legislative intervention occurs in a status quo with 100 suits, where 60% of plaintiffs are represented and 40% are pro se. There are two types of claims: fp ₁ claims were litigated preintervention and continue to be litigated postintervention; fp ₂ claims were not litigated preintervention but are litigated postintervention, and would not be litigated but for the intervention; and fp ₁ claims have stronger case facts than fp ₂ claims (fp ₁ > fp ₂). Thus, fp ₂ claims are, as compared to fp ₁ claims, factually weaker cases that were drawn into litigation by the increase in recoverable damages. Postintervention, one can conceptualize two distinct representation dynamics. First, with respect to the 100 claims of the type already being litigated preintervention (fp ₁ claims), it seems clear, as a matter of theory, that we would expect an increase in representation. From the standpoint of counsel, other things equal, some fraction of the fp ₁ claims will move from negative to positive expected value because of the increase in available damages. In the postintervention portion of panel A, there is an increase in representation to 80% for fp ₁ claims.

Figure 1.

Decision tree representing the effect of increasing the expected value of a case on representation. The percentage of plaintiffs with counsel is reported for each decision node. The figure illustrates how changes in expected value may mobilize pro se litigation as well as represented litigation because potential plaintiffs may assess expected value as positive when counsel judges it to be negative.

At the same time, however, there are 50 fp ₂ claims drawn into litigation that are factually weaker and of a type that had not been litigated preintervention. Among these, 30 are represented, while the remaining 20 proceed pro se. When fp ₁ and fp ₂ claims are aggregated, 73% of plaintiffs (110 of 150) secure representation in the postintervention period. Relative to the status quo of 60% representation, the effect of the increase in recoverable damages was to significantly increase the probability of representation in the postintervention pool of litigants as compared to the preintervention pool.

In the second scenario, depicted in panel B of figure 1, the intervention occurs in the same status quo. As distinguished from the first scenario, (1) twice the number of factually weaker fp ₂ claims are newly drawn into litigation (100 instead of 50), and (2) a smaller fraction of them secure representation (40% instead of 60%). Now, when fp ₁ and fp ₂ claims are aggregated, in the postintervention period, 60% of plaintiffs (120 of 200) secure representation. Consistent with reasonable theoretical expectations, fp ₁ claims, which were already being litigated preintervention, became more likely to secure counsel as a result of the increase in available damages. However, these representation gains were exactly offset by lower levels of representation in the newly mobilized and factually weaker fp ₂ claims, and thus the overall postintervention level of representation was unchanged as compared to the preintervention level.

Finally, the third scenario, depicted in panel C of figure 1, represents a decrease in representation. As distinguished from the second scenario, we make only one change: in the postintervention period, a smaller percentage of claims already being litigated preintervention (fp ₁ claims) secure representation (70%). This remains consistent with reasonable theoretical expectations: there is an increase in representation among claims already being litigated preintervention, but by a lesser degree than in the first two scenarios. In the third scenario, when fp ₁ and fp ₂ claims are aggregated, 55% of plaintiffs (110 of 200) secure representation in the postintervention period. This scenario suggests the clear logical possibility that an intervention aimed at increasing representation via increasing damages could have the effect of reducing representation in the overall population of claims litigated postintervention.

These examples illustrate why it is difficult to generate strong predictions about the effect of an increase in expected value on the probability of representation. It could increase levels of representation, have a null effect, or decrease levels of representation. A prediction of which state of affairs will result from the intervention would hinge on expectations about the magnitude of the effect on representation for claim types already being litigated preintervention (fp ₁ claims), the number of claims that will be newly drawn into litigation as a result of the intervention (fp ₂ claims), the fraction of those fp ₂ claimants that will secure counsel, and the propensity of unrepresented fp ₂ claimants to proceed pro se.

The possibility of changes in the pool of Title VII claims after the CRA of 1991 points to both a limitation and a strength of our study. The limitation is that we are not able to evaluate the effects of pre- and post-CRA of 1991 incentives on the identical population of claims, which would allow the cleanest possible (experiment-like) assessment of the effects of changes in expected value on representation. However, it will always be the case, in reality, that when legislators seek to increase representation via increasing expected value, the increase in expected value will have the potential to affect aggregate filings and the nature of cases filed. Studying patterns of representation in this complex reality, as we do, thus has the potential to yield insights that are useful to those who design public policy in the real world.

