From the hardwood to the gridiron to the dorm: influences on attendance to women's collegiate basketball.

Depken, Craig A., II ; Williams, Courtney ; Wilson, Dennis P. 等

This paper investigates factors that influence attendance to Division I women's collegiate basketball programs. The focus on women's basketball is warranted for several reasons. First, while women's basketball has become considerably more popular over the past 25 years, the literature focusing on the sport is considerably smaller than for other collegiate sports. Second, the extent to which factors found to influence attendance to college football and men's basketball games also impact the attendance to women's basketball games has not been investigated. Third, the extent to which college football enhances or detracts from attendance to women's basketball, and to other sports, has not been investigated.

While women's basketball is traditionally considered a non-net-revenue sport, the relative popularity of the sport over time has unquestionably increased. As evidence, in 1982 there were 272 Division I women's basketball teams, whereas in 2009 there were 343 teams. In 1982, the 32 teams that participated in the first women's NCAA tournament played to an average in-arena audience of 2,166 people. In the 2005 tournament, 64 teams played to the highest average in-arena audience of 6,520; in 2009 average in-arena attendance to the tournament had fallen to 5,213 (NCAA, 2010b).

The increased attendance to the women's NCAA tournament suggests that the game is improving in popularity at its highest level but does not directly address the influences on attendance at the team level. Using an unbalanced panel describing 324 schools with Division I women's basketball programs from 2000-2009, the empirical strategy relates regular season per-game attendance to a number of covariates, including women's basketball team quality, recent and past post-season appearances and success by the women's basketball team, various institutional characteristics, whether the school in question has a football team, conference affiliation, and a time trend.

A potentially confounding issue is that attendance to women's basketball might be influenced by the overall sport culture or "school spirit" on campus, which, while admittedly difficult to measure, might be correlated with whether a school has a football team. Thus, having a football team might be endogenous to the estimation, for which instrumental variables is an appropriate remedy. We use the age of the institution to instrument for having a football program and find that ordinary least squares results in estimates that are biased down.

To preview the results, contemporaneous and recent team quality positively influence per-game attendance to women's basketball programs, as does recent participation in the NCAA women's tournament (but not the National Invitational Tournament), winning games in the previous season's NCAA and NIT tournaments (with NCAA tournament wins having a greater impact), and the number of NCAA and NIT tournaments in which the team participated over the past seven years. Women's basketball attendance is lower at private and urban schools, at schools with a greater proportion of female students, and at schools with more students. Many of these influences are qualitatively (if not quantitatively) similar to findings by Depken (forthcoming) in Division I collegiate men's basketball.

Although results from the pooled sample of all Division I schools suggest that football is complementary to women's basketball on average, the pooled data combine high-profile and very popular programs such as Tennessee and Connecticut with low-profile and considerably less popular programs such as Nicholls State and Texas Southern. To test for further differences across school size, the pooled sample is split into various subsamples. The first is between basketball teams that play in Bowl Championship Series affiliated conferences and those that do not; football appears to be a complement to women's basketball in BCS affiliated conferences but not for teams in non-BCS affiliated conferences. A second approach sorts schools into quintiles based on the number of students enrolled; for the biggest and smallest schools, football appears to be a complement to women's basketball whereas for mid-size schools, football and women's basketball appear to be independent of each other.

College Football, College Basketball, and the Academy

The literature focusing on women's basketball is quite small and underdeveloped. Kerstetter and Kovich (1997) investigate the profile of Division I women's basketball spectators but do not estimate attendance models. Hallmark and Armstrong (1999) investigate gender equity issues in the media's coverage of the women's and men's NCAA basketball tournaments and find that women's broadcasts use fewer camera shots and graphics. Madrigal and James (1999) investigate home-court advantage in the Big 10 conference and find that higher quality teams have a greater home-court advantage than their lower-quality counterparts; however, they do not relate team quality to game attendance. Brown and Jewell (2006) estimate the MRP of women collegiate basketball players and find that the best players might generate approximately $250,000 for high quality teams, but good players on poor teams generate no rents for their programs; however, their study does not directly investigate attendance to women's basketball games.

The literature investigating the impact of college football on campus is substantially more wide-ranging. (1) Price and Sen (2003) estimate models of game-day attendance for college football games while Kaempfer and Pacey (1986) and Fizel and Bennet (1989) investigate the impact of televised games on attendance. Other authors have investigated whether football success influences alumni donations (Baade & Sundberg, 1996; Rhoads & Gerking, 2000), funding from state legislatures (Humphreys, 2006), the number of applications (McEvoy, 2005; Toma & Cross, 1998; Murphy & Trandel, 1994), and the quality of incoming freshmen (McCormick & Tinsley, 1987; Tucker & Amato, 1993; Tucker, 2005). Some have argued that football might act as a substitute for academic pursuits, which might be reflected in lower graduation rates (Tucker, 1992; Magnold, Bean, & Adams, 2003), while others have found the opposite for the general student body (Tucker, 2004) and for football athletes themselves (Matheson, 2007). Borland, Goff, and Pulsinelli (1992) consider the internal value of a football program to a university.

The local economic impacts of college football have also been investigated. Coates and Depken (2009, forthcoming) study the effect of college football games on taxable sales in several cities in Texas and find little net increase in tax revenue from hosting a college football game. Their findings support those of Baade, Baumann, and Matheson (2007, 2008) who also find no statistical evidence that college football games contribute a net positive to the home team's local economy.

While the literature on college football is substantial, there has been relatively little investigation into how college football influences other sports on campus. Rishe (1999) found that more profitable football programs tend to spend more on women's athletics (per athlete), thus suggesting that football has a positive spillover to other sports on campus at least in the dimension of institutional funding; this evidence is reinforced by Agthe and Billings (2000). Thus, while there is some evidence of pecuniary spillovers from football to non-net-revenue generating sports, it has not been directly established that having a football program increases attendance to other sports on campus.

