NCAA enforcement and competitive balance in college football.

1. Introduction

The National Collegiate Athletic Association (NCAA) is the largest and best known of the four independent organizations created to regulate collegiate sports in the United States. (1) These organizations serve to protect the amateur status of college athletes, sanction athletic events (especially championships), and to otherwise regulate the behavior of member schools (and those affiliated with them) in the context of college athletics. These associations, especially the NCAA, appear to be cartels in which members agree to codify rules that regulate both the input (student athlete) and the output (sporting-event) markets (Fleisher et al. 1988; Fleisber, Goff, and Tollison 1992). As with any cartel, there are incentives for one or more members to violate the terms of the agreement. This in turn necessitates enforcement and punishment mechanisms to mitigate the incentive for members to cheat.

Centralized rule enforcement within a cartel is usually established to deter behavior deemed suboptimal to the group for which it is established. However, a lack of data often limits investigation into the impact of internal rule enforcement within industries, yet a few authors have studied the impact of antitrust investigations on the concentration of American industry (Feinberg 1980; Garbade, Silber, and White 1982; Block and Feinstein 1986; and Nissan 1998 are notable examples). The NCAA, as an organization of collegiate athletics, provides a unique instance in which the production of industry members is well defined and internal enforcement efforts, in the form of investigations and probations, are well advertised and documented.

While previous analyses of the impact of NCAA enforcement have predominantly focused on team-specific performance, the more general question of how NCAA enforcement affects competitive balance, or the skewness of on-field success, has received less attention. Anecdotal evidence indicates that college football has been dominated by relatively few teams over the past few decades, such as Miami (Florida), Alabama, Nebraska, Notre Dame, and the University of Southern California, to name a few. Therefore, whether NCAA enforcement efforts contribute to or detract from the competitive balance of Division IA football conferences is of interest.

The NCAA is an organization whose members agree to a codified (but alterable) set of regulations that aim to reduce the ability for some members to use their substantial resources to skew the distribution of player talent and on-field success away from smaller, perhaps poorer, schools. Even though schools are limited in the amount of in-kind and direct payments they can offer student athletes, the nonmonetary benefits of attending a school, including academic training, coach and program reputation, and the possibility of national exposure to professional scouts, might put smaller schools or lesser programs at a disadvantage. In response, some members of the NCAA might be motivated to cheat on the NCAA agreement in order to obtain the marginal benefit of having a prospective student athlete attend their schools. This moral hazard is the general target of NCAA regulations. Although the NCAA regulatory structure does not ensure perfect parity, the stated objective of the NCAA is to limit the desire for schools to violate the agreement and jeopardize the amateur status of college athletes. To encourage solidarity, the NCAA has a structured penalty schedule that can be applied to schools that violate the agreement, which, while appealable, is generally binding.

How much cheating actually declines after an increase in enforcement activity is an interesting topic, but data to investigate this issue are not available. Consequently, a second-best empirical alternative is to investigate how competitive balance changes with enforcement activity, which is a test of the impact of the NCAA's role as the enforcer of the membership agreement. Thus, the level of cheating is not directly addressed in this study; rather, we investigate whether competitive balance improves after increases in enforcement activities.

This article investigates two conflicting hypotheses regarding the role of NCAA rule enforcement within Division IA football conferences. The NCAA claims to act in the best interest of the "amateur spirit" of collegiate sports, thereby ensuring a "level playing field" and, in turn, encouraging competitive balance, or a more equitable distribution of playing talent and on-field success (see Zimbalist 2001 for a detailed industry analysis of college athletics). However, others posit that as an organization, the NCAA protects the relative dominance of "big-time" programs (for example, see Noll 1999) and as a result reduces competitive balance. To date, these disparate claims have only been tested indirectly. Which hypothesis provides an appropriate characterization of the NCAA is the empirical question addressed in this article.

Using data describing 11 major Division IA football conferences from 1953 through 2003, the impacts of NCAA enforcement efforts are estimated. We empirically relate competitive balance, as measured by the Herfindahl-Hirschman Index (HHI), to various measures of enforcement and punishment and additional control variables in an instrumental variables approach. The empirical results indicate that NCAA enforcement activities enhance competitive balance, whereas more severe punishment tends to reduce competitive balance. Evaluated at the sample means, NCAA enforcement activity leads to a net improvement in the competitive environment, which supports proponents of the NCAA who argue that it ensures a more "level playing field." While the reduction in competitive balance caused by more severe punishment is likely the source of anecdotal claims that the NCAA protects the status of dominant programs, these obvious reductions are generally more than offset by the less-obvious improvements in competitive balance resulting from NCAA enforcement efforts.

The remainder of this article is structured as follows. The next section outlines a brief history of NCAA regulation. Section 3 develops several testable hypotheses about the impacts of rule enforcement on competitive balance. Section 4 describes the data, the empirical methodology, and the results. The final section offers concluding remarks and avenues for future research.

2. NCAA Enforcement and the Associated Literature

In the first half of the 20th century, the growing popularity of college athletics and the substantial increase in potential revenues for many schools provided incentives to engage in actions thought to be detrimental to student athletes and amateur athletics in general. (2) These "nefarious" tactics motivated the NCAA to regulate both the input and output market of college athletics. In 1946, the "'Principles of Conduct of Intercollegiate Athletics" was drafted, and two years later this document was incorporated into the NCAA constitution and became known as the "Sanity Code." The Sanity Code imposed regulations on the recruitment and retention of student athletes: Scholarships for athletic activities were restricted to only tuition and fees, and the financial contact of alumni with prospective student athletes was limited.

Although the Sanity Code was not the first attempt to regulate the behavior of member schools, the NCAA considered this regulation more significant because it created an enforcement mechanism to handle violations. A Compliance Committee became the arbiter of suspected violations that had been investigated by the Fact-Finding Committee. Initially, the only form of punishment available to the Compliance Committee was the termination of the violator's NCAA membership, decided by a vote of all NCAA members. However, NCAA members deemed this punishment too draconian and stripped the Compliance Committee of its punitive role in 1951. (3)

In 1953, NCAA members agreed to provide the Committee of Infractions (formally the Compliance Committee) with a continuum of powers with which to penalize rule violators in a manner less severe than termination; possible penalties included limiting the number of scholarships, limiting television appearances, and limiting participation in postseason play. Credibility was provided to the Committee of Infractions as punishments could be exacted without general membership approval. Accordingly, since 1953 the NCAA has had and employed a credible enforcement mechanism with which to maintain solidarity of the organization and, by its own claims, to promote the spirit of amateur sport and competitive balance. (4)

The increasing popularity of college athletics, especially football, men's basketball, and, to a lesser extent, women's basketball, has dramatically increased the financial returns available to university athletic departments. Over the past 20 years, the number of bowl games comprising postseason collegiate football has increased, contemporaneously, with increased media and stadium revenues available to football programs. (5) Therefore, if athletic departments, coaches, alumni, and boosters expect that violating NCAA regulations will improve on-field performance in the near-term, the potential gains from cheating have likely increased. According to Becket's (1968) explanation of criminal activity, if the gains associated with illegal behavior increase, either the probability of detection and punishment or the severity of the penalties imposed must also increase to offset the incentives of such illegal behavior.

The NCAA penalty schedule is similar to the legal penalties imposed in general society: More egregious violations carry increased severities in penalties, whereas minor offenses receive rather trivial punishments. NCAA punishments may include temporary limitations on scholarships, television appearances, and postseason play. While the NCAA does have the ability to prohibit a school or program from competing against other members, the so-called "death penalty," the only instance of this extreme punishment was (temporarily) applied to the football program at Southern Methodist University in 1987.

The intuition underlying the various forms of punishment in the NCAA seems to be two-fold. First, probations and further penalties seem intended to reduce the revenue of the sanctioned program or school, thereby penalizing an athletic program for "ill-gotten'" gains--past, present, and future-obtained through cheating. Second, penalties appear to be intended to reduce the ability of the program/school to recruit new student athletes, thereby reducing the school's competitive status and ability to generate revenue in the future.

