The Effect of the College Football Preseason Coaches Poll on BCS Game Participants

Tags: Coaches Poll, BCS games, preseason Coaches Poll, ranked teams, BCS game, BCS standings, probit model, BCS formula, conference, BCS ranking, national championship, Preseason Coaches, preseason rankings, Harris Poll, regular season, unfair advantage, Bowl Championship Series, winning percentage, college football, conferences, bowl games, BCS Game Independent variables, BCS bowl games, independent variables, team, ranked team, Strength of schedule, National Football League, Championship Series, Major League Baseball, dependent variable, preseason poll, BCS conference, BCS conferences, Iteration, College Football Poll, game participation, Pre Hamill, Major Hamill, OLS, log likelihood
Content: The Effect of the College Football Preseason Coaches Poll on BCS Game Participants Dillon Hamill Senior Thesis in Economics Professor Preston 26 April 2012
1. Introduction
Hamill 2
Sports have become of one of the greatest forms of entertainment we know. Whether it be the National Football League or Major League Baseball, the two most popular sports leagues in America, or college volleyball or lacrosse, sports everywhere captivate a certain audience. One of the greatest contributors to this captivation is the constant theme throughout of a champion. In every sport, at every level, one team is determined to be the best. This conclusion can be made because at the end of every regular season a playoff is enacted, with various teams knocking each other out until one is left standing. This occurrence, while varying in number of participating teams depending on the sport, occurs in every sport at every level, whether it be High School, college, or professional, except one: Division 1-A college football. Instead of a playoff, college football has the Bowl Championship Series. For this reason, the system is very controversial. But additional controversy stems from what compromises the bowl championship series, and one of these components is the Coaches Poll. My thesis will examine the effect the preseason Coaches Poll has on who plays in BCS games, specifically if teams ranked higher in the poll have greater chances of making it into these games than lower ranked teams of equal or better quality. The BCS standings are a three part formula, consisting of the Coaches Poll, the Harris Poll, and a computer ranking system. Out of these three components, the Coaches poll is the one I want to examine for the following reasons. It has been a part of the BCS formula since the inception of the BCS in 1998, allowing for as many observations as possible. There are also many controversies associated with the poll, notably the claims that coaches are biased towards
Hamill 3 their own teams/conferences, coaches don't even fill out ballots, coaches are uninformed about the teams they are voting for, and the fact only 59 out of the 120 coaches vote. Most importantly though, I want to study the effect of the Coaches Poll on the BCS because it is the only component to be released before the season starts, when teams have not played a single game; the Harris poll and computer rankings aren't released until the first BCS standings are released after the 6th week of the season. That is what intrigues me: do rankings that come out when no team has yet performed unfairly give higher ranked teams an advantage? 2. Background Information a. The Bowl Championship Series The Bowl Championship Series is a series of contracts that ties conferences to bowl games, which serve as postseason play for college football. There are eleven conferences in college football with a total of 120 teams, and with the number of bowl games growing every year (the 2010 season has a record 35), more and more teams are getting to participate in them. Examples of these bowl tie-ins include the Alamo Bowl, which pits the #2 ranked team from the Pacific-10 Conference against the #3 ranked team from the Big-12 Conference, and the Capitol One Bowl, which matches teams from the Big Ten Conference and Southeastern Conference (SEC). While these bowls are entertaining, the real prestige in modern-day college football lies in the five self-titled "BCS games". Four of these games are just like any of the other 30 bowls, as they also match teams from specific conferences against each other, depending on the game,
Hamill 4 while the fifth game is the national championship. The difference here is that these games pit the best of the best teams in college football against each other, usually in the form of a team that won its Conference Championship; the Rose Bowl invites the Pac-10 Champion to play the BigTen Champion. The prestige of these games lies in the aforementioned problem of college football: there is no playoff, and as a result these games, not including the national championship, are the highest goal many of the teams can strive to play in. The media coverage of these games is more substantial than the other bowl games, the payouts for playing in these games is significantly larger, and there are even television shows devoted to examining who will play in BCS games (BCS Countdown is the main one, airing on ESPN). Clearly, BCS games are viewed as better than the rest, and as such will be the focus of my study. The BCS began in 1998 as the means of determining a national champion in men's division 1-A college football, replacing the Bowl Alliance agreement that ran from 1995-1997. The initial formula, from 1998 to 2004, consisted of 1/3 poll average (combined Coaches Poll and Associated Press Poll), 1/3 computer average (3 total computers), and 1/3 strength of schedule. However, after complaints about the leniency on computers/mathematical formulas, a new formula was developed that put more emphasis on human opinion. Starting in 2005, the BCS formula was 1/3 AP Poll, 1/3 Coaches Poll, and 1/3 computer rankings (now 6 computers). The final change came the following year, when the Associated Press decided it no longer wanted to be a part of the BCS. The Associated Press Poll was replaced with the Harris Poll, which is voted on by 114 people, all of whom are former players, coaches, administrators, and members of the media as well. This has been the BCS formula to date, and the Coaches Poll is the only measure which has remained exactly the same since the beginning of the Bowl
