Lock-In and Team Effects: Recruiting and Success in College Football Athletics

How important is recruiting to a football program's success? While prior research has attempted to answer this question, we utilize an extensive panel set covering 13 years of games along with a two stage least squares approach to investigate the effects of recruiting on team success. We also split our sample to investigate whether recruiting displays heterogeneous effects across schools. Additionally, we find evidence that the benefits of recruiting are driven by team-specific effects, indicating that team success may be more heavily derived from the ability of teams to nurture their recruits than their ability to exploit athlete's raw abilities.


I Introduction
College football has become a massive business with hundreds of millions of dollars in revenue each year. Through television contracts, merchandising, and ticket sales, many colleges are seeing total revenues in the tens of millions of dollars per team. Run through the National Collegiate Athletic Association (NCAA), a nonprofit association, the competition for football teams to have success is arguably tougher, and more lucrative than ever. To this end, a vital part of every college football team's success is thought to derive from the ability of the coaching staff to attract ever higher levels of talented student-athletes. Higher rated recruits are expected to correlate with better on-field success. As of 2012, three teams spent more than two million dollars each on recruiting expenses alone, while fifty teams spent more than one million each. 3 The top 25 teams and their recruiting expenses are shown in Table 1.
Given the large sums of money flowing into football programs, understanding the importance of recruiting on football success can provide more insight into how football programs should be spending their budgets and time. This is especially true for schools that are less likely to compete for the highest valued bowl games, where getting a slightly better recruit may not necessarily provide enough boost to the team's on-field success to catapult them into the higher tier of bowl games. 4 However, for schools that are more likely to compete in the highest tier end-of-season games, a marginal boost in recruiting may be more valuable if it increases the likelihood of playing in such games.
To this end, we investigate three major points relating recruiting and team success. First, we improve upon the prior literature's work on measuring the effect of recruiting on team winning percentages by utilizing a more advanced econometric technique, increased sample size, and improved control variables. Secondly, we stratify the school football programs into three evenly sized groups based on their overall win percentages. Each of these groups are then analyzed separately to identify whether there are heterogeneous effects of recruiting on win percentages. Finally, we discuss the importance of including 3 See CSM0810 Report provided by ESPN 4 The sample period under discussion utilizes data during which the NCAA's post season consisted of single games at the end of the year, with the Bowl Championship Series (BCS) games providing the largest cash prizes for participation. team fixed effects in a panel data setting.
Our findings correspond well with prior literary findings. We find that recruiting has a positive effect on team success as measured by end-of-year win percentage. Our results would suggest that a single standard deviation increase in recruiting ratings would improve the team's win percentage by almost 32%. However, our stratification analysis indicates that this effect is largely limited only to those teams that already have a track record of on-field success. Those teams that average fewer wins per season are noted to receive less benefit for any marginal increase in recruiting rankings. These results suggest that teams tend to face lock-in effects with their peer groups.
However, upon controlling for team-specific effects, we find that the benefits of recruiting are largely eliminated. We believe this indicates two possibilities. Either recruit ratings as given a priori are relatively imprecise at determining a recruit's ex ante on-the-field presence, or that much of a recruit's impact on a team's success is through the available quality of training facilities, support staff, and coaches. Provided that recruit quality does have a positive correlation with team success, it is unlikely the former is the case. Thus, we argue that questions about football team success should concentrate less on how teams accumulate quality recruits, but instead to question what differentiates teams on their ability to exploit the quality of their facilities and coaching staff.
The remainder of the paper is organized as follows. Section II will explore prior research on football recruiting and team success. Sections III describes the dataset used while Section IV discusses the paper's methodology. Results are reported in Section V. Section VI concludes.