Empirical Specification and Data

The empirical strategy relates per-game attendance to women's basketball games to the current and recent women's basketball team quality, various institutional characteristics, and the existence of a football program. The potential influences on attendance to women's basketball are similar to those used to help explain attendance to men's collegiate basketball games in Depken (forthcoming). The full specification is:

ln[ATT.sub.it] = [[beta].sub.0] + [[beta].sub.1][WBWPCT.sub.it] + [[beta].sub.2][LAGWBWPCT.sub.it] + [[beta].sub.3][LAGWNCAA.sub.it] + [[beta].sub.4][LAGWNIT.sub.it] + [[beta].sub.5][LAGWNCAAWINS.sub.it] + [[beta].sub.6][LAGWNITWINS.sub.it] + [[beta].sub.7][LAGWNCAACHAMP.sub.it] + [[beta].sub.8][LAGWNITCHAMP.sub.it] + [[beta].sub.9][WNCAAWNITPREV7.sub.it] + [[beta].sub.10][URBAN.sub.i] + [[beta].sub.11][PRIVATE.sub.i] + [[beta].sub.12][PCTFEMALE.sub.it] + [[beta].sub.13][lnENROLL.sub.it] + [[beta].sub.14][FOOTBALL.sub.it] + [lambda]CONF + [phi]TIME + [epsilon]it,

where [lnATT.sub.it] is the natural logarithm of per-game attendance for school i during season t, the [beta]'s, [lambda]'s, and [phi]'s are parameters to be estimated, and 8 is a zero-mean stochastic error term.

The independent variables include the team's winning percentage from the current basketball season (WBWPCT), the previous season's winning percentage (LAGWBWPCT), whether the program participated in the previous season's women's NCAA basketball tournament (LAGWNCAA) or the previous season's women's National Invitational Tournament (LAGWNIT), the total number wins the team had in each tournament (LAGWNCAAWINS and LAGWNITWINS), two dummy variables that take a value of 1 if the team won the previous season's NCAA or the NIT tournament (LAGWNCAACHAMP and LAGWNITCHAMP), and the total number of combined NCAA/NIT tournaments in which the team participated over the past seven years (WNCAAWNITPREV7).

The next four variables control for the impact of institutional characteristics on attendance at women's basketball games and include a dichotomous variable which takes a value of 1 if the school is characterized as being urban by the Department of Education (URBAN), a dichotomous variable which takes a value of 1 if the school is private (PRIVATE), the percentage of the student population that is female (PCTFE-MALE), and the natural logarithm of the total student enrollment of the school (lnENROLL).

To control for the impact of football on women's basketball attendance, a dichotomous variable which takes a value of 1 if the school has a football program (FOOTBALL) is included. Finally, we include a vector of women's basketball conference dummy variables (CONF) to control for time-invariant unobserved heterogeneity across different conferences and a vector of year dummy variables (TIME) to control for unmeasured conference-invariant time-based influences on the attendance to all women's basketball teams.

Apriori, it is anticipated that higher quality basketball programs attract greater per-game attendance; therefore, the parameter estimates [[beta].sub.1] and [[beta].sub.2] are expected to be positive. Participating in the previous season's women's NCAA tournament is anticipated to have a non-negative impact on attendance as is participating in the previous season's NIT tournament. However, if the NCAA tournament is more prestigious then [[beta].sub.3] is expected to be greater than [[beta].sub.4]. If success in post-season play contributes to attendance beyond just participating, then [[beta].sub.5] and [[beta].sub.6] are expected to be positive, and if winning in the NCAA is more important than winning in the NIT then [[beta].sub.5] is expected to be greater than [[beta].sub.6]. If winning either tournament increases attendance in the next season then [[beta].sub.7] and [[beta].sub.8] are expected to be positive, and if winning the NCAA tournament is more prestigious, then [[beta].sub.7] is expected to be greater than pg. Finally, if attendance is enhanced by a history of participating in post-season play, [[beta].sub.9] is expected to be positive.

The impacts of the various school characteristics on per-game attendance are empirical questions. Urban schools might enjoy greater attendance as they are located in more densely populated areas where transaction costs of attendance might be lower for non-students; if this is the case then [[beta].sub.10] is expected to be positive. Yet, urban schools might compete with a greater number of entertainment substitutes for women's basketball. Private schools tend to be smaller and might have student bodies that are less interested in intercollegiate sports, including women's basketball, compared with public schools; if this is the case [[beta].sub.11] will be negative. If a greater percentage of the student body that is female increases interest in female sports, [[beta].sub.12] will be positive. On the other hand, there is a perception that women are generally less interested in sports and therefore a greater percentage of female students might decrease per-game attendance--that is, [[beta].sub.12] will be negative. Finally, larger schools might enjoy greater per-game attendance through a direct demand effect, in which case [[beta].sub.13] will be positive, but might also have student bodies engaged in a more diverse portfolio of extracurricular activities which might reduce attendance, in which case [p.sub.13] will be negative.

If football is a net complement (substitute) to women's basketball, then the existence of a football program on campus is expected to increase attendance to women's basketball games and [[beta].sub.14] will be positive (negative). On the other hand, if men's football and women's basketball are segmented, i.e., independent of each other, then [[beta].sub.14] will be insignificant.

Data

Data conducive to investigating attendance to women's basketball programs were gathered from the National Collegiate Athletic Association, the Integrated Post-Secondary Education Data System (IPEDS), and from individual institutions. The data represent an unbalanced panel of 324 universities and colleges that played NCAA Division I women's basketball between 2000-2009.2 The descriptive statistics of the data are reported in Table 1.

From Table 1, the average per-game attendance during the seasons under investigation was 1,499 but had considerable variation across schools; the lowest average attendance was 109 people per game (Nicholls State in 2006) and the highest average attendance was 15,796 (University of Tennessee in 2007). Approximately 20% of the women's programs participated in the previous year's women's NCAA basketball tournament or women's National Invitational Tournament (NIT). Among the schools in the sample, the average program participated in approximately two post-season tournaments over the respective previous seven years.

Approximately 73% of the observations correspond with a school with a football team, although only 36% corresponded with a school that had a football team playing in the Football Bowl Subdivision, the highest level of college football. Considering the school characteristics, 26% of the schools are considered urban according to IPEDS, 33% are private, the average school was 55% female, and the average enrollment was approximately 16,000 students.

[FIGURE 1 OMITTED]

Figure 1 provides a scatter plot of per-game attendance against women's basketball team quality. It is apparent that some observations appear to be outliers; failure to control for outlier status could bias the estimation results. Most but not all of these apparent outliers are associated with teams that play in one of the so-called Big Six conferences (Atlantic Coast, Big East, Big Ten, Big Twelve, Southeastern, or Pacific Ten). (3) Although conference-specific fixed effects are included in the estimating equations, the variable (OUTLIER) was created to take a value of one if an observation is identified as a multivariate outlier according to the procedure developed by Hadi (1992, 1994). (4) This variable is included as a separate regressor to further control for unobserved heterogeneity specific to the schools with extraordinary levels of attendance.