Hamill 5 Championship Series, reinforcing its significance to the BCS and henceforth my study. As previously mentioned, the BCS currently contains five games: the national championship and four other games (Rose, Orange, Sugar, and Fiesta Bowls). However, from 1998-2006, the BCS consisted of only four games, and as such the four BCS bowl games would rotate hosting the national championship. This changed after 2006, when another game was added to create additional revenue for all parties involved (conferences, coaches, athletic directors, etc). There have been 57 BCS games to date, meaning a total of 114 teams have participated. The last significant part of the Bowl Championship Series lies in the selection process of teams for its games. Since this information ties directly in with my data collection, I will include its explanation in that section to come. b. The Coaches Poll The Coaches Poll is a weekly ranking of NCAA Division I-A football teams based on the votes of 59 members of the AFCA Division I-A Board of Coaches. The poll is sponsored by the USA Today and administered by the AFCA. Each season opens with a preseason poll that is updated every week during the regular season. The results are released as the USA Today Coaches Poll ranking of the top twenty-five teams, showing the number of points each team received, where points are allocated as 25 for a first-place rank, 24 for a second-place rank, etc., and summed across the ballots of all participating coaches. While only the top 25 teams are officially ranked, any team that received a top 25 vote from a coach is listed at the bottom of the poll, and then ranked accordingly to how many votes that team received; in essence, upwards of
Hamill 6 40-45 teams are ranked in any given poll. The Coaches Poll results are listed and promoted on their own, but they are also combined with ranking systems (described previously) to produce an official BCS ranking. The final regular season poll, conducted after the conference championship games around the first week of December, has added significance because it contributes to the BCS formula for selecting teams to play in the national championship and the eligibility of teams for invitations to other BCS bowl games. Beginning in 2005, the AFCA began making public each coaches ballot for this poll, though ballots are not publicly available for other polls during the season. Public disclosure of these ballots was a requirement for keeping the Coaches Poll a part of the BCS rankings formula in the wake of controversy following the 2004 final poll and BCS rankings, in which there was evidence of voter bias from one of the coaches. Starting in 2009, weekly coaches ballots became public following further complaints of voting bias. 3. literature review In "Voter Bias in the Associated Press College Football Poll", Coleman et al. (2010) investigate multiple weekly biases of the 65 voters in the Coaches Poll in the year 2007. The authors collected ballots from each voter for the final nine weeks of the season to get a greater sample size of games played among teams, trying to rid themselves of potential effects of preseason rankings (which I am in fact examining). In their regressions, they used team rank as assigned by voters as their Dependent Variables, and differing bias factors as their independent
Hamill 7 variables, including geographical and TV factors. The authors found evidence of bias on numerous occasions: towards teams from a voter's state, in conferences represented by a voter's state, in BCS conferences (which will be explained in coming sections), and towards teams that played in televised games. This research is impactful on my study because coaches could potentially be influenced by these same biases. If that is the case, and these biases are at play in the preseason, then the preseason Coaches Poll could have an even more significant effect on who plays in BCS games; not only could teams be unfairly penalized by where they are ranked to start the season, but these rankings could be influenced by numerous other biases out of their control. Matthew Kotchen and Matthew Potoski (2011) looked at potential biases of voters in the Coaches Poll. The authors were driven by the impact of the Coaches Poll on the Bowl Championship Series. They examined weekly biases in the poll, not in the preseason, and focused the majority of their study on the financial impact of the BCS. The authors collected data from the ballots of the final regular season poll in the years 2005 to 2010. They ran two regressions, both of which used coach ranking as their dependent variable, as the preceding paper did. The difference in their independent variables was that in one of their regressions they used computer rankings as the main independent variable, in which bias is not possible, and in the other they did not include them; instead, they used a fixed-effect regression that accounted for heterogeneity of team quality each season. The authors found evidence that coaches distort their rankings in order to reflect own team's financial interests and reputations: on average, coaches ranked their own team two spots move favorably, ranked teams from their own conference one