To date, there has been relatively little direct analysis of NCAA enforcement efforts in college football. However, three studies provide a background for the current study. Fleisher, Goff, and Tollison (1992) investigated the probability that a school's football program would be investigated by the NCAA and found that a dramatic increase in the winning percentage of a school's football program had a strong correlation with an NCAA investigation for possible rules violations. They also found that football programs tend to suffer significant reductions in on-field success after NCAA sanctions are imposed.

Eckard (1998) considered competitive balance in seven major Division IA football conferences before and after NCAA enforcement began in 1953, using the variance of relative team positions over time to measure competitive balance. He found that competitive balance improved for five of the seven conferences, but he generally concluded that the NCAA acts as a traditional cartel by protecting the stability of relative team rankings in the conferences under its control.

Depken and Wilson (2004) investigated how competitive balance across all Division IA football teams has been affected by major changes in NCAA regulations over time. While Division IA football has become less balanced over time, several institutional changes have had a beneficial impact on competitive balance, specifically the creation of the NCAA, the creation of a credible enforcement mechanism, and the relegation of many schools to Division IAA status. However, several other major rule changes, such as increased academic standards and the creation of the Bowl Championship Series, have reduced competitive balance. These findings indicate that pressure groups may motivate many of the institutional changes in the NCAA.

While the Depken and Wilson results indicate a correlation between various NCAA rule changes and overall competitive balance in Division IA football, it is perhaps unlikely that the relationships they show hold when analyzing competitive balance at the conference level. However, changes in rules can act as proxies for increased vigilance and enforcement on the part of the NCAA, the direct measure of which is the innovation in this study.

3. Hypotheses about NCAA Enforcement

From 1953 through 2003, the number of Division IA football teams varied from a minimum of 104 teams (in 1987) to a maximum of 147 teams (in 1977). Although the average number of football programs placed on probation has increased over time, as of 2002, the NCAA employed only 15 field investigators. While NCAA investigators may be quite adept at detecting cheating, whistle-blowers (both named and anonymous) almost certainly play an important role in rules enforcement. This follows from standard cartel theory, which would suggest that the more schools, sports, coaches, players, parents, fans, teachers, and administrators involved with NCAA athletics (however tangentially), the less likely it will be that cheating on the part of any athletic program or team will remain secret. (6)

Investigating how the competitive balance of Division IA football conferences changes when competitors are investigated and placed on probation extends the literature on centralized rules enforcement. While the NCAA claims that enforcement is specifically aimed at maintaining the spirit of amateur sport, many view NCAA enforcement as ensuring the continued dominance of historical "football powers." If the NCAA's claims are true, enforcement may have the positive byproduct of improving competitive balance. However, if detractors are correct, NCAA sanctions may lead to a decline in competitive balance.

The conceptual model employed for NCAA enforcement is that of Becker (1968). In his approach to the economics of crime, the deterrence of illegal behavior is affected by the severity of the punishment (if illegal behavior is detected) and the probability of being caught engaged in such illegal behavior. In empirical applications of Becker's theory, Grogger (1991) and Levitt (1998) discriminate between the influences of deterrence and incapacitation though punishment. Deterrence effects are created by a higher probability of detecting prohibited behavior and more severe sanctions when caught engaged in prohibited behavior. Incapacitation refers to the removal of lawbreakers from the pool of potential rules violators. This incapacitation is distinguished from general deterrence because it precludes criminal actions that otherwise would have taken place.

Incapacitation is an appropriate characterization of programs placed on probation by the NCAA for a rule violation. While programs on probation are allowed to continue operations, if a program on probation for a major rules violation is found guilty of an additional major rule violation, the NCAA can impose its most draconian sentence: "the death penalty." Thus, NCAA probations may have two distinct influences on the amount of cheating that occurs. Those programs on probation may be effectively incapacitated, while those programs not on probation might be deterred from future rules violations.

If a team becomes less competitive and loses more games when placed on probation, the lost games must be transferred to other schools in the form of wins. If more wins are transferred to below-average teams, sanctions will improve competitive balance. On the other hand, if more wins are transferred to above-average teams, sanctions will act to reduce competitive balance.

Enforcement activity can, therefore, have two different impacts on competitive balance. If the NCAA uses dramatic short-run improvements in winning percentage as a trigger to initiate investigations, as suggested by Fleisher, Goff, and Tollison (1992), programs with less variance in performance will be less likely to be investigated, ceteris paribus. As dominant programs with high winning percentages would have less variation in their on-field success, they may be relatively immune from investigation. In this case, if probations reduce on-field performance, NCAA enforcement would likely reduce competitive balance. On the other hand, if NCAA investigative activity concentrates on programs that perform at a relatively high level regardless of previous performance, NCAA enforcement may well improve competitive balance.

The severity of punishment is likely to have only a deleterious effect on competitive balance. The longer a program is on probation, the less likely it is to be highly competitive contemporaneously. Moreover, the longer a program is on probation, the greater the damage to the program's reputation, which affects the program's ability to recruit high-quality student athletes and its competitiveness well after the probation period is over. The longer the pending probations, the more likely it will be that those schools on probation are put at a disadvantage relative to the other programs in their conference. Hence, longer probationary periods are expected to reduce competitive balance.

The net effect of enforcement and the severity of punishment indicates whether enforcement of the NCAA membership agreement (generally speaking) leads to a net increase or decrease in competitive balance. If competitive balance suffers a net decline, claims that the NCAA protects or enhances the rents of dominant programs would be supported. On the other hand, if competitive balance enjoys a net improvement, this would support the claims of the NCAA's proponents.

An empirical test of which regime characterizes the NCAA would estimate both the impact of enforcement and the impact of punishment on competitive balance. From such estimation, it is possible to test which regime is supported by the data, rather than relying simply on conjecture and rhetoric. The following section empirically analyzes the impact of NCAA enforcement activity on the competitive balance of several major Division IA football conferences from 1953 through 2003.

4. The Effect of NCAA Enforcement on Competitive Balance

Empirical Methodology and Data

The dependent variable in the empirical specification is the competitive balance for conference i in year t, measured using either the HHI or the natural logarithm of HHI. To measure on-field performance, and to account for ties (which were possible before 1996), each team is credited with two points for a win, one point for a tie, and zero points for a loss, similar to the point system used in the National Hockey League. (7) Performance points for each team's conference win-loss-tie record and total performance points for each conference in each year are calculated. "Market shares" (MS [member of] 0, 100) are then calculated as each team's performance points as a percentage of aggregate conference performance points, and the Herfindahl is calculated as HHI = [[summation].sup.N.sub.i]M[S.sup.2.sub.i] where N is the number of teams in the conference. (8)

The performance across teams in a given conference is made comparable by limiting the focus to intraconference games. Using only conference games abstracts from differences in the strengths of out-of-conference scheduling, which could distort competitive balance across schools in a conference. Although it is standard practice for each team in a conference to play the same number of conference opponents, teams do not necessarily play every possible conference opponent. In conferences with many members, it is not possible for every team to play all other teams in the conference; therefore, schedules often rotate over a set number of years. In a given year, one team may have a relatively strong in-conference schedule while another team enjoys a relatively weak schedule. Over time these scheduling inconsistencies are not expected to permanently distort the measure of competitive balance.

To determine the impact of enforcement and punishment on competitive balance, the following relationship is estimated:

(1) HH[I.sub.i,t] = [[beta].sub.1]HH[I.sub.i,t-1] + [[beta].sub.2] [TEAMS.sub.i,t] + [[beta].sub.3] [ENFORCEMENT.sub.i,t] + [[beta].sub.4] [PUNISHMENT.sub.i,t] + [[epsilon].sub.i,t]

where the [beta]'s are parameters to be estimated, i identifies the conference, and t identifies time from 1953 through 2003. The independent variables include the once lagged dependent variable (HH[I.sub.i,t-1]), the number of teams in the conference, and variables used to control for the severity of punishment and the level of enforcement in a conference in a given year. The error term is initially assumed to be [[epsilon].sub.i,t] = [[alpha].sub.i] + [[gamma].sub.t] + [u.sub.i.t], where [[alpha].sub.i] is a conference specific effect, assumed constant over time; [[lambda].sub.t] is a time-specific effect, assumed constant across conferences: and [u.sub.i,t] is white noise. The composite error term is accommodated using panel estimators that allow for conference-specific heteroscedasticity and autocorrelation.