II Literature Review
Research into the effects of recruiting on football performance has been a topic of interest recently. With the rise in popularity of college sports and the explosion of publicly available data, it has become easier to investigate recruiting questions. Brown (1993) discusses how valuable football recruits can be to their football programs. Brown argues that the rent capture of a football recruit is equal to the difference between the player's marginal rev-enue product and the maximum payments allowable by NCAA scholarships. He utilizes a two-stage least squares estimation technique and identifies that football players often "... generate revenues well in excess of their effective wage..." indicating that a player with possible National Football League (NFL) capabilities can be worth $646,150 in annual revenues. Langelett (2003) investigates the more direct question of whether recruiting has an effect on a team's performance as measured by the team's endof-year standing in the BCS poll. He also makes use of a two-stage equation to answer this question. Langelett's final analysis indicates that recruiting does, in fact, have a positive effect on a team's performance. The paper also notes that teams' freshman classes are, on average, the most important determinant of a team's end-of-year success. 5 . However, Langelett's analysis does not account for head coaching changes, nor for schedule difficulty. Also, as will be discussed in Section IV, the usage of individual lagged periods may mask much of the effect of recruit quality. Dumond et al. (2008) analyzes the determinants of a recruit's decision making process and finds that a school's football success plays an important role in a recruit deciding on which school to attend. Dumond et al. and Langelett both demonstrate the importance of considering how recruiting may be a bi-directional process. On one hand, better recruits may increase a team's performance, however, a team's performance also effects its ability to draw better recruits. This directionality problem lies at the heart of why using a two-stage least squares regression is vital to account for the multiple pathways of causality.
An important consideration when researching the effects of recruiting is also brought up by Bergmen and Logan (2014). Bergmen and Logan note that each school may have different abilities to not only draw in recruits, but to also train, strengthen, and improve said recruits as well. Therefore, they analyze the importance of school fixed effects on recruiting. Their research indicates that after controlling for school fixed effects, the magnitude of recruiting on team performance falls by as much as 25%. They note that even after including fixed effects, increased recruit quality still brings better team performance on average. Unfortunately, their OLS may miss important time specific effects that we consider in our research here.
Our paper builds upon the prior literature by improving the analytical 5 See also Caro (2012) and data components of the research. We utilize a two-stage least squares regression technique similar to Langelett's methodology, but we include measures for head coaching changes, conference designations, opposition quality, and fixed school effects as per Bergmen and Logan. In addition we utilize a panel dataset to facilitate the consideration of time instead of a more traditional cross-sectional approach, as well as including data for all Football Bowl Subdivision (FBS) teams over a twelve year span. 6 A new addition to the literature is also provided by our stratification analysis in an attempt to consider differential effects on recruiting for different tiers of schools.

III Data
The recruiting ratings was acquired from Rivals, a company that specializes in college recruiting information. Rivals provided the recruitment ratings for most of the FBS teams in the NCAA. They ranked each recruit on a star based system from 5 to 1 with 5 stars representing the best recruits available. The class rankings used in our analysis are generated from the average ratings for each year, weighted by the number of recruits within each star ranking. A summary of the data used here can be found in Table 2.
Yearly schedules were acquired from publicly available information provided by the NCAA. For the years 2001 through 2011, the NCAA data was utilized. To supplement the data additional years (2012-2013) were provided by ESPN. Each year's schedule was used to calculate every team's win percentage as well as the average win percentage of all of their opponents. To account for team specific recruiting capabilities, team fixed effect dummies were built based on the NCAA's database. Each team was given its own dummy variable for a total of 113 FBS participants. 7 Data on head coaching changes and conference yearly conference alignment were acquired from two internet based sources. Head coaching changes were collected from College Football Poll. 8 Head coaching changes are an 6 FBS is the designation provided to all former Division-1A schools. 7 As of 2013 there were 124 FBS teams, however, some were newly promoted and thus did not have data for the entire period while others were missing some of the control variable data. Therefore, any teams without the full 13 years of data were dropped after calculating opponent's win percentage.
important event to consider given that the choice of a program to get a new head coach may be an indicator of a program's expectations not meeting its on-field performance. New head coaches may also bring different philosophies on recruiting and play calling that might alter the program's recruiting efforts. A head coaching change however, might also indicate that a school's coach was extremely successful and was drawn away to coach for a larger team or even moving into the NFL. Additionally, each college football program is associated with one of 13 conferences. 9 Generally, each conference plays the majority of its schedule against in-conference opponents, with only two or three games played against non-conference opponents. In most years, conference membership was relatively stable. However, in 2012 a major realignment occurred in which the Big East Conference and Western Athletic Conference were each disbanded. Their membership was then distributed amongst the 11 other conferences including the newly created American Athletic Conference. A few of the older conferences also witnessed membership changes, for example, Nebraska, Colorado, Texas A&M, and Missouri joined the Pacific-12 Conference, Big Ten, and Southeastern Conference, while West Virginia and Texas Christian joined the Big 12 Conference. Switching conferences may be expected to effect recruiting and team success, as such, we include dummy variables for each team-year's conference alignment. 10