Hamill 8 spot more favorably,1 and ranked teams they defeated more favorably, which would boost their own team's reputation. This paper thus proves that biases are indeed found in the Coaches Poll at the end of the season, and it is likely they are found in the preseason poll as well. If that is the case, it is further proof that teams can be unfairly penalized in a preseason poll, affecting their ability to play in their sport's most prestigious games. In Death to the BCS: The Definitive Case Against the Bowl Championship Series, Wetzel, Peter and Passan (2010) blast the BCS, revealing its every flaw in great detail. One chapter, though, is particularly relevant to my study in its critique of the Coaches Poll. As in the previous two papers, Wetzel et al. expose numerous voter biases in the Coaches Poll. They reveal that some of the coaches in the Coaches Poll ranked teams a certain way when they admittedly had never seen them play. Two examples include Missouri coach Gary Pinkel ranking an undefeated Utah team 15th in the final 2008 standings, and LSU coach Les Miles ranking an unbeaten Cincinnati team 8th at the end of 2009. Another coach, South Carolina's Steve Spurrier, even stated his belief that most coaches do not know that much about other teams. Additionally, coaches don't always cast their ballots, a duty that is often left to administrative assistants or media relations directors, people who know even less about college football. All of the potential controversies listed in the introduction have been proven by these three sources, giving my paper further motivation to discover if the preseason Coaches Poll significantly impacts who plays in BCS games. 4. Hypotheses 1 For every BCS game a team plays in, that team's conference gets a payout. While the team participating in the game gets the majority of this revenue, every other team in that conference gets a share as well.
Hamill 9 I will test the following hypotheses: First, being ranked in the preseason Coaches Poll has a significant impact on where a team is ranked in the final BCS Standings. Second, there is a general positive correlation between being ranked in the preseason Coaches Poll and participating in a BCS game. And third, there is a step ladder effect of significance when breaking down the preseason rankings into groups of five, with a 1-5 ranking in the preseason Coaches Poll having the greatest effect on BCS game participation, then 6-10, then 11-15, etc. 5. Data The data I have collected is based off the final aspect of the BCS: the selection process. Out of the eleven conferences in Division 1-A college football, six of them (ACC, Big East, Big Ten, Big-12, SEC and Pac-102) are considered "BCS" conferences. This means that the champion of each of these conferences receives an automatic berth to a BCS game, regardless of record or ranking in the BCS. Normally, since their final BCS ranking is irrelevant in determining whether they play in a BCS game, I would have excluded those teams from my observations. However, only 5 of the 78 automatic-qualifying teams that have played in the BCS would not have qualified as an at-large, and thus they are still relevant to my analysis. In order to qualify for an at-large berth in a BCS game, teams must be ranked in the top fourteen of the final BCS standings and have at least nine wins. There are other ways to earn an automatic berth into the BCS, but unlike champions from BCS conferences, these invitations are dependent on BCS ranking and as such are pertinent to 2 Renamed Pac-12 in 2011 after addition of two teams to conference. Since this paper only counts observation through 2010 season, Pac-10 will be used in the remainder of the writing.
Hamill 10 my study. If Notre Dame finishes in the top eight in the final BCS standings, it receives an automatic berth to the BCS,3 as does the highest ranked champion from a non-BCS conference (MWC,WAC, Sun Belt, MAC and C-USA) if that team finishes in the top twelve of the final BCS standings. No more than one such team from one of these conferences can earn an automatic berth to the BCS; however, a second such champion can still qualify as an-at large team, the other type of entrant to the BCS I will be examining. This distinction belongs to teams from either the five non-BCS conferences, who did not receive an automatic berth, or nonchampions from a BCS conference. I have collected data from 70 teams across 13 years, 1998-2010, which comes out to 910 observations. However, one of these teams, South Florida, joined Division 1-A football in 2001 and as such there are three less observations. These 70 teams are relevant to my study because they are found in at least one of the following categories: - played in a BCS game - were ranked in the preseason Coaches Poll, 1-25, in at least one year - were a part of the final BCS standings in at least one year - if a non-BCS conference team, won their respective conference with at least nine wins in at least one year The specific data I have collected represents both dependent and independent variables: Preseason Coaches Poll Ranking (PreCoachRank) represents the ranking of each team in the preseason Coaches Poll. All teams ranked from 1-35 are numbered as such, and the rest are given a value of 75, which represents a median between the lowest ranked teams in the poll and all 3 Notre Dame is classified as an independent in college football, meaning it is not affiliated with any conferences. While three other teams share this distinction with Notre Dame, ND has a prestigious football history and rapid fan base and as such the founders of the BCS wanted to ensure their presence in BCS games if they were qualified.