Although the level of the HHI does not reveal much about the relative standing of any particular team, changes in the concentration measure can be interpreted as changes in competitive balance. If NCAA enforcement activity or severity of punishment reduces the winning percentages of the targeted schools, the Herfindahl measure will increase (decrease) if those wins are redistributed to above-average (below-average) teams, corresponding to a reduction (improvement) in competitive balance.

A potential concern is the inclusion of a lagged dependent variable in Equation 1. As shown by Nickell (1981), lagged dependent variables can introduce inconsistency on the order of 1/T, where T is the number of time periods in the sample. For the conferences in our sample, the average number of time periods (i.e., years) is 36; therefore, we might expect a bias of approximately 2.7% in standard panel models. However, the lagged dependent variable is included for two reasons. First, there is no general theory for how the competitive balance of a conference evolves over time; therefore, the lagged dependent variable attempts to capture the various inputs to the competitive balance of a collegiate football conference, including coaches, training, injuries, weather, and other difficult-to-measure inputs to the competitive balance production function.

Second, it is likely that college football conferences do not fully adjust from a shock (whether random or otherwise) in a single year. For example, if a football program is investigated and placed on probation for four years, during which time the program can offer fewer scholarships, it is likely that the full impact of the probation will be felt over time, rather than in a single season. This implies that modeling a lagged adjustment process is appropriate. The parameter estimate of the lagged dependent variable can be used to determine the long-run effect of the various explanatory variables included in Equation 1. The long-run impacts of NCAA enforcement and punishment activity have not, to date, been estimated, and thus the insight obtained by including the lagged dependent variable seems worth the potential loss of consistency.

The remaining explanatory variables attempt to control for exogenous changes to the environment in which college football teams participate at the conference level. The variable TEAMS measures the number of teams in a conference in a given year. Over the past 50 years, the various conferences included in our sample (and many that are not) have expanded and contracted on an irregular basis. A general empirical finding is that more games in a season or more teams in a league improves competitive balance (see Fort and Quirk 1995; Depken 1999). Therefore, as the number of teams increases (decreases), the Herfindahl is expected to decline (increase). The inclusion of this variable controls for "changes" in the competitive balance that are simply a function of having more teams in the conference, rather than having less disparity among existing teams.

A comprehensive measure of enforcement in the NCAA has yet to be developed. It will prove extremely difficult, if not impossible, to determine the annual amount of effort, both labor and monetary, that the NCAA has dedicated to the enforcement of its agreement since 1953. Without ready knowledge of exactly which and how many resources the NCAA commits to enforcement, the empirical researcher is left to measure enforcement as accurately, but also as broadly, as possible. In this study, the variable ENFORCEMENT is measured two ways: total football-related investigations and total football-related probations. (9)

The number of football-related investigations is analogous to arrests (or indictments), and the number of football-related probations is analogous to the number of convictions; arrests and convictions are commonly used in the general crime literature. While neither measure captures all efforts related to enforcement (for example, resources expended that did not result in an investigation), the two measures arguably capture those enforcement efforts expected to influence the level of cheating by a conference's schools. Both investigations and new and old (that is, yet-to-expire) probations in a given conference arguably signal a credible threat that the NCAA will impose additional sanctions on cheaters. A priori, it is expected that the total number of pending probations will have a greater impact on competitive balance than total investigations.

Fleisher, Goff, and Tollison (1992) suggest that dramatic improvements in winning percentage can trigger an NCAA investigation (and possibly probation). A dramatic improvement in a team's winning percentage might also be associated with a change in the competitive balance measure. This implies that investigations and probations are likely correlated with [u.sub.it] in Equation 1, for which an instrumental variables approach is desirable.

As shown by Depken and Wilson (2004), several changes to the NCAA membership agreement correlated with changes in the competitive balance of Division IA football as a whole. As additional rules and regulations are added to the NCAA membership agreement, it is arguably more difficult for schools to comply. The rule changes would seem to be natural candidates to instrument for the number of investigations and probations in a conference in a given year. The following policy changes were used as instruments: the increase in minimum grade point average (GPA) standards in 1966, the relegation of some schools to Division IAA status in 1982, Proposition 48 in 1985, and Proposition 16 in 1992. An additional instrument is whether the conference is a BCS Conference. (10) Using these instruments, Durbin-Wu-Hausman tests indicate that, except for the lagged competitive balance measure, only the enforcement variable, whether measured by investigations or probations, is correlated with [u.sub.it] in Equation 1.

The variable PUNISHMENT captures that aspect of the economic theory of crime associated with the fines incurred if 'caught and convicted' of violating the NCAA membership agreement. Although probations often carry additional penalties, including loss of scholarships, postseason play, or television appearances, these additional penalties did not prove statistically relevant when added to Equation 1. Therefore, punishment is measured as the average length of probations pending (measured in years) in a given year in a given conference. For example, if three teams were on probation in a conference in a given year, with School A on probation for two years, School B on probation for three years, and School C on probation for tour years, the average length of the pending probations would be 9/3 = 3 years. In many cases there are no pending probations in a conference in a given year, in which case the average length of pending probations is coded as a zero.

The data describe 16 different college football conferences from 1953 to 2003: the Atlantic Coast, Big East, Big 6, Big 7, Big Eight, Big Ten, Big XII, Conference USA, Mid-American, Mountain West, Pacific Coast, Athletic Association of Western Universities (AAWU), Pacific Eight, Pacific Ten, Southeastern, Southwest, and the Western Athletic Conferences. Historical team records were obtained from David Wilson. (11) The HHI is calculated for each conference and year with market shares based on intraconference performance points. Several conferences have existed for only a few years (for example, Conference USA and the Big XII). However, several of the conferences in our sample can be combined to form two distinct time series: one for the Big XII and one for the Pacific 10. In 1947, the Big 6 became the Big 7, which in turn became the Big 8 in 1960, which in turn merged with four teams from the defunct Southwest Conference to become the Big 12 in 1996. (12) Similarly, the Pacific Coast Conference dissolved in 1959, and several members formed the AAWU, which became the Pacific Eight in 1968, which in turn joined with Arizona and Arizona State to become the Pacific Ten in 1979. In these two cases, the historical lineages of the two conferences are included in a single time series. Thus, the unbalanced panel used in the estimation comprises 11 time series.

The length of probations, the number of schools on football-related probation, and the total number of football-related investigations for each conference-year in the sample were obtained from the NCAA's Major Infractions Database. (13) While the NCAA maintains a Secondary Infractions Database, these data are not utilized because of the minor nature of the infractions and subsequent penalties. (14)

Descriptive statistics of the 401 observations included in the sample are reported in Table 1. The average Herfindahl was approximately 1490. During the sample period there was approximately one football-related investigation and one football-related probation in a conference every two years, although in many conferences there was no enforcement activity for many years. In conferences in which at least one football-related investigation took place, the average number of investigations was 1.40, and the average number of probations pending was 1.08. This indicates that if a conference came under NCAA scrutiny to the extent that an investigation was initiated, the NCAA often investigated more than one team, and investigations were more frequent, on average, in conferences with more pending probations. This might indicate that the NCAA uses past "bad" behavior of conference members as a guide to where to direct enforcement efforts.

The largest number of football-related investigations in a single year was in the Pacific Coast Conference in 1957, when four of nine members were investigated. However, a number of conferences experienced three football-related investigations in a single year: the Atlantic Coast Conference in 1983 and 1990; the Big 8 Conference in 1989; the Big Ten Conference in 1976, 1991, and 1999; the Pacific Ten Conference in 1982 and 1997; the Southeastern Conference in 1985 and 1986; and the Southwest Conference in 1965 and 1987. The largest number of football-related probations in place at a given time was five in the Southwest Conference in 1987; the Pacific Ten Conference had four schools on football-related probation in 1983. The average length of pending probations over the entire sample period was approximately one year; however, in conferences in which at least one probation was pending, the average length of probations increased to 1.69 years. Probations have become longer over time, perhaps reflecting more egregious cheating, a smaller chance of detection, or greater potential gains to cheating. The Big Ten in 2001, 2002, and 2003; the Big XII in 1996 and 1997; the Pacific Ten in 2003; the Southwest in 1994; and the Western Athletic Conference from 1997 through 2000 each had one five-year probation pending.