If we discover that the act had the effect of reducing the overall probability of representation, this would provide cautionary evidence about the feasibility of solving the problem of underrepresentation by increasing expected value. Pro se litigants impose cost and delay on an adversarial system that is designed on the assumption of attorney representation, and numerous judicial administrative bodies and scholarly commentators have accordingly identified growing pro se litigation as a serious problem of judicial administration (Rosenbloom Reference Rosenbloom2002; Landsman Reference Landsman2009; Thompson Reference Thompson2010). Further, the great difficulty that pro se litigants have in navigating judicial proceedings fuels their alienation from the judiciary, compromises its legitimacy, and erodes its authority among the growing ranks of pro se litigants (Rosenbloom Reference Rosenbloom2002; Landsman Reference Landsman2009; Thompson Reference Thompson2010). From a policy design point of view, materially increasing pro se litigation has policy costs.

On the other hand, if we find that the CRA of 1991 brought about a substantively significant increase in representation, this would suggest that the positive operation of the increase in expected value predominated over any incentivization of more pro se claims within the overall pool of suits litigated. Of course, one could not generalize carelessly from this finding that it will hold in all contexts. Fuller understanding of this subject can be built only on a body of studies focusing on a range of contexts. We endeavor to offer the first.

A. The Origins of the Act

The bill that led to the Civil Rights Act of 1991 was initially motivated to override a series of five Supreme Court decisions handed down in May and June of 1989 interpreting Title VII of the Civil Rights Act of 1964. These cases, and several others that were decided while the override process was under way, all sought to cut back Title VII’s private enforcement regime. They spanned burdens of proof, standards of evidence, standing, statutes of limitations, attorney’s fees, and expert witness costs (Govan Reference Govan1993; Farhang Reference Farhang2010, chap. 6). In each of the cases the Supreme Court adjusted Title VII’s private enforcement regime to the detriment of women and minority plaintiffs, reducing opportunities and incentives for private enforcement actions.

B. The Content of the Act

In the CRA of 1991, a Democrat-controlled Congress overrode a large majority of the seven cases. The express goal of the statute was to substantially restore Title VII law to its condition prior to the decisions being overridden (Govan Reference Govan1993; Farhang Reference Farhang2010, chap. 6). However, the CRA of 1991’s most important changes went well beyond overriding the Supreme Court. The law added new economic incentives to Title VII’s private enforcement regime that were calculated to increase plaintiffs’ ability to secure representation by counsel and thereby to bolster private enforcement. Prior to the amendments, while winning plaintiffs could recover attorney’s fees, money damages were limited to back pay. The CRA of 1991 provided that plaintiffs could additionally recover compensatory damages, including for emotional pain and suffering, as well as punitive damages (42 U.S.C. § 1981a[a]).Footnote ² This change in available damages transformed claims that had previously carried little or no money value into claims with potentially substantial monetary worth. It thereby increased the expected value of many Title VII lawsuits from the standpoint of both plaintiffs and their lawyers.

A related factor that likely increased the perceived expected value of Title VII claims was a provision that created a right to trial by jury in cases seeking compensatory or punitive damages. Prior to 1991, no right to trial by jury existed for Title VII claims. The addition of the jury trial right was tied to the new damages provisions. The Seventh Amendment of the US Constitution guarantees the right to jury trial in actions brought in federal court to enforce “legal” rights and remedies, including suits seeking compensatory or punitive damages (Murphy 1995, 351–56). At least partly in recognition of this constitutional requirement, the CRA of 1991 created an express right to jury trial in suits seeking such damages (42 U.S.C. § 1981a[c]; Govan 1993). Thus, in claims seeking compensatory or punitive damages under the CRA of 1991, the liability determination of whether a plaintiff had been discriminated against and, if so, the damages determination of how much money to award her would be in the hands of a cross section of the population rather than a federal judge (at the time most likely to be white, male, and Republican). Both civil rights activists and those advocating the interests of the business community perceived that juries would, on balance, be more favorable to plaintiffs than would federal judges (Govan 1993).Footnote ³ That is, they believed that, as compared to bench trials, the right to trial by jury under the CRA of 1991 would increase Title VII plaintiffs’ chances of prevailing, as well as the magnitude of damages likely to be awarded. In the probable event that plaintiffs’ lawyers share this perception, the right to trial by jury, along with the new damages provisions, would thus increase their estimate of the expected value of Title VII claims.