Hamill 11 teams in Division 1-A football. Final BCS Ranking (Fin BCS Rank) represents the ranking of each team in the final BCS standings released each year. All teams ranked in the BCS standings from 1-25 are numbered as such, and the rest are given a value of 70, which is again a median. BCS Game Participation (BCSGame) is a binary variable, with a value of 1 if the team participated in a BCS game and a value of 0 if the team did not. Final Games Played (FinGames) represents the amount of games played by a team during the regular season. Final Winning Percentage (Win %) represents the winning percentage of each team at the end of the regular season, and is calculated as follows: Wins/FinGames. Strength of Schedule (SOS) represents how difficult each team's schedule is each year. SOS is calculated as follows: (2(Opponents Record) x Opponents' Opponents Record)/3 This means the collective winning percentage of a team's opponents is multiplied by 2, and that value is then multiplied by the collective winning percentage of a team's opponent's opponents winning percentage. That number is then divided by three to determine a team's overall strength of schedule. Teams are then ranked 1-120, with 1 representing the hardest schedule in college football and 120 representing the easiest. BCS and Non-BCS Conferences (Major and Minor) are both binary variables that represent whether each team is a member of a BCS or Non-BCS Conference, respectively. Preseason Coaches Poll Ranking 1-5, 6-10, 11-15, 16-20, 21-25 (Pre 15, Pre 6-10, etc) are all binary variables which represent where each team is slotted in the preseason Coaches Poll in each year. Others Receiving Votes 26-30 (ORV 26-30) is a binary variable which represents the first five teams left out of the preseason Coaches Poll each year. All Other Teams (AOT) is a binary variable which represents all teams not slotted 1-30 in the preseason Coaches Poll each year.
The Summary statistics of all variables are presented in Table 1:
Hamill 12
Summary Statistics of All Variables in Regression Analysis
Observations Mean
Std. Dev.
Min
PreCoachRank 907 Fin BCS Rank 907
47.32
29.60
1
52.63
27.02
1
BCS Game 907
FinGames
907
.1257
.3317
0
11.71
.6207
10
Win %
907
SOS
907
.5984 42.09
.2110 27.54
.0000 1
Major
907
.8225
.3823
0
Minor
907
Pre 1-5
907
.1775
.3823
0
.0717
.2581
0
Pre 6-10
907
Pre 11-15
907
.0717
.2581
0
.0717
.2581
0
Pre 16-20
907
Pre 21-25
907
.0717
.2581
0
.0728
.2599
0
ORV 26-30 907
.0650
.2467
0
AOT
907
.5733
.4949
0
Max 75 70 1 14 1.000 117 1 1 1 1 1 1 1 1 1
There was a higher mean and standard deviation on Pre 21-25 than on the other preseason variables due to two teams being tied for 25th one year in the Coaches Poll. There is also a lower
Hamill 13 mean and standard deviation on ORV 26-30 than the other preseason variables due lack of data in this category for two years.
6. Methodology
I estimate the effect of preseason ranking on final outcome using the following equation: Final Outcome = 0 + 1(Preseason) + 2(Win%) + 3(FinGames) + 4(SOS) + 5(Major) + Final outcome is measured in two ways, final BCS ranking and a dummy variable equal to one if the team plays in a BCS game. Preseason rank is included as a continuous variable in one specification and as a set of dummy variables (ranked 1-5, ranked 6-10, etc.) in the other. Strength of schedule is included as a continuous variable, wins is winning percentage and games is total games played. In specifications with the dummy dependent variable, I estimate both the linear probability model (OLS) and the probit model. Results from these regressions are presented in Table 2:
Preseason Ranking (1)
Coefficients of Independent Variables in Regression Analysis
Final BCS Ranking (1) .1234*** (.0235)
Final BCS Ranking (2)
Played In BCS game OLS (3) -.0014*** (.0004)
Played in BCS game OLS (4)
Played in BCS game Probit (5) -.0011** (.0005)
Played in BCS game Probit (6)
Preseason Ranked 1-5 (2)
-17.38*** (2.535)
.2839*** (.0391)
.1430** (.0800)
Preseason Ranked 6-10 (3)
-14.96*** (2.450)
.1549*** (.0377)
.0567 (.0564)
Preseason
-4.231*
.1364***
.2065***
Ranked 11-15
(2.395)
(.0369)
(.0935)
(4)
Preseason
-.0907
.0209
.0413
Ranked 16-20
(2.317)
(.0357)
(.0700)
(5)
Preseason
-2.375
-.0347
.0216
Ranked 21-25
(2.289)
(.0353)
(.0680)
(6)
Others
.5865
-.0353
-.0488
Receiving
(2.404)
(.0370)
(.0363)
Votes 26-30 (7)
