Well I've finally gotten around to putting together the long over due final part of our three part series examining the accuracy of the AP Pre-Season Poll. The past week I was immersed in a stringent training routine for a cornhole tournament. It involved numerous 12 oz curls, which consumed much of my time. However, our hard work was all worthwhile when we were eliminated in the first round.

In our first part, we quantified the success (or failure) of the AP Pre-Season Poll. In the second part, we attempted to shed some light on these numbers by comparing the success of the AP polls to that generated by a computer. While these analyses have shown that the AP Pre-Season poll is typically not at all successful in predicting the eventual national champion, let alone the final rankings.

This begs the questions though, "so what?". What harm is there in publishing a pre-season poll when voters can revise these rankings as the season plays out? Isn't it only important that they voters get it right in the end?

Comparison of Pre-Season and Final Rankings for Undefeated-Untied Teams

Year	Team	Pre	Final
1950	Oklahoma	6	1
	Army	2	2
	Princeton	NR	6
	Wyoming	NR	12
1951	Tennessee	1	1
	Michigan St.	2	2
	Maryland	16	3
	Princeton	18	6
	San Fransisco	NR	14
1952	Michigan St.	1	1
	Georgia Tech	3	2
1954	Ohio St.	20	1
	UCLA	8	2
	Oklahoma	2	3
1955	Oklahoma	2	1
	Maryland	5	3
	Miami (OH)	NR	15
1956	Oklahoma	1	1
	Tennessee	12	2
1957	Auburn	15	1
	Arizona St.	NR	12
1960	Yale	NR	14
	New Mexico St.	NR	17
1961	Alabama	3	1
	Rutgers	NR	15
1962	Southern Cal	NR	1
	Mississippi	6	3
1964	Alabama	6	1
	Arkansas	NR	2
1968	Ohio St.	11	1
	Penn St.	8	2
1969	Texas	4	1
	Penn St.	3	2
1970	Arizona St.	19	6
	Toledo	NR	12
	Dartmouth	NR	14
1971	Nebraska	2	1
	Toledo	NR	14
1973	Notre Dame	8	1
	Penn St.	7	5
	Miami (OH)	NR	15
1976	Pittsburgh	9	1
	Rutgers	NR	17
1991	Miami (FL)	3	1
	Washington	4	2
1994	Nebraska	4	1
	Penn St.	9	2
1997	Michigan	14	1
	Nebraska	6	2
1998	Tennessee	10	1
	Tulane	NR	7
1999	Florida St.	1	1
	Marshall	NR	10
2004	Southern Cal	1	1
	Auburn	17	2
	Utah	20	4

Getting it right in the end is all that matters, but when voters have pre-conceived biases, it is undoubtedly difficult to fairly judge factors that should differentiate two teams (e.g. strength of schedule, quality of victories). As an anonymous commenter pointed out after the first part in the series:

. . . putting a team in the top 5 for the pre-season poll can be a self-fulfilling prophecy -- in the sense that once a team is put up that high, losing is the ONLY way they can fall. Simply "finding a way to win" should not be ample credentials for staying "in place" in the polls.

Fans of college football have undoubtedly seen this habit of voters play out in the polls in recent years. To attempt to quantify this, years in which there were two or more undefeated-untied teams, were examined, comparing those team's pre-season and final rankings. Of the 56 years the AP has published a pre-season poll, there have been 23 seasons when two or more teams have finished undefeated-untied and, therefore have a claim to the national title.

In 16 of those 23 seasons, those undefeated teams finished the season in the exact order they were ranked in the pre-season poll. Only 7 times did the AP voters reconsider their pre-season rankings. It should be noted that in 8 of those seasons, one of the undefeated teams was from a minor conference and could be considered by some, not a legitimate contender for a national title.

The table to the right shows the undefeated teams from these seasons with years the AP voters reversed their rankings shown in red, the seasons the voters held to their rankings in green, and years where the voters held to their rankings but the second undefeated team was not from a major conference show in black.

In the seven years the voters changed their ranking order, most of the time, the teams were ranked closely in the pre-season poll, no more than four places apart in 1950, 1968, 1969, and 1973.