For purposes of later analysis, it is important to emphasize that the CRA of 1991 changed Title VII law in two distinct ways. First, it overruled a set of Supreme Court decisions, substantially restoring the law to its condition in April 1989. Second, the law created wholly new damages and jury trial provisions. Thus, as compared to the Title VII status quo as it existed in April 1989, the new regime of Title VII law created by the CRA of 1991 was distinguished only by the new damages and jury trial provisions.

C. The Motivations for the New Damages and Associated Jury Trial Provisions

The legislative history of the CRA of 1991 reveals that one of Congress’s primary motivations in enacting the new damages provisions, in the face of extensive pro se Title VII litigation, was to mobilize counsel (Farhang 2010, chap. 6). Legislative supporters of the new damages provisions maintained that many plaintiffs with meritorious claims were unable to secure counsel (190). In support of this belief, Congress relied on evidence, including a report from the Federal Courts Study Committee, suggesting that the availability of fee awards under existing law was often insufficient to attract Title VII counsel because of uncertainty and long delay before fees could be recovered, narrow judicial construction of fee-shifting rules, and difficulty in recovering fees at market rates even when fee petitions were successful (190–91). Both the House and Senate reports on the issue explicitly maintained that there was an insufficient bar of attorneys willing to represent job discrimination plaintiffs with meritorious claims and that this problem resulted from inadequate economic incentives to do so (191–92). Compensatory and punitive damages, legislative advocates believed, would allow contingency arrangements to counteract the discounting of fees resulting from delay and uncertainty of fee recovery (192). To a significant degree, the new damages provisions were expressly justified as an instrument to mobilize more lawyers in the service of Title VII enforcement. Did it work?

A. Longitudinal Trends in Proportion of Plaintiffs Represented

In our empirical analysis we focus on explaining change over time in the likelihood that plaintiffs were represented. Another measurement approach would be to have as the dependent variable a count of represented plaintiffs within some unit of time. This would allow assessment of whether the CRA of 1991 increased the absolute number of represented Title VII filings. We do not add this to our analysis because we have insufficient data, which cover only 20 years. Twenty annual counts, or 40 biannual counts, would yield little data for analysis, and the data in the early years are too sparse to cut into smaller units of time. Disaggregating the data into separate models for different claim types, and controlling for important case characteristics, would be more than such data could plausibly bear. By exploiting the granularity of our case-level data and estimating the probability of representation with that unit of analysis, we greatly increase the statistical power of the models presented below, and we are able to probe the data more carefully by disaggregating modes into distinctive claim types and incorporating important case-level controls. We note, however, that inferences about the absolute number of represented plaintiffs can be made. If we find that there was a statistically significant and material increase in the probability of representation at the individual case level, in combination with the obvious postintervention growth in the number of cases depicted in figure 1, it will be evident that there was an increase in the number of represented cases associated with the heightened probability of representation.

In figure 3 we present locally weighted least squares (loess) curves fit through the annual percentage of cases in which the plaintiff was represented by counsel in Title VII cases and in our comparison group. The loess curves provide a very broad sense of the longitudinal patterns in the data.Footnote ⁷ The solid line in figure 3 indicates that the predicted values of the proportion of Title VII plaintiffs represented by counsel in our sample declined from 67% in 1981 to 50% in 1987; it ranged between 50% and 47% from 1987 to 1992; and from 1992 to 2000, it rose by 24 percentage points, to 73% represented.

Figure 3.

Locally weighted scatter plot smoothing (loess) curves estimating the annual percentage of federal employment discrimination lawsuits with a plaintiff that was represented by an attorney. Source: 659 cases of 1,000 random sample of the Federal Court Cases: Integrated Database (ICPSR no. 8429), subset to cases coded “job discrimination” between 1981 and 2000 in the Northern and Eastern Districts of California. Bandwidth = 80%.

The dotted loess curve represents the percentage of plaintiffs that were represented by an attorney in comparison group complaints (labeled comparison group A). Until the very end of the series, the predicted values of representation are considerably higher in this group than for Title VII claims. It declined from 83% in 1981 to 62% in 1987; it ranged between 62% and 65% from 1987 to 1992; and from 1992 to 2000, it rose by 7 percentage points, to 72% represented. We are mindful of the possibility that where comparison group claims are litigated jointly with Title VII claims, there may be spillover effects whereby, from the standpoint of prospective counsel, the expected value of comparison group claims is influenced by the presence of Title VII claims in the same suit. The CRA of 1991’s increase in the expected value of Title VII claims may have affected the attractiveness to counsel of post-CRA of 1991 comparison group claims with cofiled Title VII claims. The dashed loess curve in figure 3 drops complaints from the comparison group with concurrently filed Title VII claims (labeled comparison group B). In this set of cases, the downward trajectory in the proportion represented does not change in the early 1990s; from 1992 to 2000, it declined by 15 percentage points, to 63% represented. In our statistical models of representation in comparison group claims below, we will be able to control for the presence of concurrent claims under other job discrimination statutes, including Title VII.