***significant at the 99% level **significant at the 95% level *significant at the 90% level
The first regressions I ran were standard OLS regressions, which tested the correlation between preseason Coaches Poll ranking, both in general and split 1-5, 6-10, etc., and final BCS ranking. The purpose of these regressions were to see how much of an effect the preseason Coaches Poll had on where teams end up ranked in the BCS and thus whether they are eligible to play in a BCS game. The results of the first regression showed a positive relationship between preseason Coaches Poll ranking and final BCS ranking, as the coefficient on PreCoachRank (column 1, row 1) was .1234, significant at the 99% level. Thus, generally being ranked in the preseason Coaches Poll had a significant effect on where a team ends up in the BCS standings, controlling for winning percentage, strength of schedule, conference affiliation and final games played; a decrease in preseason ranking by one caused the final BCS ranking of a team to decrease by .12 spots. The results of the second regression showed that there was a divide within the Coaches Poll in terms of the effect on Final BCS Ranking. The coefficients on Preseason 1-5 (column 2,
Hamill 15 row 2) and 6-10 (column 2, row 3), - 17.38 and - 14.96 respectively, were both significant at the 99% level, and the coefficient on Preseason 11-15 was significant at the 90% level. Thus, being ranked in the top 5 in the preseason poll reduced final BCS ranking by 17 spots, controlling for all other variables. Similarly, being ranked 6-10 in the preseason poll reduced final BCS ranking by almost 15 spots. However, the remaining coefficients were insignificant, meaning being ranked lower than 15th in the Preseason Coaches Poll did not have a significant effect on where a team was ranked in the final BCS rankings. We would thus expect this pattern to continue in terms of the Preseason Coaches Poll's effect on who plays in the actual BCS games. Since there was an established effect of the preseason Coaches Poll on the final BCS standings, I then ran a regression of BCSGame on PreCoachRank, inserting the same control variables. Because BCSGame is a binary dependent variable, I ran two regressions, probit and OLS. The purpose of these regressions was to see if being generally ranked in the preseason Coaches Poll gave teams an unfair advantage in being able to participate in BCS games, controlling for other variables. With the probit model, I controlled for all teams that had a winning percentage greater than .6, so the model could better estimate the relationship between playing in a BCS game and being ranked in the preseason Coaches Poll. The results showed a positive relationship between BCSGame and PreCoachRank, as PreCoachRank had a coefficient value (column 5, row 1) of -.0011, significant at the 95% level. Thus, the higher a team was ranked in the preseason Coaches Poll, including those in the Other Receiving Votes category, the more likely it was that teams would play in a BCS game, which supported the second hypothesis. The results of the OLS model also showed a positive relationship between BCSGame and PreCoachRank, as the coefficient on PreCoachRank (column 3, row 1) was -.0014, significant at
Hamill 16 the 99% level. This result showed that a decrease by one in the preseason rankings, increased the likelihood of playing in a BCS game by .14%, controlling for all other variables. Since there was a general result that being ranked high in the preseason Coaches Poll led to increased BCS game participation, I ran a further regression of BCSGame on each of the split variables, Pre 1-5, Pre 6-10, etc, inserting the same control variables as in previous regressions. The purpose of this regression was to see how high a team had to be ranked in the preseason Coaches Poll to have a significant unfair advantage over all teams ranked lower than it. Because BCSGame was again the dependent variable, I used both a probit and OLS model. The results of the probit model showed that being ranked 11-15 (column 6, row 4), not 1-5 (6,2) or 6-10 (6,3), in the preseason had the greatest effect on BCS game participation, with a z-value of 2.96, significant at the 99% level. Thus, being ranked in the preseason top 11-15 is an unfair advantage over all other teams ranked lower. Being ranked 1-5 was still an unfair advantage, as this variable had a z-value of 2.37 and was significant at the 95% level. However, these results rejected the third hypothesis, as the highest ranked teams in the preseason did not have the highest BCS game participation. While the coefficients on the variables for Pre 6-10, Pre 16-20 and Pre 21-25 were also positive, they were not significant. The OLS model revealed that being ranked 1-5 in the preseason Coaches Poll showed the greatest correlation with BCS game participation with a coefficient of .2839, significant at the 99% level. This meant that being ranked 1-5 in the preseason increased the probability of playing in a BCS game by 28.4 percentage points when controlling for all other variables. There was also 12.9 percent greater chance of playing in a BCS game being ranked 1-5 than being ranked 6-10 (15.5% increased probability) and 14.8% greater chance than being ranked 11-15 (13.6%