Notable exceptions are 1954 when pre-season #20 Ohio St. passed #8 UCLA and #2 Oklahoma for the national championship. Oklahoma fell below both Ohio St. and UCLA after a narrow 13-6 victory over unranked Colorado. Ohio St. benefited from victories over then #2 Wisconsin and #12 Michigan. UCLA, despite also defeating two ranked teams, would fall victim to a weak Pacific Coast Conference schedule and finish second in the AP. It should be noted that the United Press Coaches Poll did rank UCLA number one that year, resulting in a split national championship. Sadly, these two teams did not get to meet in the bowl season as the Rose Bowl selected 8-4 Southern Cal to face Ohio St.

In 1962, an unranked Southern Cal team would claim the national title despite pre-season #6 Mississippi also winning all of their games. Southern Cal cracked the top ten after winning their first game against pre-season #8 Duke (Duke was ranked in a pre-season poll?). Their steady climb continued throughout the year and they assumed the top spot when Alabama lost to Georgia Tech. Mississippi on the other hand could not impress the voters with what proved to be a weak early schedule against the likes of Kentucky, Houston, Tulane, and Tennessee-Chattanooga. While an eventual win over then #4 LSU helped, it was too little, too late.

The final example of the AP voters reconsidering their pre-season ranking came in 1997 when Michigan leapfrogged Nebraska to claim the AP National Title. Michigan made its jump over Nebraska after a lopsided victory over then #2 Penn St. That week Nebraska struggled against unranked Missouri. Without a chance to meet in a bowl game, a win over Washington St. earned the Wolverines their second Associated Press title; however, as was the case in 1962, the writers and coaches did not agree, and a split national title was awarded.

Early in the pre-season poll's history, even in years where the order of the final rankings matched that of the pre-season rankings, the AP voters have shown the ability to reassign their rankings during the course of the season based on the performance of the teams to date. Take the 1951 season where five undefeated teams were ranked in the same order at season's end as they were in the beginning. This, however, does not reflect the dynamics of the poll as the season played out. In that year, there were four changes in the number one ranking. At one point, eventual #2 Michigan St. was ranked fifth below Maryland and Princeton.
In recent years though, this is the exception, not the rule. In the first 28 years of the pre-season poll's existence, the voters reconsidered their pre-season rankings six times, meaning that in the last 28 years, the voters have done so only once (Michigan-Nebraska, 1997). The recent trend has been win-and-stay-put. Take the 1991 season, shown graphically below. Did Washington really ever have a chance at the National Championship as long as Miami did not lose? A much more dramatic example came in the 2004 season, when five teams finished the regular season undefeated, but there was only room at the BCS inn for pre-season #1 Southern Cal and #2 Oklahoma. With the exception of a week 1 flip-flop between Auburn and Utah and a brief tie for number two between Oklahoma and Auburn, the voters keep each team ranked in the same order they had cast at the beginning of the season. It is difficult to argue that strength of schedule and quality of victories are considerations in the voters' minds when the ballots seem to be so mindlessly filled out.
We hope this series has shed some light onto preseason polls, namely:

• The polls do predict the National Champion more often than one might think (10 out of 56, 18%)
• Most of the teams predicted to finish in the top 5, do finish near the top (within 5 places of original rank 57% of the time)
• On average, 40% of the teams ranked in the pre-season poll finish the season unranked.
• The likelihood of a team ranked 6-10 finishing in the predicted spot is the same for teams ranked 21-25 (3.2%)
• The new pre-season poll is only marginally better than simply using the final poll from the previous year.
• It is even possible for a computer to out pick the humans.
• There is a strong correlation between the position of a team in the pre-season poll their position in the final poll.

While this analysis has been fun, unfortunately, the consequences of pre-season polls can be very serious. Since in the past human polls determined the National Champion, and today, human polls are the largest contributor in determining which two teams play for the National Championship, if a quantifiable bias exists in the polls, every step should be taken to remove it. In this case the solution is simple:

Don’t publish a pre-season poll!

While this doesn’t prevent individual pollsters from having their own pre-season biases, it will prevent the kind of group-think the voters so frequently demonstrate.

The Harris Interactive Poll took the first step when it announced that it would not issue its first poll until late September, when most teams have three games under their belt. The problem of course is that voters in that poll are still influenced by the AP and Coaches Polls, so the pre-season media biases are still present. In the first Harris Poll published last year, no team was further than two places from the AP Poll rankings.