Estimation Results

The initial empirical results are based on ordinary least squares with White (1980) standard errors robust to heteroscedasticity applied to the pooled sample of all 324 teams; the results are reported in Table 2. (5) All specifications in Table 2 include conference-specific fixed effects and initially included year dummy variables which were dropped because they were jointly insignificant. Model (1) in Table 2 is a parsimonious model which includes only variables describing the women's basketball team. The greater the current-season winning percentage, the greater the per-game attendance, as expected. Additionally, the higher the previous season's winning percentage the greater per-game attendance. These results are consistent with the findings of Depken (forthcoming) in men's collegiate basketball and the general sports economic literature. Interestingly, playing in the previous season's NCAA tournament does not directly influence per-game attendance, although playing in the women's NIT tournament corresponds with a reduction in per-game attendance. However, winning games in either the previous year's women's NCAA or NIT tournament does increase per-game attendance; each win in the NCAA (NIT) tournament increases per-game attendance by approximately 13% (10%), although a team needs to win two NIT games to overcome the direct negative effect of participating in that tournament. Winning the NIT championship does not influence per-game attendance whereas winning the NCAA tournament corresponds with an approximately 40% increase in attendance the next season. (6) Finally, having established a history of participating in post-season events, as reflected in the total number of women's NCAA or NIT tournaments over the previous seven years, contributes to increased per-game attendance.

Model (2) in Table 2 adds several school characteristics. Urban schools have attendance which is about 6% lower than their non-urban counterparts and private schools have about 17% less attendance than their public counterparts. Schools with greater proportions of female students and schools with greater numbers of students also have lower per-game attendance. The basketball oriented variables all have essentially the same magnitudes and significance as in Model (1), the only difference being that in Model (2) the direct impact of participating in the previous season's women's NCAA tournament is marginally significant.

Model (3) in Table 2 includes a dummy variable that takes a value of one if the team was identified as a multivariate outlier. Those teams that are outliers have approximately 82% greater per-game attendance, all else equal. (7) The other parameters do not noticeably shift once controlling for the outlier status.

Model (4) in Table 2 adds a dummy variable for the presence of a Division I football team, regardless of whether the team plays at the Football Bowl Subdivision (FBS; formerly Division I-A) or the Football Championship Subdivision (FCS; formerly Division I-AA) level. While the parameters on the previously included regressors do not change appreciably in magnitude or significance, the parameter on FOOTBALL is positive and statistically significant; schools with a football team enjoy per-game attendance that is 8.62% greater than schools without a football program.

However, there are two potential problems with the pooled results reported in Table 2. First, the results might be biased if the presence of a football team is correlated with the OLS error term. This endogeneity bias would arise if schools with greater interest in sports are more likely to have a football program and also have greater attendance to other sports, including women's basketball. A second concern is that, although the OLS results include the outlier dummy variable and conference-specific fixed effects, the pooling of teams of such diverse size and popularity such as Tennessee and Nicholls State is not appropriate.

Concern #1: Endogeneity

The first concern is addressed by considering the dichotomous variable FOOTBALL as an endogenous treatment for which instrumental variables is appropriate. In this case the standard two-stage least squares estimator is not appropriate. Rather, a Heckman-like approach is taken wherein the endogenous variable FOOTBALL is modeled as a probit, the outcome variable lnATT is modeled as a linear equation, and both are estimated simultaneously using Stata's treatreg command.

Like other Heckman-type models, it is preferable to have at least one variable that appears in the treatment equation that does not appear in the outcome equation. For a variable to be a valid instrument it must be correlated with the endogenous variable, in this case having a football program, but not correlated with the primary variable of interest, in this case attendance at women's basketball games. We propose the age of the institution as an instrument for having a football program (AGE). It is not anticipated that attendance to women's basketball games is correlated with the age of the institution. However, in the past, starting a football team entailed significantly lower start-up and maintenance costs than today, both in nominal and real terms. (8) Because starting a football team today entails considerable costs, including stadium and practice facilities, coaching staffs, and Title IX considerations, relatively younger institutions may be less likely to have a football program. Thus, the age of the institution is expected to be (positively) correlated with having a football team on campus, although there is no reason to expect that the age of the institution is correlated with attendance to women's basketball games.

A second instrument is the number of other schools within the state that have a football program, either playing in the FBS or FCS (TOTFBSCHOOLS). It is not anticipated that attendance to women's basketball games is correlated with the number of other in-state football programs. However, the number of other in-state football programs might be (negatively) correlated with the odds that a particular institution has a football program. (9) More football programs in a given state might reduce each institution's actual and potential fan base, increase competition for in-state recruits, and might cause increased expenditure on facilities and coaching staffs. All of these issues might reduce the probability that a school would maintain or start a football program.

Table 3 presents estimation results from several probit specifications that vary by the number of variables included. In the final specification the two variables discussed above (AGE and TOTFBSCHOOLS) and two variables included in the outcome equation also thought to influence the odds of having a football team are included: whether the school is urban (URBAN) and the percentage of the student body that are females (PCTFEMALE). The results in Table 3 suggest older schools are more likely to have a football team, all else equal, but that schools in states with more football programs, urban schools, and schools with a larger proportion of female students are less likely to have a football program. The last column in Table 3 provides the marginal probabilities associated with each of the dependent variables; for instance, urban schools are approximately 18% less likely to have a Division I football program, all else equal.

Table 4 reports the estimates of the treatment models using various groups of variables in the treatment equation. (10) Model (1) in Table 4 includes only institution age as an explanatory variable for the existence of a Division I football program; in this specification all variables retain their significance and magnitude except for LAGWNCAACHAMP and FOOTBALL, which are now statistically insignificant (although the parameter on FOOTBALL is roughly the same magnitude as in the pooled OLS results). However, as more variables are added to the treatment equation, the parameter on FOOTBALL becomes significant and increases in magnitude. The various models suggest that schools with football programs enjoy as much as 27% greater per-game attendance than non-football schools, all else equal.

Concern #2: Pooled Data

While it seems that the impact of football on per-game women's basketball attendance might be biased down in pooled OLS, a second concern centers on whether the data should be pooled. Is it true that all schools with a football team, no matter how large or popular, experience the same increase in attendance or is the result being driven by a select number of Division I schools? To address this concern, the sample is initially split into teams that play basketball in a Bowl Championship Series (BCS)-affiliated conference and those that do not. (11) This distinction might seem somewhat ad hoc as there are only five non-football schools that play basketball in a BCS-affiliated conference (DePaul, Marquette, Providence, Seton Hall, and St. John's, which all play in the Big East Conference). However, it is a commonplace distinction in modern discussion concerning college sports and is therefore the initial distinction made here.