Hamill 17 increased probability). Still, Pre 6-10 and Pre 11-15 were both still significant at the 99% level. After these three groups, the drop off is substantial, with the next highest coefficient belonging to the Pre 16-20 group at .019, an insignificant result. Unlike the probit model, the OLS model accepts the third hypothesis, as the group with the highest probability of playing in a BCS game, controlling for all other variables, is Pre 1-5, followed by 6-10 and then 11-15. 7. Conclusion The purpose of this study was to get a definitive answer as to whether the preseason Coaches Poll has a significant impact on who plays in Bowl Championship Series, college football's version of prestigious postseason play. That is, did higher ranked teams in the preseason have a decisive edge over lower ranked teams in attaining this goal? Results that lead to a yes answer would mean that there is a huge problem with college football and how it handles its postseason play, and will be evidence for a necessary change to the system. If the results lead to a No Answer, then the preseason Coaches Poll would prove insignificant and suggest that preseason rankings seem to fairly balance out over the course of the season, with no damaging effects to a team's ability to compete in postseason play. After running all the regressions, the answer was clear: higher ranked teams in the preseason Coaches Poll have an unfair advantage over lower ranked and unranked teams. The two regression models used in the experiment, probit and OLS, both proved this point on differing levels. While the probit model revealed that only teams ranked 1-5 and 11-15 held this unfair advantage, the OLS model showed that the top 15 teams, represented by three groups (Pre
Hamill 18 1-5, Pre 6-10, Pre 11-15), were unfairly ranked higher than all remaining teams in Division 1-A college football, controlling for winning percentage, strength of schedule, conference affiliation, and final games played. As such, these results proved college football's playoff system needs to change. How it changes is up for debate, but one thing is certain: the preseason Coaches Poll does not belong.
Appendix Model: OLS Dependent variable: Final BCS Ranking Independent variables: Preseason Coaches Ranking Controls: Win %, Final Games, SOS, Major
Hamill 19
source model Residual Total
SS
df
MS
380202.895 5 76040.5789 281320.373 901 312.231269
661523.268 906 730.158132
Number of obs =
F( 5, 901) =
Prob > F
=
R-squared =
Adj R-squared =
Root MSE
=
907 243.54 0.0000 0.5747 0.5724 17.67
finbcsrank precoachrank win fingames sos major _cons
Coef. Std. Err.
t P>|t|
.1233834 -83.72254 -2.214143 .1249424 -3.142464 120.1582
.023509 3.206668 .9620972 .0288107 2.063207 11.88387
5.25 -26.11 -2.30 4.34 -1.52 10.11
0.000 0.000 0.022 0.000 0.128 0.000
[95% Conf. Interval]
.0772447 -90.01595 -4.102355 .0683986 -7.191715 96.83495
.1695221 -77.42913 -.3259302 .1814863 .9067875 143.4815
Model: OLS Dependent variable: Final BCS Ranking Independent variables: Preseason Coaches Ranking 1-5, 6-10, 11-15, 16-20, 21-25, ORV 26-30 Controls: Win %, Final Games, SOS, Major
Source Model Residual Total
SS
df
MS
393765.36 10 39376.536 267757.908 896 298.836951
661523.268 906 730.158132
Number of obs =
F( 10, 896) =
Prob > F
=
R-squared =
Adj R-squared =
Root MSE
=
907 131.77 0.0000 0.5952 0.5907 17.287
finbcsrank pre15 pre610 pre1115 pre1620 pre2125 orv2630 win fingames sos major _cons
Coef. Std. Err.
t P>|t|
-17.38065 -14.95716 -4.230679 -.0906779 -2.375392 .5865239 -81.04115 -2.071087 .1071703 -3.089647 126.1826
2.534913 2.44954 2.395142 2.316993 2.289358 2.403716 3.131007 .941944 .0284966 2.03602 11.39416
-6.86 -6.11 -1.77 -0.04 -1.04 0.24 -25.88 -2.20 3.76 -1.52 11.07
0.000 0.000 0.078 0.969 0.300 0.807 0.000 0.028 0.000 0.129 0.000
[95% Conf. Interval]
-22.3557 -19.76466 -8.931421 -4.638043 -6.868521 -4.131045 -87.18611 -3.91976 .0512424 -7.085571 103.8202
-12.40559 -10.14966 .4700627 4.456687 2.117736 5.304093 -74.89619 -.2224131 .1630982 .9062767 148.5449
Model: Probit Dependent variable: BCS game Independent variables: Preseason Coaches Ranking Controls: Win %, Final Games, SOS, Major, if Win% > .6
Iteration 0: Iteration 1: Iteration 2: Iteration 3: Iteration 4: Iteration 5:
log likelihood = -255.26821 log likelihood = -142.60047 log likelihood = -121.25897 log likelihood = -116.88781 log likelihood = -116.63203 log likelihood = -116.631
Probit regression, reporting marginal effects
Log likelihood = -116.631
Number of obs = 452
LR chi2(5) = 277.27
Prob > chi2 = 0.0000
Pseudo R2
= 0.5431
bcsgame
dF/dx Std. Err.