Empirical Results and Discussion

Various specifications of the primary model described in Equation 1 are reported in Table 2. One major distinction between the models is the form of the dependent variable. The left half of Table 2 reports results using the level of the Herfindahl as the dependent variable. While parameter estimates can be interpreted as the marginal influence on the Herfindahl, a measure from 0 to 10,000, the intuitive interpretation of the parameter estimates can be elusive. The fight side of Table 2 presents results with the natural log of the Herfindahl as the dependent variable, in which parameter estimates reflect the percentage change in the Herfindahl due to a unit change in the explanatory variable. The upper half of Table 2 reports models using the number of football-related investigations in a conference as the measure of enforcement. The lower half reports models using the number of pending football-related probations as the measure of enforcement.

For each functional form and measure of enforcement, three models are estimated: traditional instrumental variables two-stage least squares (IV-2SLS), instrumental variables generalized method of moments (IV-GMM), and the dynamic panel estimator (DPD) developed by Blundell and Bond (1998). The former two specifications allow for generalized heteroscedasticity and autocorrelation but do not correct for the potential endogeneity concerns introduced by the lagged dependent variable; the lagged dependent variable is treated as nonstochastic. The anticipated bias in the first two specifications is perhaps as great as 2-3%. The dynamic panel data estimator uses the correct instrument for lagged dependent variable. However, each specification yields very similar results, indicating that the potential bias (either through the lagged dependent variable or weak instruments) does not seem to be serious.

An additional concern is introduced by the potential for multivariate outliers in the data, especially in the case of the AAWU, which was reduced to five teams for several years. The dramatic reduction in the number of teams in that conference artificially increased the HHI, which is partially accounted for by the inclusion of the TEAMS variable. However, if the number of teams is not fully able to accommodate the artificial yet dramatic increase in the HHI, both inefficiency and potential bias can be introduced into the estimation. To determine if there are any multivariate outliers, the procedure proposed by Hadi (1992, 1994) was used to identify four outliers in the data--the AAWU for the years 1959, 1960, and 1962 and the Western Athletic Conference in 2001. A dummy variable was created that takes a value of one for these four conferences in the appropriate years and zero otherwise; this dummy variable is included in the estimation to accommodate the outliers. The dummy variable did not dramatically alter the estimates reported here but did improve the precision of the parameter estimates.

Initial estimation was undertaken including conference- and year-specific fixed effects. However, the estimation results were less than satisfactory because of colinearity problems. The year-specific effects were, therefore, dropped and replaced with a linear time trend beginning in 1953 and its quadratic. Neither of these time variables proved significant, and therefore, the final estimations include only conference fixed effects, that is, the composite error term in estimating Equation 1 is actually [[epsilon].sub.it] - [[alpha].sub.i] + [u.sub.i].

The first-stage regressions are not reported here but are available from the authors upon request. While the conference-specific effects were all individually insignificant, the null hypothesis that they were jointly insignificant was soundly rejected for all specifications. In general, only a few of the instruments proved individually statistically significant, although they were jointly significant for each specification. The excluded instruments consistently significant at the 10% level were the relegation of schools to Division IAA status and whether the conference was a BCS conference, both of which were positively related to both the number of investigations or probations. As shown by Bound, Jaeger, and Baker (1995), the strength of the instrument set is inversely proportional to the joint significance of the excluded instruments. Table 2 reports the first-stage F-statistics on the excluded instruments. While the F-statistics are not as high as might be hoped, they do indicate that the instrument set is sufficiently strong in explaining the variation of investigations or probations. Moreover, for each specification in Table 2, the Jansen test of overidentifying restrictions cannot be rejected, indicating that the instruments appear to he exogenous and valid.

Turning attention to the second-stage regression results reported in Table 2, focusing first on the upper panel, the estimation results are fairly robust across the various specifications. The parameter estimate on the lagged dependent variable is consistently positive and less than one, indicating that the evolution of conference-level competitive balance is stable and that the number of teams in a conference reduces the Herfindahl, as would be expected. The number of football-related investigations consistently correlates with more competitive balance (i.e., a lower concentration of performance points). In both the IV-2SLS and IV-GMM specifications, each additional investigation reduces the Herfindahl by approximately 155 points; the DPD estimate is a bit lower in absolute value. In all three specifications, the parameter estimates are statistically different from zero in a two-tailed test and are safely characterized as less than zero. The empirical results indicate that an increase in NCAA enforcement efforts, as reflected in investigatory activity, improves competitive balance, as reflected in a reduction in the HHI.

However, the severity of punishment, measured by the average length of pending probations, is associated with a less-competitive environment. A one-year increase in the average length of pending probations reduces competitive balance, reflected in an increase in the HHI of 27 to 35 points. This result is consistent with that of Fleisher et al. (1988), who found that teams performed worse after being put on probation.

To put the impacts of enforcement and punishment in perspective, the right side of the upper half of Table 2 reports the results using the natural logarithm of HHI as the dependent variable. The lagged dependent variable indicates that approximately 18-22% of the previous year's competitive balance is carried over into the next year. This would indicate that a shock to a conference's competitive balance (whether random or not) would take between 4.5 and 5.5 years to be fully dissipated. This time frame is consistent with the eligibility of NCAA student athletes and seems to indicate that the full effect of punishment is felt only over time.

Each additional team in a conference reduces the Herfindahl by approximately 8.6%. Each football-related investigation improves competitive balance by approximately 6%, whereas a one-year increase in the average length of pending probations reduces the competitive balance measure by approximately 1.7%.

The bottom half of Table 2 presents similar estimation results, replacing the number of football related investigations with the number of football-related probations. The results are qualitatively consistent with those in the upper half of Table 2. Approximately 24-27% of the once-lagged competitive balance is carried over into the next year, and the number of teams in the conference reduces the Herfindahl. The number of pending probations in a conference has a statistically significant and negative impact on the Herfindahl, indicating an improvement in competitive balance, and the length of probations reduces competitive balance. Notice that the marginal impacts of the number of probations and the length of pending probations are greater than the analogous estimates using investigations as the measure of enforcement. This might indicate that probations (actual punishment) rather than investigations (threatened punishment) are the more appropriate measure of enforcement. While each probation is associated with an investigation, but not vice-versa, actual probations might influence behavior and, hence, competitive balance more so than investigations alone.

The instantaneous effects of enforcement and punishment are reflected in the parameter estimates presented in Table 2; however, the parameter estimates on the lagged dependent variable indicate a partial adjustment process in the evolution of competitive balance in college football conferences on the order of approximately four to five years. This is intuitively appealing in the case of college football, because NCAA rules limit student athletes to four years of eligibility in five years. Therefore, if a football program is placed on probation and loses scholarships for a number of years, the lost scholarships would be expected to affect the program's competitiveness over time. This indicates that the full impact of enforcement and punishment might not be realized instantaneously.

The long-run impacts of enforcement and punishment can be calculated as [[beta].sub.LR] = [[beta].sub.i]/(1 - [[beta].sub.tag]), where [[beta].sub.it] is the parameter estimate of the variable in question and [[beta].sub.tag] is the parameter estimate of the lagged dependent variable (see Gujarati 1999, Ch. 14). Returning to the top panel of Table 2, the empirical results indicate that an additional football-related investigation improves the Herfindahl by 160-200 points, or by 7.6-10.3%, in the long run. On the other hand, if the average length of probations increases by a year, the long-run impact is to reduce competitive balance by increasing the Herfindahl by 36-45 points, or approximately 2%.

When using total probations as the measure of enforcement, the long-run impacts of punishment and enforcement are more pronounced. An additional probation generates a long-run improvement in competitive balance of 222-240 points, or 11.9-12.9%, ceteris paribus. The long-run impact of a one-year increase in the average length of pending probations is a reduction in competitive balance by approximately 92-105 points, or 5-6%, ceteris paribus.