IV Methodology
There are two major considerations to take into account when analyzing the effect of recruiting on football success. As Langelett described, a bidirectional problem exists in which recruiting and team success influence each other. As such, we utilize a two-stage least squares regression technique 9 A few are listed as unaffiliated schools. Such schools were aggregated into an 'unaffiliated' category.
10 Moving from one conference to another will tend to alter the viewing audience leading to a change in recruiting patterns. For example, Nebraska's move to the Big Ten Conference led to its televised games being predominantly shown in Mid-Western homes, whereas when it was a Big 12 team, its television appearances tended to be shown in the Great Plains and South Central areas. Recruits in each area may be more drawn toward teams that they can watch with more regularity. defined as follows: (1) Where Recruiting i,t is a variable indicating the recruiting class rating of team i at time t. 11 WinPer i,t is defined as team i 's year t win percentage, while WinPer i,t−1−>t−2 contains each lagged win percentage of team i over the prior two years. Two distinct lags are used to reflect the the likelihood that upon entering a player's junior year in high school, players with college football aspirations may be paying extra attention to program success. We expect a positive correlation between a team's success and recruit ratings.HC i,t−1 is a dummy variable indicating 0 if a team has not changed its head coach in year t while T i is a set of dummy variables for each team. We have no expectations on the effect of a head coaching change on recruiting or team success. On the one hand, an unsuccessful team my fire their head coach and lure a new (presumably better) head coach to the program, resulting in a positive correlation. However, it may also be the case that some teams lose their extremely successful head coaches to bigger programs or to the NFL, which might lead to subsequently lower success by teams. The T variable accounts for school-specific fixed effects as mentioned by Bergmen and Logan. To account for conference specific effects, a dummy variable indicating each team's conference alignment is included as C i . Finally OppPer i,t measures team i 's oppositional win percentage in year t. This controls for difficulty of schedule by accounting for the fact that certain teams may play a schedule with tougher opponents than another team. To calculate this, we remove the the win percentages of opposing Football Championship Subdivision (FCS) teams. FCS teams play in a different division of college football with only a handful playing FBS teams each year. Without full schedules for those teams, it was not possible to calculate their win percentages, so they were dropped from generating the OppPer i,t variable.
The variable of interest is Recruiting i,t−1−>t−5 which indicates the five year trailing moving average of recruit ratings. 12 The moving average is used as a moving average will better capture the total ability of a team at any specific point in time rather than the use of lags which are more appropriately used 11 The recruiting class rating is derived from the average rating of all recruits to each school. In effect, it is an average recruiting rating.
12 Five years are used as there are generally five years of classes on a football team; red-shirt freshman, freshman, sophomore, junior, and senior.
to answer questions about specific class effects (i.e. are freshman or seniors more important to a team's success). Here, we want to isolate the specific effects of increasing the total quality of the recruits a team has available. The trailing moving average provides for a single value that indicates the quality of recruits at a team. The outcome variable of interest, Recruiting i,t−1−>t−5 is expected to show a positive correlation, indicating that if a team can improve its recruit rating, then a team will likely experience greater on-field success.
In order for the two stage least squares estimation technique to be valid, it must meet the exclusion restriction. Our argument centers on how recruits make decisions on their school of choice. As Dumond et. al. note, two important determinants of a recruit's decision to attend a school is the team's conference alignment and its success. This is modeled in equation (1). However, we note that recruits are unlikely to be paying attention to a team's opposition. Instead, an athlete is likely to make a decision on attending any specific school based on its success, which may be impacted by the quality of opposition. Thus, we include difficulty of schedule in equation (2), but not in equation (1).
A second method will utilize the same regression technique, but instead will split the sample into two evenly sized groups based on sample period win percentage. Each team has its average win percentage calculated for the entire sample, then they are split into halves. Each tier runs a two stage least squares regression using (1) and (2). This technique allows for exploration regarding the value of recruiting to each group. The reason why this might be a valuable tool of analysis is that there have been some studies indicating that NCAA rules and regulations create lock-in effects in which certain tiers of teams can utilize better recruits more effectively than others. 13 For this model, we expect to see different magnitudes on equation (2)'s Recruiting variable for each tier. If lock-in effects exist, then it is expected to find statistically insignificant effects for the less successful teams.