z P>|z|
x-bar [ 95% C.I. ]
precoa~k win fingames sos major*
-.0010741 2.019402 .0521408 -.0022387 .1186769
.0005462 .2934501 .0207861 .000751 .0278266
-1.99 10.07 2.60 -3.23 3.15
0.047 0.000 0.009 0.001 0.002
36.9845 .770139 11.7898 42.0465 .809735
-.002145 -3.4e-06 1.44425 2.59455 .011401 .092881 -.003711 -.000767 .064138 .173216
obs. P pred. P
.2522124 .0803415 (at x-bar)
(*) dF/dx is for discrete change of dummy variable from 0 to 1 z and P>|z| correspond to the test of the underlying coefficient being 0
Model: OLS Dependent variable: BCS game Independent variables: Preseason Coaches Ranking Controls: Win %, Final Games, SOS, Major
Source
SS
df
MS
Model 31.8151041 5 6.36302082 Residual 67.8563402 901 .075312253
Total 99.6714443 906 .110012632
Number of obs = 907 F( 5, 901) = 84.49 Prob > F = 0.0000 R-squared = 0.3192 Adj R-squared = 0.3154 Root MSE = .27443
bcsgame
Coef. Std. Err. t P>|t| [95% Conf. Interval]
precoachrank win fingames sos major _cons
-.0013702 .7060272 .0439416 -.0016675 .0161976 -.6898774
.0003651 .0498022 .0149422 .0004475 .0320433 .1845663
-3.75 14.18 2.94 -3.73 0.51 -3.74
0.000 0.000 0.003 0.000 0.613 0.000
-.0020867 .6082854 .0146161 -.0025457 -.0466907 -1.052107
-.0006536 .8037691 .0732671 -.0007893 .0790858 -.3276475
Hamill 20
Hamill 21
Model: Probit Dependent variable: BCS Game Independent variables: Preseason Coaches Ranking 1-5, 6-10, 11-15, 16-20, 21-25, ORV 26-30 Controls: Win %, Final Games, SOS, Major, if Win % > .6
Iteration 0: Iteration 1: Iteration 2: Iteration 3: Iteration 4: Iteration 5: Iteration 6:
log likelihood = -255.26821 log likelihood = -138.37225 log likelihood = -116.67888 log likelihood = -111.73601 log likelihood = -111.35608 log likelihood = -111.35341 log likelihood = -111.35341
Probit regression, reporting marginal effects
Log likelihood = -111.35341
Number of obs = 452
LR chi2(10) = 287.83
Prob > chi2 = 0.0000
Pseudo R2
= 0.5638
bcsgame
dF/dx Std. Err.
z P>|z|
x-bar [ 95% C.I. ]
pre15* pre610* pre1115* pre1620* pre2125* orv2630* win fingames sos major*
.1430175 .0566718 .2064596 .04129 .0216043 -.048844 1.994871 .052615 -.0020791 .1042499
.0799555 .0563889 .0934748 .0700401 .0679906 .0363495 .3047789 .0206644 .0007384 .0282671
2.37 1.18 2.96 0.68 0.34 -0.97 9.74 2.68 -2.98 2.75
0.018 0.237 0.003 0.499 0.732 0.334 0.000 0.007 0.003 0.006
.128319 .121681 .103982 .073009 .079646 .057522 .770139 11.7898 42.0465 .809735
-.013692 .299728 -.053848 .167192 .023252 .389667 -.095986 .178566 -.111655 .154863 -.120088 .0224 1.39752 2.59223 .012113 .093117 -.003526 -.000632 .048847 .159652
obs. P pred. P
.2522124 .0747255 (at x-bar)
(*) dF/dx is for discrete change of dummy variable from 0 to 1 z and P>|z| correspond to the test of the underlying coefficient being 0
Model: OLS Dependent variable: BCS Game Independent variables: Preseason Coaches Ranking 1-5, 6-10, 11-15, 16-20, 21-25, ORV 26-30 Controls: Win %, Final Games, SOS, Major
Source Model Residual Total
SS
df
MS
36.112426 10 3.6112426 63.5590183 896 .070936404
99.6714443 906 .110012632
Number of obs =
F( 10, 896) =
Prob > F
=
R-squared
=
Adj R-squared =
Root MSE
=
907 50.91 0.0000 0.3623 0.3552 .26634
bcsgame pre15 pre610 pre1115 pre1620 pre2125 orv2630 win fingames sos major _cons
Coef. Std. Err.