B. Change-Point Models

Because loess curves smooth over sharp disjunctures in the data, they are not useful for identifying break points. When performing statistical tests to evaluate whether there are structural breaks in a longitudinal series, there are two basic and distinct approaches. First, change-point methods are agnostic about the existence or location of a break in the series and rely only on patterns in the data, with no structure being imposed on the data by the researcher. In the second approach, the researcher relies on theory and knowledge of historical events to impose temporal structure in the statistical model and thereby to test hypotheses with time variables incorporated into it. The first approach alone risks being atheoretical and structureless; the second approach alone risks imposing more structure on the data than is warranted (Wawro and Katznelson Reference Wawro and Katznelson2014). In our analysis, we combine the two approaches, seeking to leverage the advantages and obviate the weaknesses of each. With the first approach we assess whether the raw data reveal a structural break in the proportion of plaintiffs represented by counsel in the time window in which we anticipate effects of the CRA of 1991. With the second approach we assess whether variables modeling our anticipated cut point are statistically and substantively significant in a fully specified statistical model.

To analyze structural breaks in the raw data, we use a combination of two methods. Barry and Hartigan (Reference Barry and Hartigan1993) develop a Bayesian model that computes probabilities that there was a structural break in a series for each time interval in the data. Bai and Perron (Reference Bai and Perron1998, Reference Bai and Perron2003) use regression to identify break points in a series and to calculate confidence intervals around it. Before we present the results, it is important to establish reasonable expectations about when to expect a break in the series caused by the CRA of 1991, if one exists.

The CRA of 1991 applied to alleged discriminatory acts occurring on or after the law’s effective date of November 21, 1991 (Taylor 1994). Prior to filing a Title VII case in federal court, plaintiffs must first file an administrative charge with the EEOC. The statute of limitations for a Title VII charge in California is 300 days. It is not possible to know what fraction of claims are filed very close in time to the events sued upon and what fraction are filed just before the statute of limitations expires. After an EEOC charge is filed, the EEOC investigates the claim, and upon concluding its investigation, it issues a finding of whether there is “probable cause” to believe that illegal discrimination occurred. When it does so, it also issues a “right-to-sue” letter (regardless of the direction of the probable cause finding). After a claim has been before the EEOC for 180 days without resolution, a claimant is entitled, if it so requests, to receive a right-to-sue letter and terminate the administrative process. In 1992, the average length of time the EEOC took to process a charge through to issuance of a right-to-sue letter—including both claims in which the EEOC issued a finding and claims removed from the EEOC process after 180 days and prior to issuance of a finding—was 366 days (General Accounting Office 1999, 9). After a right-to-sue letter is issued, the claimant has 90 days to file a lawsuit in federal court.

On the basis of the assumption that cases spent an average number of days in the EEOC administrative process, we can estimate the lower and upper bounds of when alleged Tile VII violations occurring in December 1991 (the month following passage) would register as federal court filings. If EEOC charges were filed sometime in December 1991, the claim spent the average 366 days before the EEOC, and litigation was filed within a week of issuance of the EEOC right-to-sue letter, the lawsuit filing would occur in the range of December 1992 to January 1993. If EEOC charges were filed in the last week of the 300-day statute of limitations period, the claim spent the average 366 days before the EEOC, and litigation was filed in the last week of the 90-day window following issuance of the EEOC right-to-sue letter, the filing would occur in the range of December 1993 to January 1994. We believe that the upper bound is more realistic given that lawyers balancing portfolios of cases are more likely to focus on deadlines than to act at the earliest possible opportunity. The upper-bound prediction thus would anticipate a change in representation near the start of 1994.