The long-run impacts of NCAA enforcement and punishment on the competitive balance of collegiate football conferences have not, heretofore, been estimated. The empirical results indicate that the NCAA does influence member behavior using the credible enforcement mechanism established in 1953. Although the evidence is indirect, the improvements in competitive balance associated with enforcement indicate that increased vigilance by the NCAA increases the overall expected costs of cheating, reduces cheating, and yields a more equitable distribution of player talent within a conference. However, the improvements in competitive balance gained through enforcement do come at a cost: more severe punishments reduce competitive balance.

Thus, it is natural to question whether the benefits of enforcement activity outweigh the costs, at least in the dimension of competitive balance. Proponents of the NCAA claim that enforcement enhances competitive balance, which would be consistent with investigations]probations and punishment combining to provide a net reduction in the HHI. On the other hand, the NCAA's critics claim that enforcement protects dominant programs, which would be consistent with investigations/probations and punishment combining to have either no impact or to cause an increase in the HHI.

To formally test which characterization of the NCAA is supported by the data, two linear restrictions are tested for each specification. The first tests whether the benefits outweigh the costs calculated using the average number of investigations/probations and the average length of probations for those conferences that had at least one football-related investigation in that year (H1 in Table 2). This test restricts the sample to include only those conferences that had direct enforcement activity in a given year. The second test is more general and uses the average number of investigations/probations and length of probations for the entire sample (H2 in Table 2). This test focuses on whether enforcement and punishment have a beneficial impact on competitive balance even when no enforcement activity occurred in that conference.

As can be seen in Table 2, the evidence indicates that, in general, the benefits of investigations/ probations outweigh the costs of punishment, at least in the dimension of competitive balance. More often than not the null hypothesis that the costs are greater than or equal to the benefits of enforcement (i.e., the net effect of enforcement is a reduction in competitive balance, an increase in the HHI) can be rejected. Of those conferences with at least one football-related investigation, the null hypothesis is soundly rejected in each of the 12 specifications. On the other hand, when using the entire sample, the net benefits of enforcement and punishment are naturally lower and less statistically robust. In only 10 of the 12 specifications can the null hypothesis be rejected at conventional confidence levels. Overall, the evidence indicates that in conferences in which enforcement actually takes place, the net impact is a noticeable improvement in competitive balance. In those conferences in which enforcement does not explicitly occur, the impact of enforcement and punishment is a bit less dramatic; it either slightly improves or has no net impact on competitive balance.

Is it possible or likely for the Herfindahl to experience a three-figure decline, as suggested by the net-benefit tests presented in Table 2? Moreover, if a conference experiences a three-digit drop in its concentration, is it economically meaningful? Of the 401 conference-years in the sample, 57 correspond with an annual decline in the HHI of more than 150 points. While many of these changes were caused by conference expansion, this is not universally true. Table 3a reports the conferences that experienced an annual decline in the HHI of more than 150 points and also had at least one football-related investigation in that year. Table 3b reports those conferences with a reduction in the HHI of more than 60 points and at least one new football-related probation.

Both Table 3a and 3b also report the net change in the number of six-win teams in the conference the year the HHI dropped]5 Over the sample period, the majority of teams played either 10 or 11 games, not including conference championships and postseason bowl games. Therefore, a net increase in the number of six-win teams would not only correspond with an improvement in competitive balance but might also be important from the point of view of fans. NCAA bylaws currently require a team to have at least six wins against Division IA opponents in order to qualify for a postseason bowl. From Table 3a, the average improvement in the competitive balance of conferences with at least one football-related investigation corresponds with an average of 0.26 more six-win teams. The increase in the number of six-win teams is more pronounced in conferences in which a new football-related probation is enforced--on average, 0.52 more teams win at least six games. Thus, the impact of NCAA enforcement efforts on competitive balance is not merely a statistical artifact. The improved competitive balance often leads to additional teams becoming bowl eligible, which likely enhances fan interest.

Nevertheless, the NCAA could impose sufficiently harsh punishments such that competitive balance was reduced in net. For example, suppose one team in a conference was placed on probation for four years, yielding an average length of pending probations of fours years. If an additional team were placed on probation for, say, 16 years, the average length of probation would increase from 4 to 10 years. Using the estimation results from the bottom half of Table 2, a six-unit increase in the average length of probations would lead to a 485-point increase in the Herfindahl, whereas the gain from additional enforcement would be a 182-point reduction in the Herfindahl. Thus, the extraordinary length of the second probation would cause a net increase in the HHI of approximately 300 points, or approximately 16% of the sample-average HHI. Such an increase in the HHI would represent a dramatic reduction in the competitive balance of the conference in which the probate team played, and therefore the NCAA might avoid such draconian punishment, to include the Death Penalty.

The empirical results support the NCAA's claim that its enforcement efforts promote a more level playing field among its members. However, the empirical results also provide insight into a possible source of the concern expressed by the NCAA's critics. In a partial analysis, it is true that more severe punishments reduce competitive balance. However, these obvious reductions are generally more than offset by the less-easily observed improvements in competitive balance associated with greater NCAA vigilance, which are likely reflections of reduced cheating on the part of a conference's members.

5. Conclusions

This article extends the analysis of Fleisher, Goff, and Tollison (1992); Eckard (1998); and Depken and Wilson (2004) by presenting an investigation of the historical impact of NCAA enforcement activity on the competitive balance of a sample of 11 major Division IA college football conferences. Fleisher, Goff, and Tollison suggest that NCAA enforcement might help dominant football programs, a sentiment which Eckard seems to support. However, these claims contradict the NCAA, which insists that its enforcement efforts are aimed at promoting competitive balance. We develop empirical hypotheses that allow the data to determine which claim is correct. The behavioral model follows Becker (1968) by allowing punishment and deterrence effects to affect competitive balance.

Whereas Eckard tests for statistical differences in the variance of relative team rankings, we employ the HHI. Using both linear and lin-log specifications, competitive balance is related to once-lagged competitive balance, the number of teams in the conference, and measures of enforcement and punishment. Because of a lack of strong theoretical guidance with regard to how to measure enforcement efforts, two measures are used: the total number of football-related investigations and the total number of football-related probations in a given year. We measure the severity of punishment as the average number length of pending probations measured in years.

Our empirical analysis recognizes that if Fleisher, Goff, and Tollison (1992) are correct, it is likely that the number of investigations/probations is correlated with shocks to competitive balance--the variables are stochastic regressors. Therefore, we use the NCAA regime changes investigated by Depken and Wilson (2004) to instrument for investigations and probations. The results from various panel instrumental-variables estimators indicate that enforcement efforts tend to correlate with improved competitive balance, whereas punishment tends to correlate with reductions in competitive balance. Moreover, the full effects of enforcement and punishment are only fully dissipated after four to five years.

Statistical tests indicate that, on average, the benefits of investigations/probations tend to outweigh the costs of increased punishment, at least in the dimension of competitive balance. While we are not able to directly test the impact of investigations/probations and punishment on the amount of clandestine cheating in the NCAA, it is likely that the improvements in competitive balance that are correlated with investigations and probations are reflections of reduced cheating by NCAA members. The empirical evidence indicates that the net impact of investigations/probations and punishment is much greater, both in absolute and statistical terms, in the conferences in which investigatory behavior actually takes place. While there seems to be a small amount of positive spillover to other conferences in which NCAA enforcement efforts are not directly applied, these changes are not economically significant (although they are statistically significant).

Overall, the empirical results support the NCAA's claim that enforcement of its membership agreement enhances competitive balance. The detractors of the NCAA seemingly focus only on the obvious reduction in competitive balance that occurs when a team is put on probation and suffers a reduction in on-field success. The empirical results do indicate that teams placed on probation lose more games to above-average teams, which might be viewed as an implicit rent-protecting scheme perpetrated by the NCAA. However, the obvious reductions in competitive balance are generally more than offset by less-obvious improvements in competitive balance brought about by reduced cheating by all schools, whether these schools are on probation or not. Future analysis in this area might focus on the dollar costs of football-related probations, the efficiency of the cost of enforcement being borne by those who may have had no connection with the actions in question, and whether the impacts of football-related probations extend beyond the gridiron.