t P>|t|
.2838578 .154923 .1364489 .0208725 -.0346672 -.0352813 .6373046 .0402242 -.0012966 .0073107 -.7162422
.0390553 .03774 .0369019 .0356979 .0352721 .037034 .0482394 .0145125 .000439 .0313689 .1755496
7.27 4.11 3.70 0.58 -0.98 -0.95 13.21 2.77 -2.95 0.23 -4.08
0.000 0.000 0.000 0.559 0.326 0.341 0.000 0.006 0.003 0.816 0.000
[95% Conf. Interval]
.2072072 .0808539 .0640247 -.0491887 -.1038927 -.1079648 .5426293 .0117418 -.0021583 -.0542545 -1.060779
.3605084 .2289921 .2088732 .0909337 .0345583 .0374022 .7319799 .0687067 -.0004349 .0688758 -.3717058
Bibliography
Hamill 22
"Alabama is No. 1, followed by LSU then Oklahoma State". USAToday.com. January 2012. http://www.usatoday.com/sports/college/football/usatpoll.htm "BCS, Alliance & Coalition games, year-by-year". BCS Football.org. February 2011. http://www.bcsfootball.org/news/story?id=4809942 "BCS Standings, BCS Rankings". December 2011. http://www.collegefootballpoll.com/bcs_standings.html "CBS SportsLine ­ 1999 Preseason Coaches Poll". CBSSports.com. 1999. http://www.cbssports.com/u/ce/feature/0,1518,1235607_56,00.html "CNN/SI ­ College Football ­ Ohio State tops preseason coaches poll". CNNSI.com. August 1998. http://cnnsi.com/football/college/news/1998/08/06/coaches_poll/ "Coaches Poll Preseason 2001". National Champs.Net. August 2001. http://www.nationalchamps.net/NCAA/polls/Coaches_Poll_2001_Preseason.htm Coleman, B. Jay, Andres Gallo, Paul Mason, and Jeffery Steagall. "Voter Bias in the Associated Press College Football Poll". University of North Florida. 1 August 2010. http://digitalcommons.unf.edu/cgi/viewcontent.cgi?article=1002&context=bmgt_facpub "College Football Rankings, College Football Poll". CollegeFootballPoll.com. January 2012. http://www.collegefootballpoll.com/1999_archive_computer_rankings.html Kotchen, Matthew and Potoski, Matthew. "conflicts of interest Distort Public Evaluations: Evidence from the Top 25 Ballots of NCAA Football Coaches". NBER Working Paper Series. November 2011. http://www.nber.org/papers/w17628 "NCAA College Football Polls and Rankings". ESPN.com. 2012. http://espn.go.com/collegefootball/rankings "NCAA Division 1 FBS teams, rosters, game schedules, photos, articles and videos". Fanbase. 2011. http://www.fanbase.com/NCAA-D-I-FBS Sorenson, Soren. "College Football 1998 Division 1-A Teams Chi Square Linear Win Difference ­ Ratio". August 2001. http://www.phys.utk.edu/sorensen/cfr/cfr/Output/1998/CF_1998_Ranking_Best.html "USA Today-ESPN Top 25 Coaches Poll". National Champs.Net. August 2000. http://www.nationalchamps.net/NCAA/polls/Coaches_poll.htm
Hamill 23 Wetzel, Dan, Josh Peter, and Jeff Passan. Death to the BCS. Penguin Group Inc., New York. 2010. Pg. 115

File: the-effect-of-the-college-football-preseason-coaches-poll-on.pdf
Title: Microsoft Word - HamillDillon Thesis
Author: LabAdmin
Published: Tue May 29 09:56:37 2012
Pages: 23
File size: 0.15 Mb


, pages, 0 Mb

Polymer handbook, 6 pages, 0.29 Mb

Crossing the chasm, 61 pages, 1.68 Mb

Al-Tusi, connaissez-vous, 12 pages, 0.42 Mb
Copyright © 2018 doc.uments.com