Figure 4A represents both the Barry-Hartigan and the Bai-Perron change-point analyses run on the mean annual proportion of plaintiffs represented in Title VII claims. The top portion of the graph depicts the posterior probability of representation derived from the Barry-Hartigan model. It declined in the first several years of the series, was stable from 1983 to 1994, rose sharply from 1994 to 1996, and roughly stabilized again from 1996 through 2000. The lower panel of the figure represents the posterior probability of a change point in each year. That probability moves in a very low range from 1983 to 1993; it then rises sharply from 1993 to 1995, where it peaks; it then returns in 1996 roughly to its 1983–93 level, where it remains through 2000. The dark vertical dotted line is the change point identified by the Bai-Perron method. The Bai-Perron method identifies 1995 as the most probable break point in the series (like the Barry-Hartigan model), with 95% confidence intervals at 1994 and 1998. The 1994 lower bound in the Bai-Perron model coincides with the year the Barry-Hartigan model detects the first sharp jump in probability of representation after a decade of relative stability.

Figure 4.

Change-point models of representation for Title VII cases (panel A) and non–Title VII employment discrimination cases (panel B). The upper plot in both panels depicts the posterior probability of representation by year and the lower plot in both panels depicts the posterior probability of a change point in any given year derived from the Barry-Hartigan algorithm. The vertical dashed lines depict the change point (1995) and the vertical dotted lines depict the 95% confidence interval (1994 and 1998) identified using the Bai-Perron algorithm.

The change-point models thus identify a break point in the series of representation roughly coincident with the upper bound suggested by the scenario in which claims arising in December 1991 were filed with the EEOC late in their statute of limitations period, spent a year in the EEOC, and then were filed in federal court late in their 90-day window, in the range of December 1993 to January 1994.

Figure 4B represents the structural break analyses run on the mean annual proportion of plaintiffs represented in our comparison group of 247 federal job discrimination complaints containing claims not affected by the CRA of 1991. The top portion of the graph indicates that the probability of representation derived from the Barry-Hartigan model did not change over the 20-year series. The bottom portion indicates that the probability of a break point in each year was essentially constant at a very low level. The Bai-Perron model was also unable to identify any meaningful break point in the series. Whatever caused the sharp upward shift in Title VII representation beginning around 1994 did not operate on these comparison group claims. Still, these structural break analyses cannot rule out the possibility that some other factor in the early 1990s, distinctively associated with Title VII claims, was the real cause of the growth in Title VII representation while not affecting the comparison group. In order to investigate representation in Title VII claims further, we turn to statistical models that allow incorporation of controls for a variety of potentially important factors.

C. Model Specification

We use logistic regression to test the effect of the CRA of 1991 on representation by counsel, which is our dependent variable. In this modeling strategy, we explicitly model a change in representation with independent variables measuring a predicted break effectuated by the CRA of 1991. Here, we impose the location of the temporal break based on the theory we seek to test and on rules governing the timing of Title VII litigation. As discussed, the 1994 break point is the outer bound of the time in which claims based on discriminatory acts in December 1991 would reach federal court if lawyers filed near relevant deadlines. It has been preliminarily validated by our change-point analyses.

We include three time variables:

1. 1. Time is a linear annual time trend variable:

   
   This variable is necessary to ensure that any significant increase in the probability of representation after the CRA of 1991 is not actually a simple function of a broad time trend toward increasing representation by counsel that had nothing to do with the CRA of 1991.

2. 2. CRA captures the 1994 break point:

   
   This dummy variable is one way of detecting whether there was an increase in the probability of representation starting in 1994. It models the act as taking its full effect starting in 1994, in a single stepwise transition, when the value of the variable switches from 0 to 1.

3. 3. CRA × Time is an interaction of the CRA dummy and the linear time trend and is another way of detecting an increase in the probability of representation starting in 1994, but with a different functional form. In contrast to CRA, this variable takes the value of 0 prior to 1994 but then has a linear growth pattern starting in 1994. It models the CRA of 1991 as having triggered an incremental pattern of growth in the probability of representation starting in 1994.

We adopt a flexible model specification that can detect either a single stepwise increase in mean representation occurring at a discrete break point (the CRA dummy) or an incremental growth pattern beginning at that break point (CRA × Time) because we lack a firm theoretical basis to impose only one functional form in the model. On the one hand, it is possible that diffusion of information about the new law to prospective plaintiffs, and corresponding effects on their claiming behavior and efforts to secure counsel, were fully realized at a discrete point in time. Likewise, it is possible that the existing plaintiffs’ job discrimination bar had sufficient unutilized capacity to very rapidly and fully absorb new opportunities for litigation. In this scenario, we would predict a single stepwise increase to a new mean level of representation following passage of the law. This would follow the functional form of the CRA dummy.