V Results
We first split the sample to test for lock-in effects. Table 3 presents these findings. The primary variable of interest is the coefficient on Recruiting. Both halves of the sample report positive correlations between recruiting quality and team success. As hypothesized, the results for the lower half of FBS teams finds this relation to be statistically insignificant. This indicates that for those teams for whom consistent on field success does not occur are, on average, unlikely to see any benefit from increasing their recruit quality. However, more successful teams do see a statistically significant beneficial effect from pulling in more highly rated recruit classes. This lends to credence to the prior research discussing the possibility of lock-in effects amongst peer groups in football teams. Table 4 provides the main findings for the research presented here. The cross-sectional results of only using equation (2) can be found in model (1). All three variables are found to be statistically significant. Head coaching is found to be negatively correlated with team success, indicating that (at least in the short term), bringing in a new head coach will likely lead to an 8% lower end of year win percentage. Opposition win percentage is inversely associated with own-team win percentage. This finding is expected as it provides evidence that a team facing a more difficult schedule will likely win fewer games over the course of a season. Model (1) would associate a standard deviation increase in difficulty of schedule would likely lead to a 10% lower win percentage (approximately one extra lost game for that season).
Models (2) and (3) utilize the more nuanced two stage least squares methodology. Without accounting for team specific effects, we continue to find evidence of the importance of recruiting, with the magnitude doubling in size after accounting for bi-directional causality. Both head coaching changes and opposition win percentage continue to be significant in both size and magnitude. This corresponds well to both Langelett and Bergmen's findings. However, upon controlling for team-specific fixed effects, the effect of recruiting turns insignificant. 14 In essence, team specific unobservables account for much of each team's success. Such a finding, while surprising, provides for a new avenue of consideration in future research. These results would suggest that recruiting raw ability is less important than the utilization of the team's facilities. In other words team success may be more akin to the discussion of the effects of nature versus nurture. In our case, these findings represent a push toward the importance of nurturing in the generation of team performance from recruits.
One may argue that our results are more heavily driven by Rival's recruit rankings providing less predictive power than expected. However, without team effects, the findings indicate a relatively strong correlation between recruit quality and team success. Additionally, if individual success can be identified as being drafted in the first round of the NFL, then Rival's ratings are still well capable of identifying quality football players. For example, in the 2014 NFL draft, 32 players were selected in the first round, of which 14 were juniors. 15 As a percentage of their high school year class (2011), the junior represented 11% of the 2011 class's five star recruits, 1.5% of their four star players, and .27% of the three star recruits. Rival's recruit ratings would appear to correlate with the possibility of first round draft status, indicating that they are at least somewhat predictive of player's on the field potential.

VI Conclusion
In this paper we expand upon the prior literature investigating the effect of recruiting on college football team success. By introducing a split sample analysis, our research indicates the possibility of lock-in effects amongst college teams. Less successful teams, on average, tend to derive less benefit from quality recruits compared to more successful teams. This may be due to a lower capacity to utilize recruit abilities.
We also find that much of the prior findings may be driven by misspecification in econometric modeling techniques. Specifically, the inclusion of both a two stage least squares panel regression technique combined with team specific control variables yields results that run counter to prior findings. These results are demonstrated to be largely the result of team specific control variables soaking up the effect of recruit quality.
However, these results also open up new avenues of possible research. The findings presented here indicate that future research into football team success may derive more benefit from focusing on team specific characteristics as opposed to recruit quality. Coaching quality and team facilities are both examples of less studied factors that may influence team success and the ability of recruits to perform on the field. 16 It should be noted that there is no standard for defining how recruiting expenses should be calculated. Therefore, recruiting expenses for some universities may not be wholly accurate. Most likely these are underestimating recruitment expenses since some costs may be classified under different groupings such as coaching staff salaries, staff and administration salaries/benefits, travel, and fund raising/marketing.  0.28 0.19 *: p < 0.10 **: p < 0.05 ***: p < 0.01 Standard deviations in parentheses.