The results for the BCS and non-BCS conference subsamples are reported as Model (4) and Model (5) in Table 4. For BCS-conference teams, lagged winning percentage does not have a statistically meaningful relationship with current-season per-game attendance; however for non-BCS conference teams lagged winning percentage retains its importance, both economically and statistically. Moreover, for BCS-conference teams the impact of winning the NCAA tournament is statistically significant, whereas for non-BCS conference teams the parameter on LAGNCAACHAMP is not estimated because no non-BCS conference team won the NCAA tournament during the sample period. Urban schools affiliated with BCS-conferences actually enjoy greater per-game attendance than their non-urban, BCS-affiliated counterparts; urban schools in non-BCS affiliated conferences do not have different attendance than non-urban, non-BCS affiliated schools, all else equal. For BCS-affiliated schools, greater enrollment does not impact women's basketball attendance but for non-BCS affiliated schools, greater enrollment reduces women's basketball attendance. For BCS-affiliated schools the impact of FOOTBALL is approximately 28% and statistically significant whereas for non-BCS-affiliated schools the impact of FOOTBALL is not statistically significant, suggesting that for BCS-affiliated schools football and women's basketball seem to be complements but for non-BCS-affiliated schools the two sports seem independent.

However, using BCS affiliation is only one way to differentiate schools. A second approach is to differentiate schools by the number of enrolled students and see if the impact of football on women's basketball differs in this dimension. For instance, at very large schools football might be more likely to act as a substitute for women's basketball. This might be true because the two sports overlap at the beginning of the basketball season and a highly popular football team might reduce attendance to women's games early in the season which, in turn, would lower the team's average attendance. On the other hand, at very small schools football and basketball might be complementary if football provides for a greater sports culture on campus which spills over to women's basketball even while the football season is on-going. The schools in the sample were sorted into enrollment quintiles and the full specification in Model (3) in Table 4 was re-estimated for each; the results are reported in Table 5.

The results in Table 5 suggest that the impact of football on women's basketball attendance differs substantially across the different quintiles. Among the smallest schools in the sample, schools with football teams enjoy approximately 145% increase in per-game attendance than their non-football counterparts, all else equal. This result might be driven by a very concentrated sense of school spirit that inspires individuals affiliated with smaller schools to show greater support for their sports teams, all else equal. However, as school size increases, football and women's basketball appear to become independent sports; the impact of FOOTBALL in the second, third, and fourth quintiles is not statistically meaningful. Among the largest schools in the country, it appears that football and women's basketball are again complements to each other; football schools enjoy an approximately 70% greater per-game attendance to women's basketball relative to non-football schools, all else equal.

Football Program Quality

Finding a positive impact to certain schools of the presence of football on women's basketball attendance invites further inquiry into whether the quality of the football program influences women's basketball attendance. To test this, additional data were collected describing the football team's winning percentage, whether the football team plays in the FBS, and whether an FBS (FCS) team participated in a bowl game (postseason playoffs) during the season coinciding with the women's basketball season. The descriptive statistics of these additional variables are reported in final rows of Table 1.

The average winning percentage among those schools with football programs is .500 as there were no ties in college football during the sample period. Among the schools with a football team, approximately 51% of the observations correspond with a team playing in the FBS. Among those schools with football teams in the FBS, 25% of the observations correspond with teams playing in post-season bowls and 7% of the observations correspond with FCS schools with football teams that participated in post-season play.

Table 6 reports random-effects estimation results for various specifications using the restricted sample of schools with football programs. In each model, the variables describing the quality of the football team were all statistically insignificant except the post-season variables. FBS teams that played in a post-season bowl experienced a boost in per-game attendance to women's basketball of approximately 5.4%, whereas FCS schools that played in the post-season playoffs experienced a drop in per-game attendance of approximately 7.2%. The different results might arise because FBS bowl games are most often played at some distance from each school's campus and there is only one game involved; thus, the buzz created by participating in a bowl game might spill over to other winter sports, in the form of increased attendance, and yet the single bowl game does not detract from attendance during the period between the end of the football season and the bowl. On the other hand, FCS playoff games are played at the higher-seeded team's campus (except for the final game) and therefore an FCS playoff run might more directly draw fans away from attending women's basketball games. The overall lack of a systematic statistical relationship between women's basketball attendance and football program quality suggests that the results in the previous subsection are not being caused by a select few schools that happen to have football programs.

Discussion and Conclusions

The empirical exercise in this study focuses on attendance to women's collegiate basketball games and addresses two gaps in the literature: first, whether variables found to influence attendance to men's basketball and football games also influence attendance to women's games, and second, whether football acts as a complement or a substitute to women's basketball. The evidence suggests that recent and the not-too-distant quality of the women's basketball program does increase per-game attendance, consistent with the findings of a large number of empirical studies focusing on the attendance to professional sports and most closely matching the results in Depken (forthcoming) pertaining to men's collegiate basketball. It is shown that participating in the women's NIT can have a positive influence on a team's attendance in the following season; this is only the case if the team wins two games in the NIT. On the other hand, participating in the women's NCAA tournament boosts attendance in the next season as does winning games in the NCAA tournament. Finally, we find that winning either the NIT or the NCAA tournaments has a lasting positive influence.

Unique to this study is testing whether football is a complement, substitute, or independent of women's basketball. The evidence suggests that, on average, football is a complement to women's basketball but the relationship is not the same for all schools. Indeed, while BCS schools show complementarity between the two sports, and non-BCS schools show independence of the two sports, the relationship is actually more nuanced. Splitting schools into enrollment quintiles, the empirical evidence suggests that football and women's basketball are complements at the smallest and largest schools in the sample but are independent at mid-sized schools (with between 5,600 and 25,000 in total enrollment).

To the extent that the influence of football on women's basketball attendance is considered causal, any increase (decrease) in attendance to women's basketball games should be considered a benefit (cost) of having a football program on campus. Moreover, schools considering starting or discontinuing a Division I football program might consider the impact of football on the revealed interest in other sports on campus. Future research might focus on the extent to which football is a complement, substitute, or independent of other non-net-revenue sports.

References

Agthe, D. E., & Billings, R. B. (2000). The role of football profits in meeting Title IX gender equity regulations and policy. Journal of Sport Management, 14(1), 28-40.

Baade, R. A., Baumann, R. W., & Matheson, V. A. (2007). Down, set hike: The economics impact of college football games on local economies. College of Holy Cross, Department of Economics Faculty Research Series, Paper No. 07-02.

Baade, R. A., Baumann, R. W., & Matheson, V. A. (2008). Assessing the economic impact of college football games and local economies. Journal of Sports Economics, 9(6), 628-643.

Baade, R. A., & Sundberg, J. O. (1996). Fourth down and gold to go? Assessing the link between athletics and alumni giving. Social Science Quarterly, 77(4), 789-803.