Appendix

Conferences and Teams Included in the Sample. School Membership Periods Reported in Parentheses if Different from Conference Years

Atlantic Coast Conference (1953-2003)

Clemson, Duke, Florida State (1992-2003), Georgia Tech (1983-2003), Maryland, North Carolina, North Carolina State, South Carolina (1953-1971), Virginia, Wake Forest

Big East Conference (1991-2003)

Boston College, Miami (Florida), Pittsburgh, Rutgers, Syracuse, Temple, Virginia Tech, West Virginia 

Big Ten Conference (1953-2003)

Illinois, Indiana, Iowa, Michigan, Michigan State, Minnesota, Northwestern, Ohio State, Penn State (1993-2003), Purdue, Wisconsin

Big 6-Big 7-Big 8-Big XII Conference (1953-2003)

Big 7 Conference (1948-1959):

Colorado, Iowa State, Kansas, Kansas State, Missouri, Nebraska, Oklahoma

Big 8 Conference (1960-1994):

Colorado, Nebraska, Missouri, Iowa State, Kansas, Kansas State, Oklahoma, Oklahoma State

Big XII Conference (1995-2003):

Baylor, Colorado, Iowa State, Kansas, Kansas State, Missouri, Nebraska, Oklahoma, Oklahoma State, Texas, Texas A&M, Texas Tech

Conference USA (1995-2003)

Alabama-Birmingham (1999-2003), Army (1997-2003), Cincinnati, East Carolina, Houston, Louisville, Memphis, Southern Mississippi, Texas Christian (2001-2003), Tulane

Mid-American Athletic Conference (1962-2003)

Akron (1992-2003), Ball State (1975-2003), Bowling Green, Buffalo (1999-2003), Central Florida (2002-2003), Central Michigan (1975-2003), Eastern Michigan (1977-2003), Kent. Marshall, Miami (Ohio), Northern Illinois (1975-2003), Ohio, Toledo, Western Michigan 

Mountain West Conference (1999-2003)

Air Force, Brigham Young, Colorado State, Nevada-Las Vegas, New Mexico, San Diego State, Utah, Wyoming

Pacific Coast-AAWU-Pacific 8-Pacific Ten Conference (1953-2003)

Pacific Coast (1916-1958):

California, Idaho, Oregon, Oregon State, Southern California, Stanford, UCLA, Washington, Washington State

AAWU (1959-1967):

California, Oregon (1964-1967), Oregon State (1964-1967), Southern California, Stanford, UCLA, Washington, Washington State (1962-1967)

Pacific 8 (1968-1977):

California, Oregon, Oregon Stale (1964-1977), Southern California, Stanford, UCLA, Washington, Washington State

Pacific Ten (1978-2003):

Arizona, Arizona State, California, Oregon, Oregon State, Southern California, Stanford, UCLA, Washington, Washington State

Southeastern Conference (1953-2003)

Alabama, Arkansas (1992-2003), Auburn, Florida, Georgia, Georgia Tech (1953-1963L Kentucky, Louisiana State, Mississippi, Mississippi State, South Carolina (1992-2003), Tennessee, Tulane (1953-1980), Vanderbilt

Southwest Conference (1953-1994)

Arkansas (1953-1991), Baylor, Houston (1976-1994), Rice, Southern Methodist, Texas, Texas A&M, Texas Christian, Texas Tech (1960-1994)

Western Athletic Conference (1962-2003)

Air Force (1980-1998), Arizona (1962-1977), Arizona State (1962 1977), Boise State (2001-2003), Brigham Young (1962-1998), Colorado State (1968-1998), Fresno State (1992-2003), Hawaii (1979-2003), Louisiana Tech (2001-2003), Nevada-Las Vegas (1995-1998), New Mexico (1962-1998), Rice (1995-2003). San Diego State (1978-1998), San Jose State (1995-2003), Southern Methodist (1995-2003), Texas Christian (1995-2001), Texas El Paso (1968-2003), Tulsa (1995-2003), Utah (1962-1998), Wyoming (1962-1998)

References

Becket, Gary S. 1968. Crime and punishment: An economic approach. Journal of Political Economy 76:169-217.

Block, Michael K., and Jonathan S. Feinstein. 1986. The spillover effect of antitrust enforcement. The Review of Economics and Statistics 68:122-31.

Blundell, Richard, and Stephen Bond. 1998. Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics 87:115-43.

Bound, John, David A. Jaeger, and Regina M. Baker. 1995. Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variables. Journal o[" the American Statistical Association 90:443-50.

Depken, Craig A. II. 1999. Free agency and the competitiveness of Major League Baseball. Review of Industrial Organization 14:205-17.

Depken, Craig A. II, and Dennis P. Wilson. 2004. Institutional change in the NCAA and competitive balance in intercollegiate football, in Economics of college sports, edited by John Fizel and Rodney Fort. Westport, CT: Praeger Publishers, pp. 179-210.

Eckard, E. Woodward. 1998. The NCAA cartel and competitive balance in college footbalh Review of Industrial Organization 13:347-67.

Feinberg, Robert M. 1980. Antitrust enforcement and subsequent price behavior. The Review qf Economics attd Statistics 62:609-12.

Fleisher, Arthur A. III, Brian L. Goff, William F. Shughart II, and Robert D. Tollison. 1988. Crime and punishment? Enforcement of the NCAA football cartel. Journal of Economic Behavior and Organization 10:433-51.

Fleisher, Arthur A. III, Brian L. Goff, and Robert D. Tollison. 1992. The National Collegiate Athletic Association: A study in cartel behavior. Chicago: University of Chicago Press.

Fort, Rodney, and James Quirk. 1995. Cross-subsidization, incentives and outcomes in professional team sports leagues. Journal of Economic Literature 33:1265-99.

Garbade, Kenneth D., William L. Silber, and Lawrence J. White. 1982. Market reaction to the filling of antitrust suits: An aggregate and cross-sectional analysis. The Review of Economics and Statistics 64:686-91.

Grogger, Jeffrey. 1991. Certainty vs, severity of punishment. Economic Inquiry, 29:297-309.

Gujarati, Damodar. 1999, Essentials of econometrics. 2nd edition. Boston: Irwin/McGraw Hill.

Hadi, Ali S. 1992. Identifying multiple outliers in multivariate data. Journal q[" the Royal Statistical Society, Series B (Methodological) 54:761-71.

Hadi, Ali S. 1994. A modification of a method for the detection of outliers in multivariate samples. Journal of the Royal Statistical Society, Series B (Methodological) 56:393-6.

Humphreys, Brad R. 2002. Alternative measures of competitive balance in sports leagues. Journal of Sports Economics 3: 133-48.

Levitt, Steven D. 1998. "Why do increased arrest rates appear to reduce crime: Deterrence, incapacitation, or measurement error?" Economic Inquiry 36:353-72.

National Collegiate Athletic Association. Major Infractions Database. Accessed March 2004. Available http://www.ncaa.org.

Nickell, Stephen. 1981. Biases in dynamic models with fixed effects. Econometrica 46:1417-26.

Nissan, Edward. 1998. Effects of antitrust enforcement on aggregate concentration. Journal of Economic Studies 25:112-7.

Noll, Roger. 1999. The business of college sports and the high cost of winning. The Milken Institute Review Third Quarter: 24-37.

Zimbalist, Andrew. 2001. Unpaid professionals: Commercialism and conflict in big-time college sports. Princeton: Princeton University Press.

(1) The other organizations are the National Association of Intercollegiate Athletics (NAIA), the National Christian College Athletic Association (NCCAA) and the National Junior College Athletic Association (NJCAA).

(2) The history of the NCAA has been well documented by other authors, (e.g., Fleisher, Goff, and Tollison [1992], Eckard [1998], Depken and Wilson [2004], and the NCAA itself).

(3) In 1950, the Council of the NCAA and the Executive Committee attempted to terminate the NCAA memberships of Boston College, the Citadel, the University of Maryland, the University of Virginia, Villanova University, Virginia Military Institute, and Virginia Tech for violation of the Sanity Code. However, a membership vote overturned the recommended termination.