On the other hand, it is possible that the diffusion of information to prospective plaintiffs, and corresponding effects on their claiming behavior and efforts to secure counsel, were more gradual. Likewise, it is possible that there was an incremental process of growth of the bar needed to litigate the new claims. In this scenario, we would predict an incremental growth in representation following passage of the law. This would follow the functional form of CRA × Time.

The social and economic processes involved in the diffusion of information to the plaintiff population, the timing of its influence on their claiming behavior, and the speed with which legal labor markets grow are complex. We are aware of no literature that would allow us to make strong predictions in favor of either a single stepwise change or an incremental growth scenario, to the exclusion of the other. Accordingly, rather than imposing one versus the other functional form on the data, we let the data speak for themselves and allow our interaction setup to adjudicate between theoretically plausible functional forms of the relationship we test for.

We include a battery of controls that may be associated with a plaintiff’s ability to secure counsel, as follows:

• dummy variables indicating the grounds of discrimination alleged, which are the nonexclusive categories of race, national origin, gender, religion, age, and disability;

• dummy variables indicating the nature of the discrimination alleged, which are the nonexclusive categories of failure to hire, termination, failure to promote, demotion, pay, hostile environment, retaliation, or a disparate impact claim;

• dummy variables indicating whether the defendant was a federal governmental entity, a state governmental entity, or a private entity, with private defendants left out as the reference category;

• dummy variables indicating concurrent claims under other statutes.

Finally, we control for judicial ideology. It has been shown that the ideology of appellate judges is an important factor contributing to the doctrine they create in civil rights cases, with greater degrees of liberalism associated with a more pro-plaintiff orientation (Segal and Spaeth Reference Segal and Spaeth2002; Sunstein, Schkade, and Ellman Reference Sunstein, Schkade and Ellman2006; Moyer and Tankersley Reference Moyer and Tankersley2012). Thus, we incorporate into our model measures of both Supreme Court and Ninth Circuit ideology. We model Supreme Court ideology with median Supreme Court justice ideology scores recently developed by Cameron and Park (Reference Cameron and Park2009); we observe similar results (clear statistical insignificance) in alternative specifications using two other widely used measures of Supreme Court ideology.Footnote ⁸ To model Ninth Circuit ideology, we followed Giles, Hettinger, and Peppers (Reference Giles, Hettinger and Peppers2001, Reference Giles, Hettinger and Peppers2002) in relying on a combination of appointing presidents’ and senators’ common space NOMINATE scores, and we compute Ninth Circuit ideology as the mean score of active judges on the circuit.Footnote ⁹ While one might anticipate that incorporating both Supreme Court and Ninth Circuit ideology into a model could raise problems of collinearity, this is not the case. The measures have a correlation coefficient of only .29, and each variable’s clear insignificance in all models (below) remains when the other is dropped. We lag both ideology measures 1 year to allow movement in judicial ideology to affect doctrine and lawyer perceptions of the expected value of lawsuits. In alternative specifications we also incorporated a variable measuring the mean common space NOMINATE scores (computed as described above) for active judges at the district court level, and this too proved insignificant. Summary statistics are presented in table 2.

Table 2.Summary Statistics of Variables Affecting Representation in Title VII Cases

Source—Federal Court Cases: Integrated Database (ICPSR no. 8429), subset to cases coded “job discrimination” between 1981 and 2000 in the Northern and Eastern Districts of California.

Note—Supreme Court ideology measures are based on Cameron and Park (2009). Ninth Circuit ideology measures are based on Giles et al. (2002).

Recall that the CRA of 1991 was precipitated by, and overruled, a series of Supreme Court decisions issued between 1989 and 1991 that were to the disadvantage of Title VII plaintiffs. It is important to stress that to the extent that the CRA of 1991’s override of these decisions—as distinguished from its new damages and jury trial provisions—enhanced the expected value of Title VII claims, the effect was to restore the law to its condition in April 1989, before the first of the overridden decisions were issued. The year 1989 is near the nadir of Title VII representation in the loess plot in figure 3, and it is in the midst of the long stable lower bound in the proportion of Title VII plaintiffs represented in the Barry-Hartigan model (see the upper quadrant of fig. 4A ). Accordingly, restoration of substantive Title VII doctrine to its state in 1989 cannot explain the subsequent jump in Title VII representation to levels not seen since the early 1980s.