Borland, M. V., Goff, B. L., & Pulsinelli, R. W. (1992). College athletics: Financial burden or boon? In G. W. Scully (Ed.), Advances in the economics of sport. Greenwich, CT: JAI Press.

Brown, R. W., & Jewell, R. T. (2006). The marginal revenue product of a women's college basketball player. Industrial Relations, 45(1), 96-101.

Coates, D., & Depken, C. A. (2009). The impact of college football games on local sales tax revenue: Evidence from four cities in Texas. Eastern Economic Journal, 35(4), 531-547.

Coates, D., & Depken, C. A. (forthcoming). Mega-events: Is the Texas-Baylor game to Waco what the Super Bowl is to Houston? Journal of Sports Economics.

Depken, C. A. (forthcoming). Is March Madness contagious? Post season play and attendance in NCAA Division I basketball. In L. Kahane & S. Shmanske (Eds.), Oxford Handbook of Sports Economics. Oxford, UK: Oxford University Press.

Fizel, J. L., & Bennett, R. W. (1989). The impact of college football telecasts on college football attendance. Social Science Quarterly, 70(4), 980-988.

Hadi, A. S. (1992). Identifying multiple outliers in multivariate data. Journal of the Royal Statistical Society, Series B (Methodological), 54(3), 761-771.

Hadi, A. 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(2), 393-396.

Hallmark J. R., & Armstrong, R. N. (1999). Gender equity in televised sports: A comparative analysis of men's and women's NCAA Division I basketball championship broadcasts, 19911995. Journal of Broadcasting & Electronic Media, 43(2), 222-235.

Humphreys, B. R. (2006). The relationship between big-time college football and state appropriations for higher education. International Journal of Sport Finance, 1(2), 119-128.

Kaempfer, W., & Pacey, P. (1986). Televising college football: The complementarity of attendance and viewing. Social Science Quarterly, 67(1), 176-185.

Kennedy, P. (1981). Estimation with correctly interpreted dummy variables in semilogarithmic equations [The interpretation of dummy variables in semilogarithmic equations]. American Economic Review, 71(4), 801.

Kerstetter, D. L., & Kovich, G. M. (1997). An involvement profile of Division I women's basketball spectators. Journal of Sport Management, 11(3), 234-249.

Madrigal, R., & James, J. (1999). Team quality and home advantage. Journal of Sport Behavior, 22(3), 381-398.

Magnold, W. D., Bean, L., & Adams, D. (2003). The impact of intercollegiate athletics on graduation rates among NCAA Division I universities. Journal of Higher Education, 74(5), 540562.

Matheson, V. A. (2007). Athletic graduation rates and Simpson's paradox. Economics of Education Review, 26(4), 516-520.

McCormick, R. E., & Tinsley, M. (1987). Athletics versus academics? Evidence from SAT scores. Journal of Political Economy, 95(5), 1103-1116.

McEvoy, C. (2005). The relationship between dramatic changes in team performance and undergraduate admissions applications. The SMART Journal, 2(1), 18-24.

Murphy, R. G., & Trandel, G. A. (1994). The relation between a university's football record and the size of its applicant pool. The Economics of Education Review, 13(3), 265-270.

NCAA. (2010a). 2010 NCAA Division I football records. Retrieved from http://www.ncaa.org

NCAA. (2010b). 2009-2010 NCAA women's basketball records. Retrieved from http://www.ncaa.org

Price, D. I., & Sen, K. C. (2003). The demand for game day attendance in college football: An analysis of the 1997 Division I-A season. Managerial and Decision Economics, 24(1), 35-46.

Rhoads, T. A., & Gerking, S. (2000). Educational contributions, academic quality, and athletic success. Contemporary Economic Policy, 18(2), 248-258.

Rishe, P. J. (1999). Gender gaps and the presence of profitability of college football. Social Science Quarterly, 80(4), 702-717.

Terza, J. V., Basu, A., & Rathouz, P. J. (2008). Two-stage residual inclusion estimation: Addressing endogeneity in health econometric modeling. Journal of Health Economics, 27(3), 531-543.

Toma, J. D., & Cross, M. E. (1998). Intercollegiate athletics and student college choice: Exploring the impact of championship seasons on undergraduate applications. Research in Higher Education, 39(6), 633-661.

Tucker, I. B. (1992). The impact of big-time athletics on graduation rates. Atlantic Economic Journal, 20(4), 65-72.

Tucker, I. B. (2004). A re-examination of the effect of big-time football and basketball success on graduation rates and alumni giving rates. Economics of Education Review, 23(6), 655-661.

Tucker, I. B. (2005). Big-time pigskin success: Is there an advertising effect? Journal of Sports Economics, 6(2), 222-229.

Tucker, I. B., & Amato, L. (1993). Does big-time success in football or basketball affect SAT scores? Economics of Education Review, 12(2), 177-181.

White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48(4), 817-838.

Endnotes

(1) The focus on Division I college football may be warranted as it is one of the most popular sports in the United States. In 2009, total attendance to 1,451 Division I NCAA intercollegiate football games was approximately 43 million (NCAA, 2010a), as compared to 73.06 million people who attended 1,212 regular season Major League Baseball games and 17.15 million people who attended 127 regular season National Football League games in the same year.

(2) Reasons for the unbalanced nature of the panel include several schools that were promoted (demoted) to (from) Division I status during the sample period, missing values for one or more variables in the IPEDS data, and missing data for one or more lagged variables. Additionally, the Citadel didn't field a women's basketball team during the academic years under consideration, although the school does participate in Division I (FCS) football.

(3) These are the names of the conferences at the time of writing.

(4) One hundred and twenty-three observations were identified as multivariate outliers according to the methodology developed by Hadi (1992, 1994). One hundred and ten of these observations were Big Six conference members.

(5) The Cook-Weisberg test statistics indicated rejection of the null hypothesis of homoscedasticity for each of the specifications reported in Table 2.

(6) As shown by Kennedy (1981), the impact of a dichotomous variable on a log-transformed dependent variable is appropriately calculated as [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where d is the parameter on FOOTBALL and V(d) is the sampling variance of [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

(7) Several teams are identified as a multivariate outlier for every year of the sample, including Connecticut, New Mexico, Notre Dame, Purdue, Tennessee, and Texas Tech. Several teams were identified as multivariate outliers for seven or eight of the years of the sample, including Baylor, Iowa State, Minnesota, Oklahoma, and Wisconsin. Three teams were identified as multivariate outliers for six of the years of the sample: Duke, Michigan State, and Penn State.