(4) NCAA rules hold for all intercollegiate sports. While this article focuses on college football, the majority of NCAA infractions occur in other sports.

(5) In 2003, there were 28 post-season bowl games, eight more than were
played in 1997. The following table presents select College Bowl Game
Payouts in 1997 and 2003:

                                            2003 Payout    1997 Payout
Bowl Game                  Site              per Team       per Team

Fiesta            Tempe, Arizona             $14-17 M        $9.45 M
Rose              Pasadena, California       $14-17 M        $11.60 M
Sugar             New Orleans, Louisiana     $14-17 M        $9.45 M
Orange            Miami, Florida             $14-17 M        $9.78 M
Florida Citrus    Orlando, Florida           $5.1 M          $3.98 M
Outback           Tampa, Florida             $2.65 M         $1.93 M
Holiday           San Diego, California      $2 M            $1.61 M

1997 payouts are measured in 2003 dollars. In 2003, the Florida Citrus
Bowl was renamed the Capital One Bowl.

Teams participating in bowl games retain a percentage of the per-team
bowl payout, the remainder shared with the other schools in the
participant's conference.

(6) The number of high profile whistle-blowers has increased over the past 20 years. Notable examples include Jan Kemp at the University of Georgia in 1986 (academic fraud); teachers and tutors at the University of Tennessee in 1998 (academic fraud); academic tutors at the University of Minnesota in 1999 (academic fraud); high school football coaches affiliated with the University of Alabama and the University of Kentucky in 2001 (bribery); and disgruntled players at the University of Georgia in 2003 (academic fraud and bribery). These instances indicate the willingness for people to report significant violations of NCAA regulations, even if they lack passionate interest in the success of a particular team or sport.

(7) Our normalization is only one possibility. Eckard (1998) awards one point lbr a win and half a point for a tie. The impact our normalization is to scale the parameter estimates: statistical inference is lint affected.

(8) Alternative measures of competitive balance focus on changes in the relative standings of a conlerence's teams. Eckard (1998) and Humphreys (2002) use the intertemporal variation in team position as a measure of competitive balance, wherein greater variance indicates better competitive balance.

(9) The two measures of enforcement are not used in the same estimation because their correlation is approximately 0.75.

(10) The season champion of each BCS Conference earns an automatic berth to one of the four BCS bowl games, from which the national championship is determined. The BCS Conferences are the Southeastern, Big East, Big Ten, Big XII, Atlantic Coast, and Pacific Ten, along with Notre Dame.

(11) The data were gathered from the schools themselves, newspapers, media guides, and the NCAA. They are available online at http://www.cae.wisc.edu/~dwilson, last accessed December 2003.

(12) The Big 6 existed from 1928 through 1947 and included Iowa State, Kansas. Kansas State, Missouri, Nebraska and Oklahoma. The Big 7 existed from 1948 through 1959 and included the same teams as the Big 6 with the addition of Colorado. The Big 8, which began in 1960 and lasted through 1994, included the same teams as the Big 7 with the addition of Oklahoma State. These eight teams joined four teams from the defunct Southwest Conference to form the Big XII in 1995.

(13) Data were also gathered on new football investigations and the different types of punishments imposed on football programs including loss of scholarships, loss of postseason play, and loss of television appearances. However, models including various types of penalties were not statistically different from the results presented here.

(14) Schools often self-report and self-punish secondary infractions, which are generally trivial violations such as a player borrowing change from a coach to use a pay phone (which violates NCAA bylaw 16.12.2.2).

(15) In this case, the number of wins includes non-conference wins.

Craig A. Depken II * and Dennis P. Wilson ([dagger])

* Department of Economics, University of Texas-Arlington, Arlington, TX 76019, USA; E-mail [e-mail].

([dagger]) Department of Economics, University of Texas-Arlington, Arlington, TX 76019, USA; E-mail [e-mail]; corresponding author.

We appreciate the helpful comments of two anonymous referees on a previous version of the manuscript. Any remaining errors are the sole responsibility of the authors.

Received October 2004; accepted July 2005.

Table 1. Descriptive Statistics of the Data

                                          Standard
Variable                        Mean      Deviation   Minimum   Maximum

Performance points
  Herfindahl (HHI)             1488.34     318.65      755.03   3000.00
Log of HHI                        7.28       0.20        6.62      8.01
Number of investigations (a)      0.54       0.79        0.00      4.00
Number of investigations (b)      1.39       0.66        0.00      4.00
Total number of probations        0.56       0.82        0.00      5.00
  (a)
Total number of probations
  pending (b)                     1.08       0.94        1.00      5.00
Average length of probations
  (a)                             0.98       1.40        0.00      5.00
Average length of probations
  pending (b)                     1.68       1.39        0.00      5.00
Number of teams in                9.13       1.72        5.00     16.00
  conference
Atlantic Coast Conference         0.123      0.33        0.00      1.00
Big East Conference               0.03       0.17        0.00      1.00
Big 10 Conference                 0.12       0.33        0.00      1.00
Big 7-Big 8-Big XII               0.13       0.33        0.00      1.00
  Conference
Conference USA                    0.02       0.14        0.00      1.00
Mid-American Athletic
  Conference                      0.10       0.31        0.00      1.00
Mountain West Conference          0.01       0.10        0.00      1.00
Pacific Coast-AAWU-Pacific
  8-Pacific 10 Conference         0.13       0.33        0.00      1.00
Southeastern Conference           0.13       0.34        0.00      1.00
Southwest Conference              0.10       0.31        0.00      1.00
Western Athletic Conference       0.10       0.30        0.00      1.00

Sample includes 401 observations describing 16 major Division IA
football conferences from 1953 to 2003; see Appendix for details.

(a) Based on entire sample.

(b) Based on 156 conference-year observations in which at least one
football-related investigation took place. The means of conference
dummy variables do not sum to 1 because of rounding.

Table 2. Instrumental Variables Panel Estimation Results

                                     Dependent Variable: PPHI

Independent Variable         IV-2SLS        IV-GMM         DPD

Lagged dependent               0.219 *       0.215 *       0.255 *
  variable                    (0.06)        (0.06)        (0.04)
Teams in conference         -120.709 *    -120.722 *    -113.219 *
                             (11.66)       (11.58)       (10.73)
Number of                   -155.051 *    -156.109 *    -113.466 *
  Investigations             (59.27)       (58.54)       (45.92)
Average length of             35.220 *      35.041 *      27.345 *
  probation                  (12.37)       (12.19)       (10.79)
R-squared                      0.761         0.76          --
First-stage F-stat on
  excluded instruments         2.19 *        2.19 *        --
Hansen J-statistic
  of overidentifying
  restrictions                 0.84          0.84          3.25
F-test: zero slopes           85.08 *       85.50 *      265.98 *
H1: Net effect of           -157.44 *     -159.216      -112.57 *
  average enforcement         [0.02]        [0.01]        [0.04]
  efforts on
  competitive balance
  measure (conferences
  with at least one
  investigation) (a)
H2: Net effect of            -49.62 *      -50.37 *      -34.76 *
  average enforcement         [0.03]        [0.02]        [0.07]
  efforts on competitive
  balance measure
  (entire sample) (b)
Lagged dependent               0.240 *       0.243 *       0.274 *
  variable                    (0.05)        (0.05)        (0.05)
Teams in conference         -123.217 *    -121.949 *    -113.686 *
                             (12.10)       (11.79)       (12.31)
Number of                   -182.536 *    -178.446 *    -161.455 *
  probations                 (63.86)       (62.81)       (68.64)
Average length of             80.922 *      79.082 *      67.461 *
  probation                  (27.00)       (26.39)       (27.93)
R-squared                      0.773         0.778         --
First-stage F-stat
  on excluded
  instruments                  2.89 *        2.89 *        --
Hansen J-statistic
  of overidentifying
  restrictions                 0.65          0.65          3.06
F-test: zero slopes           84.27 *       86.09 *       75.12 *
H1: Net effect of            -61.63 *      -60.30 *      -61.44 *
  average enforcement         [0.03]        [0.03 ]       [0.08]
  efforts on competitive
  balance measure
  (conferences
  with at least one
  investigation) (a)
H2: Net effect of            -21.85 **     -21.39 **     -23.38
  average enforcement         [0.08]        [0.08]        [0.13]
  efforts on
  competitive balance
  measure
  (entire sample) (b)