(8) For example, the New York Times reports in a 1911 story that Harvard spent approximately $31,000 (or approximately $725,000 in 2008 dollars) to field their football team in 1910 while Yale spent approximately $40,000 (or approximately $935,000 in 2008 dollars). In contrast, according to Equity in Athletics Disclosure Data, Harvard spent $2.35 million on its football program in 2007 while Yale spent $2.56 million in the same year. These two schools now play in the FCS (formerly Division I-AA) and spend considerably less on their football programs than schools playing in the FBS (formerly Division I-A).

(9) We measure the number of in-state football teams when each basketball season began play.

(10) The treatment-effect estimator is only one way to accommodate the endogeneity of having a football team on campus. An instrumental variables estimator was employed including school-specific random effects. Using this estimator the parameter on football was positive and statistically significant although the parameter estimates are slightly higher than in the treatment-effect model. Results are available from the authors upon request. The models were also estimated using two-stage residual inclusion (2SRI) methodology outlined by Terza, et al. (2008). This methodology yielded statistically significant and positive parameter estimates on FOOTBALL although the parameter estimates were slightly less than half the magnitude of the parameters reported in Table 4.

(11) The six BCS-affliated conferences are the Atlantic Coast, the Big East, the Big Ten, the Big Twelve, the Pac-10, and the Southeastern. While Notre Dame is considered an independent in football, it too is affiliated with the BCS system. However, Notre Dame plays basketball in the Big East conference and is therefore included in the BCS-affiliated subsample.

Craig A. Depken II [1], Courtney Williams [1], and Dennis P. Wilson [2]

[1] UNC-Charlotte

[2] Western Kentucky University

Craig A. Depken II is an associate professor in the Department of Economics. His research interests include industrial organization, applied microeconomics, and sport economics.

Courtney Williams has a B.A. in mathematics from the University of North Carolina at Charlotte.