                                  Dependent Variable: Log PPHI

Independent Variable         IV-2SLS        IV-GMM         DPD

Lagged dependent              0.183 *       0.179 *       0.231 *
  variable                   (0.09)        (0.02)        (0.04)
Teams in conference          -0.087 *      -0.087 *      -0.081 *
                             (0.006)       (0.006)       (0.01)
Number of                    -0.081 *      -0.085 *      -0.058 *
  Investigations             (0.03)        (0.03)        (0.02)
Average length of             0.019 *       0.019 *       0.015 *
  probation                  (0.01)        (0.01)        (0.01)
R-squared                     0.821         0.806         --
First-stage F-stat on
  excluded instruments        2.36 *        2.36 *        --
Hansen J-statistic
  of overidentifying
  restrictions                1.04          1.04          3.86
F-test: zero slopes         139.03 *      139.39 *      530.21 *
H1: Net effect of            -0.081 *      -0.085 *      -0.056 *
  average enforcement        [0.03]        [0.02]        [0.05]
  efforts on
  competitive balance
  measure (conferences
  with at least one
  investigation) (a)
H2: Net effect of            -0.025 *      -0.027 *      -0.017 **
  average enforcement        [0.05]        [0.04]        [0.09]
  efforts on competitive
  balance measure
  (entire sample) (b)
Lagged dependent              0.204 *       0.204 *       0.247 *
  variable                   (0.05)        (0.05)        (0.05)
Teams in conference          -0.088 *      -0.088 *      -0.081 *
                             (0.01)        (0.01)        (0.01)
Number of                    -0.103 *      -0.102 *      -0.090 *
  probations                 (0.04)        (0.04)        (0.04)
Average length of             0.046 *       0.046 *       0.039 *
  probation                  (0.02)        (0.02)        (0.02)
R-squared                     0.815         0.816         --
First-stage F-stat
  on excluded
  instruments                 2.94 *        2.94 *        --
Hansen J-statistic
  of overidentifying
  restrictions                0.94          0.94          3.78
F-test: zero slopes         120.67 *      122.47 *      799.54 *
H1: Net effect of            -0.034 *      -0.033 *      -0.032 **
  average enforcement        [0.04]        [0.04]        [0.08]
  efforts on competitive
  balance measure
  (conferences
  with at least one
  investigation) (a)
H2: Net effect of            -0.012 **     -0.012 **     -0.012
  average enforcement        [0.10]        [0.091        [0.13]
  efforts on
  competitive balance
  measure
  (entire sample) (b)

All specifications include conference fixed effects and a dummy
variable to control for four outliers, determined using the method
proposed by Hadi (1992, 1994). Number of investigations and number of
probations instrumented with major regulatory changes in the NCAA
membership agreement and whether a conference is a BCS Conference; see
text for details. IV-2SLS and IV-GMM are
heteroscedastic-autocorrelation (HAC) consistent estimates. DPD are
Blundell and Bond (1998) Dynamic Panel Data estimates.

(a) Net benefits calculated using average number of investigations or
probations and average length of probations among sample conferences in
which at least one investigation took place (see Table 1).

(b) Net benefits calculated using the average number of investigations
or probations and average length of probations for the entire sample
(see Table 1).

* Indicates significance at the 5% and ** 10% level in a two-tailed
test. Robust standard errors reported in parentheses; two-tailed p
values reported in brackets.

Table 3a. Sample Conferences with One-Year Decline in Herfindahl, More
than 150 Points, and at Least One Football-Related Investigation

Conference          Year     HHI      [DELTA]HHI

Atlantic Coast      1955   1759.003    -231.737
Atlantic Coast      1961   1331.361    -244.083
Atlantic Coast      1983   1632.373    -385.767
Atlantic Coast      1992   1435.185    -191.090
Big 8               1972   1511.480    -248.724
Big 8               1981   1549.745    -165.816
Big 8               1992   1485.969    -274.234
Big Ten             1969   1248.980    -163.265
Big Ten             1984   1207.407    -182.716
Conference USA      1998   1666.667    -399.491
Mid-American        1969   1791.383    -260.771
Mid-American        1980   1145.125    -244.224
Pacific 10 (AAWU)   1964   1675.900    -546.322
Southeastern        1992   1064.140    -282.798
Southwest           1957   1780.045    -283.446
Southwest           1974   1575.255    -153.061
Southwest           1976   1469.907    -239.276
Southwest           1989   1435.185    -325.018
Western Athletic    1969   1712.000    -197.262
Averages                   1498.795    -264.163

                                   Change in
Conference          %[DELTA]HHI   Six-Win Teams

Atlantic Coast           -11.64       0
Atlantic Coast           -15.49      -2
Atlantic Coast           -19.11      -1
Atlantic Coast           -11.75       0
Big 8                    -14.13      +1
Big 8                     -9.66      +1
Big 8                    -15.57      -2
Big Ten                  -11.56      -2
Big Ten                  -13.14      +2
Conference USA           -19.33      +1
Mid-American             -12.71       0
Mid-American             -17.57       0
Pacific 10 (AAWU)        -24.58      +2
Southeastern             -20.99      +1
Southwest                -13.73       0
Southwest                 -8.85      +3
Southwest                -13.99       0
Southwest                -18.46       0
Western Athletic         -10.33      +1
Averages                 -14.87      +0.26

Four other observations, not reported here, corresponded
with conferences that experienced an annual drop in the HHI of
between 140 and 150 points. The change in six-win teams
reflects total victories, including nonconference and non-Division
1A opponents. During the sample period, teams played either 10 or
11 games (not including conference championship and postseason
bowl games). Therefore, six wins would correspond to an above-average
team.

Table 3b. Sample Conferences with One-Year Decline in Herfindahl
Greater than 60 Points and at Least One New Football-Related Probation

Conference       Year     HHI      [DELTA]HHI

Atlantic Coast   1955   1759.003     -231.738
Atlantic Coast   2003   1327.161     -123.45
Big East         1994   1639.031     -146.687
Big East         2003   1658.163      -76.530
Big 8            1970   1607.143      -76.530
Big 8            1972   1511.480     -249.725
Big 8            1981   1549.745     -165.816
Big Ten          1974   1296.875      -87.50
Big Ten          1999   1146.694      -92.975
Mid-American     1969   1791.383     -260.771
Mid-American     2002   969.5291      -110.71
Pacific Ten      2002   1237.500      -87.500
Southeastern     1981   1378.772     -109.261
Southeastern     1990   1269.388      -77.551
Southwest        1957   1780.045     -283.446
Southwest        1958   1700.680      -79.365
Southwest        1966   1530.612     -102.041
Southwest        1974   1575.255     -153.061
Southwest        1976   1469.907     -239.276
Southwest        1989   1435.185     -325.019
Western
  Athletic       1997   755.0296     -108.14
Averages                1447.075     -151.766

                                 Change in
Conference       %[DELTA]HHI   Six-Win Teams

Atlantic Coast      -11.64            0
Atlantic Coast       -8.51           -1
Big East             -8.21            0
Big East             -4.41           +2
Big 8                -4.54            0
Big 8               -14.13           +2
Big 8                 9.66           +1
Big Ten              -6.32           +2
Big Ten              -7.50           +2
Mid-American        -12.71            0
Mid-American        -10.25           +1
Pacific Ten          -6.60           +2
Southeastern         -7.34            0
Southeastern         -5.75           -2
Southwest           -13.73            0
Southwest            -4.45           -1
Southwest            -6.25           -1
Southwest            -8.85           +3
Southwest           -13.99            0
Southwest            18.46            0
Western
  Athletic          -12.52           +1
Averages             -6.64           +0.52

The change in six-win teams reflects total victories, including
nonconference and non-Division IA opponents. During the sample
period teams played either 10 or 11 games (not including conference
championship and postseason bowl games). Therefore, six wins would
correspond to an above-average team.