Dennis P. Wilson is an associate professor in the Department of Economics. His research interests include applied microeconomic theory, public economics, industrial organization, and sport economics. Table 1: Descriptive Statistics of the Data Employed Variable Description PGATT Per-game Attendance WBWINPCT Women's basketball win percentage LAGWINBPCT Previous season's women's basketball win percentage LAGWNCAA Team participated in previous season's Women's NCAA Tournament (1=Yes) LAGWNIT Team participated in previous season's Women's NIT (1=Yes) LAGWNCAAWINS Number of wins in previous season's Women's NCAA Tournament LAGWNITWINS Number of wins in previous season's Women's NIT LAGWNCAACHAMP Team won previous season's Women's NCAA Tournament LAGWNITCHAMP Team won previous season's Women's NIT WNCAAWNITPREV7 Total number of Women's NIT and Women's NCAA tournaments in which team participated in last seven years URBAN School is considered urban (1=yes) PRIVATE School is private (1=yes) PCTFEMALE Percent of student body female ENROLL Total enrollment (Full Time Equivalents) FOOTBALL Football (1=Yes) OUTLIER Multivariate Outlier (p=.05) AGE Age of institution in years. TOTFBSCHOOLS Total number of Division I (FBS and FCS) football teams in the state in current year FBWINPCT Winning percentage of football program if it exists FBS Football team played in Football Bowl Subdivision FBSBOWL Football team played in FBS bowl game in current season FCSPLAYOFFS Football team played in FCS playoffs in current season Variable Mean Std. Dev. Min Max PGATT 1500 1950 109 15,796 WBWINPCT 0.5 0.194 0 1 LAGWINBPCT 0.499 0.195 0 1 LAGWNCAA 0.201 0.401 0 1 LAGWNIT 0.115 0.319 0 1 LAGWNCAAWINS 0.99 1.347 0 6 LAGWNITWINS 0.987 1.232 0 6 LAGWNCAACHAMP 0.003 0.056 0 1 LAGWNITCHAMP 0.003 0.056 0 1 WNCAAWNITPREV7 2.12 2.32 0 7 URBAN 0.272 0.445 0 1 PRIVATE 0.326 0.469 0 1 PCTFEMALE 54.48 7.06 13.03 78.69 ENROLL 163,69 112,72 907 67,082 FOOTBALL 0.732 0.442 0 1 OUTLIER 0.05 0.21 0 1 AGE 122.03 50.27 30 373 TOTFBSCHOOLS 7.2 4.19 0 16 FBWINPCT 0.504 0.221 0 1 FBS 0.375 0.484 0 1 FBSBOWL 0.187 0.39 0 1 FCSPLAYOFFS 0.051 0.221 0 1 Notes: Data gathered from the NCAA and the Integrated Postsecondary Education Data System (IPEDS). Sample contains 2,507 observations for 324 teams from 2000-2009. (a) Based on 506 observations of teams that played in the Women's NCAA Tournament during the sample period. (b) Based on 290 observations of teams that played in the Women's NIT during the sample period. (c) According to the algorithm developed by Hadi (1992, 1994). (d) The year the institution opened is determined by IPEDS and authors' calculations. Specifically, it is often necessary to trace the lineage of a particular institution to at least one parent institution; for example, the University of North Carolina at Charlotte was originally called Charlotte College. When appropriate we use the parent institution's opening year. Table 2: Influences on Attendance to Women's Collegiate Basketball (Robust Pooled OLS Results) (1) (2) (3) (4) WBWINPCT 0.741 *** 0.752 *** 0.706 *** 0.712 *** (0.069) (0.069) (0.068) (0.067) LAGWBWINPCT 0.437 *** 0.423 *** 0.435 *** 0.441 *** (0.078) (0.078) (0.076) (0.076) LAGWNCAA 0.060 0.075 * 0.078 ** 0.075 * (0.041) (0.041) (0.039) (0.039) LAGWNIT -0.130 *** -0.138 *** -0.097 ** -0.095 ** (0.048) (0.048) (0.046) (0.046) LAGWNITWINS 0.101 *** 0.101 *** 0.092 *** 0.091 *** (0.034) (0.034) (0.031) (0.031) LAGWNCAAWINS 0.128 *** 0.129 *** 0.037 ** 0.037 ** (0.019) (0.019) (0.017) (0.017) LAGWNITCHAMP -0.103 -0.093 -0.190 -0.176 (0.258) (0.252) (0.202) (0.198) LAGWNCAACHAMP 0.402 ** 0.359 ** 0.174 0.169 (0.165) (0.144) (0.117) (0.118) WNCAAWNITPREV7 0.112 *** 0.113 *** 0.104 *** 0.103 *** (0.008) (0.008) (0.007) (0.007) URBAN -0.063 ** -0.064 ** -0.051 ** (0.027) (0.025) (0.025) PRIVATE -0.175 *** -0.158 *** -0.156 *** (0.031) (0.030) (0.029) PCTFEMALE -0.003 * -0.003 * -0.002 (0.002) (0.002) (0.002) lnENROLLMENT -0.073 *** -0.068 *** -0.072 *** (0.021) (0.020) (0.020) OUTLIER 0.829 *** 0.824 *** (0.045) (0.045) FOOTBALL 0.083 *** (0.028) Pct. Change in Attendance due to Football (a) 8.62 *** Constant 5.942 *** 6.891 *** 6.863 *** 6.808 *** (0.073) (0.214) (0.208) (0.208) Observations 2507 2507 2507 2507 R-squared 0.72 0.72 0.75 0.75 Notes: Dependent variable is per-game attendance at Division I women's basketball games from the 2000-2001 through 2008-2009 basketball seasons. Conference-specific fixed effects included but not reported for brevity; results available upon request. Year dummy variables included in preliminary specification but were jointly insignificant; year dummy variables were not included in the specifications reported here. Explanatory variables are defined in Table 1. (a) Calculated as [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] , where [??] is the parameter on FOOTBALL and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is the sampling variance of [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (see Kennedy, 1981). Huber-White Standard errors reported in parentheses. *** indicates p < 0.01; ** indicates p < 0.05; * indicates p < 0.1. Table 3: Explaining the Presence of Football on Campus (1) (2) (3) FB(1=YES) FB(1=YES) FB(1=YES) AGE 0.007 *** 0.007 *** 0.007 *** (0.001) (0.001) (0.001) TOTFBSCHOOLS -0.035 *** -0.029 *** (0.007) (0.007) URBAN -0.490 *** (0.061) PCTFEMALE Constant -0.248 *** 0.065 0.146 (0.075) (0.096) (0.097) Observations 2507 2507 2507 (4) (5) (6) FB(1=YES) fb(1=yes) dPr(FB=1)/dX AGE 0.006 *** 0.006 *** 0.002 *** (0.001) (0.001) (0.000) TOTFBSCHOOLS -0.030 *** -0.030 *** -0.009 *** (0.007) (0.007) (0.002) URBAN -0.543 *** -0.543 *** -0.175 *** (0.063) (0.063) (0.021) PCTFEMALE -0.074 *** -0.074 *** -0.022 *** (0.006) (0.006) (0.002) Constant 4.430 *** 4.430 *** (0.344) (0.344) Observations 2507 2507 2507 Notes: Dependent variable is whether school has a Division I football program. Explanatory variables defined in Table 1. Institution age is measured in decades. *** indicates p<0.01; ** indicates p<0.05; * indicates p<0.1. Table 4: Influences on Attendance to Collegiate Women's Basketball (Pooled Sample Treatment-Effect Estimation Results) Sample Pooled Pooled Pooled WBWINPCT 0.712 *** 0.712 *** 0.712 *** (0.067) (0.067) (0.066) LAGWBWINPCT 0.441 *** 0.440 *** 0.437 *** (0.075) (0.075) (0.075) LAGWNCAA 0.075 * 0.076 ** 0.077 ** (0.039) (0.039) (0.039) LAGWNIT -0.096 ** -0.094 ** -0.094 ** (0.045) (0.045) (0.045) LAGWNITWINS 0.091 *** 0.090 *** 0.090 *** (0.030) (0.030) (0.030) LAGWNCAAWINS 0.037 ** 0.037 ** 0.037 ** (0.017) (0.017) (0.017) LAGWNITCHAMP -0.176 -0.176 -0.179 (0.196) (0.196) (0.196) LAGWNCAACHAMP 0.168 0.169 0.168 (0.117) (0.116) (0.116) WNCAAWNITP REV7 0.103 *** 0.103 *** 0.102 *** (0.007) (0.007) (0.007) URBAN -0.051 ** -0.051 ** -0.022 (0.025) (0.025) (0.028) PRIVATE -0.156 *** -0.155 *** -0.161 *** (0.029) (0.029) (0.029) PCTFEMALE -0.002 -0.002 0.000 (0.002) (0.002) (0.002) lnENROLLMENT -0.073 *** -0.069 *** -0.070 *** (0.020) (0.020) (0.020) OUTLIER 0.824 *** 0.826 *** 0.827 *** (0.044) (0.044) (0.044) FOOTBALL 0.065 0.177 ** 0.246 *** (0.079) (0.081) (0.079) Pct. Change in Attendance 6.38 19.02 ** 27.45 *** due to Football (a) Constant 6.827 *** 6.710 *** 6.516 *** (0.216) (0.219) (0.239) Variables in treatment AGE AGE, AGE, TOTFBSCHOOLS TOTFBSCHOOLS, TOTFBSCHOOLS, TOTFBSCHOOLS, URBAN,PCTFEM Observations 2507 2507 2507 Sample BCS Non-BCS Conferences Conferences WBWINPCT 1.007 *** 0.626 *** (0.133) (0.074) LAGWBWINPCT 0.185 0.457 *** (0.213) LAGWNCAA 0.141 * 0.044 (0.073) (0.048) LAGWNIT -0.195 ** -0.053 (0.080) (0.053) LAGWNITWINS 0.147 *** 0.067 (0.035) (0.045) LAGWNCAAWINS 0.047 ** 0.060 (0.019) (0.048) LAGWNITCHAMP -0.216 -0.349 (0.209) (0.443) LAGWNCAACHAMP 0.250 ** (0.115) WNCAAWNITP REV7 0.062 *** 0.111 *** (0.012) (0.009) URBAN 0.102 ** -0.085 (0.042) (0.054) PRIVATE -0.197 *** -0.107 *** (0.054) (0.035) PCTFEMALE 0.004 -0.001 (0.004) (0.003) lnENROLLMENT 0.079 -0.082 *** (0.049) (0.021) OUTLIER 0.792 *** 1.382 *** (0.044) (0.122) FOOTBALL 0.262 * 0.271 (0.137) (0.221) Pct. Change in Attendance 28.69 * 27.98 due to Football (a) Constant 5.242 *** 6.712 *** (0.495) (0.403) Variables in treatment AGE, AGE, equation TOTFBSCHOOLS TOTFBSCHOOLS, TOTFBSCHOOLS, TOTFBSCHOOLS, URBAN,PCTFEM URBAN,PCTFEM Observations 564 1943 Notes: The dependent variable is the natural log of per-game attendance for Division 1 women's basketball teams from 2000 -2009. Year fixed effects were jointly insignificant and were dropped from the final specifications. Each specification treats having a football team endogenous; specifications differ by the instruments used to explain whether a school has a football team. All specifications include conference fixed effects; year fixed effects were included in initial specifications but were dropped because they were jointly insignificant. (a) Calculated as [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where [??] is the parameter on FOOTBALL and V(d) is the sampling variance of [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (see Kennedy, 1981). Robust standard errors in parentheses. *** indicates p<0.01; ** indicates p<0.05; * indicates p<0.1. Table 5: Influences on Attendance to Collegiate Women's Basketball (Schools Divided into Quintiles by Enrollment) (1) (2) (3) Quintile 1 Quintile 2 Quintile 3 WBWINPCT 0.351 ** 0.777 *** 0.599 *** (0.143) (0.122) (0.152) LAGWBWINPCT 0.202 0.725 *** 0.425 *** (0.137) (0.143) (0.164) LAGWNCAA 0.046 -0.043 0.135 (0.074) (0.088) (0.114) LAGWNIT -0.012 -0.111 -0.132 (0.083) (0.107) (0.090) LAGWNITWINS -0.221 *** 0.133 0.167 *** (0.082) (0.083) (0.050) LAGWNCAAWINS 0.035 0.232 *** 0.046 (0.084) (0.066) (0.046) LAGWNITCHAMP (a) -1.078 s*** -0.342 (0.369) (0.255) LAGWNCAACHAMP (b) -0.081 (0.218) WNCAAWNITPREV7 0.113 *** 0.055 *** 0.115 *** (0.013) (0.017) (0.020) URBAN -0.064 -0.288 *** -0.078 (0.073) (0.072) (0.102) PRIVATE -0.082 0.021 -0.160 * (0.098) (0.115) (0.089) PCTFEMALE 0.007 * -0.008 -0.013 ** (0.004) (0.005) (0.005) lnENROLLMENT -0.001 -0.194 -0.761 *** (0.063) (0.153) (0.188) OUTLIER (c) 0.802 *** (0.125) FOOTBALL 0.906 *** -0.012 0.112 (0.131) (0.118) (0.138) Pct. Change in 145.41 *** -1.92 10.75 Attendance due to Football (d) Constant 5.493 *** 8.312 *** 14.004 *** (0.602) (1.397) (1.776) Observations (e) 456 496 495 Percentage of Quintile 60.86 60.56 69.72 with Football Min-Max Enrollment 907-5,657 5,658-10,569 10,573-16,481 of Quintile (4) (5) Quintile 4 Quintile 5 WBWINPCT 0.797 *** 0.711 *** (0.144) (0.114) LAGWBWINPCT 0.386 ** 0.305 * (0.165) (0.158) LAGWNCAA 0.102 0.084 (0.075) (0.062) LAGWNIT -0.091 -0.075 (0.082) (0.077) LAGWNITWINS 0.101 ** 0.045 (0.048) (0.041) LAGWNCAAWINS 0.010 0.029 (0.032) (0.020) LAGWNITCHAMP (a) -0.324 -0.088 (0.250) (0.165) LAGWNCAACHAMP (b) 0.190 0.252 *** (0.138) (0.090) WNCAAWNITPREV7 0.147 *** 0.048 *** (0.015) (0.013) URBAN -0.030 0.253 *** (0.064) (0.048) PRIVATE -0.256 *** -0.337 *** (0.076) (0.056) PCTFEMALE -0.001 -0.001 (0.004) (0.005) lnENROLLMENT 0.542 *** -0.337 *** (0.193) (0.103) OUTLIER (c) 0.751 *** 0.867 *** (0.098) (0.051) FOOTBALL -0.029 0.545 *** (0.119) (0.122) Pct. Change in -3.57 71.15 *** Attendance due to Football (d) Constant 0.516 8.909 *** (1.907) (1.066) Observations (e) 516 544 Percentage of Quintile 81.73 84.86 with Football Min-Max Enrollment 16,496-25,041 25,078-67,082 of Quintile Notes: (a) There have been no women's NIT champions from the first quintile of schools. (b) There have been no women's NCAA champions from the first and second quintiles of schools. (c) There are no multivariate outliers in the first and second quintiles of schools. (d) Calculated as [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where [??] is the parameter on FOOTBALL and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is the sampling variance of [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (see Kennedy, 1981). (e) Some observations lost in each quintile because of missing values. Robust standard errors in parentheses. *** indicates p<0.01; ** indicates p<0.05; * indicates p<0.1. Table 6: Impact of Football Team Quality on Women's Basketball Attendance (Random Effects Estimation Results) (1) (2) (3) WBWINPCT 0.671 *** 0.690 *** 0.674 *** (0.055) (0.059) (0.059) LAGWBWINPCT 0.335 *** 0.327 *** 0.320 *** (0.064) (0.069) (0.069) LAGWNCAA 0.092 *** 0.099 *** 0.099 *** (0.030) (0.033) (0.033) LAGWNIT -0.004 0.005 0.009 (0.036) (0.039) (0.039) WNCAAWNITPREV7 0.058 *** 0.059 *** 0.056 *** (0.008) (0.009) (0.009) LAGWNITWINS 0.035 * 0.033 0.037 (0.020) (0.023) (0.023) LAGWNCAAWINS 0.059 *** 0.065 *** 0.040 *** (0.012) (0.013) (0.012) LAGWNITCHAMP 0.052 0.122 0.030 (0.187) (0.199) (0.164) LAGWNCAACHAMP 0.042 0.077 0.078 (0.106) (0.099) (0.076) FBWINPCT -0.033 -0.082 -0.086 (0.049) (0.053) (0.052) FBS -0.020 -0.055 -0.054 (0.094) (0.102) (0.100) FBSBOWL 0.026 0.048 0.053 * (0.029) (0.031) (0.031) FCSPLAYOFFS -0.066 * -0.075 * -0.074 * (0.037) (0.039) (0.039) URBAN -0.095 -0.088 (0.072) (0.068) PRIVATE -0.151 * -0.155 * (0.084) (0.079) PCTFEMALE -0.001 -0.001 (0.003) (0.003) lnENROLLMENT 0.024 0.016 (0.054) (0.052) OUTLIER 0.483 *** (0.050) CONSTANT 6.082 *** 5.913 *** 6.015 *** (0.147) (0.547) (0.526) Observations 2068 1836 1836 Number teams 242 241 241 Notes: The dependent variable is per-game attendance at Division 1A women's basketball teams during the 2000-2001 through the 2008-2009 seasons. Sample includes only those schools with a football team. Sample size differs between Model (1) and Models (2) and (3) because of missing values for one or more variables. All specifications include conference-specific fixed effects, not reported here for brevity. Year fixed effects were jointly insignificant and were dropped from the final specifications. Robust standard errors in parentheses. *** indicates p<0.01; ** indicates p<0.05; * indicates p<0.1.