Welcome to Project Left Turn

Welcome to Project Left Turn, a fun look at using statistics to understand the Indiana University Little 500. I completed this project in 2011 as an MBA student at the Kelley School of Business in Bloomington, IN.

2011 Race Predictions

It probably comes as no surprise that the model is predicting the Cutters to be this year’s winner. The rest of the field is as follows:

Place	Team
1	Cutters
2	Delta Tau Delta
3	Sigma Chi
4	Phi Delta Theta
5	Beta Theta Pi
6	Black Key Bulls
7	Phi Gamma Delta
8	Sigma Nu
9	Phi Kappa Psi
10	Delta Chi
11	Acacia
12	Cru Cycling
13	Gray Goat
14	Hoosier Climber?
15	Kappa Sigma
16	Theta Chi
17	Air Force Cycling
18	Emanon
19	Delta Upsilon
20	#JungleExpress
21	Pi Kappa Alpha
22	Achtung
23	Sigma Phi Epsilon
24	Sigma Alpha Mu
25	Evans Scholars
26	Dodds House
27	LAMP
28	Sigma Pi
29	Wright Cycling
30	Phi Kappa Sigma
31	CSF Cycling
32	Delta Sigma Pi
33	Sigma Alpha Epsilon

 

Five teams are within the standard error of the model and thus have a shot at the win. Winning probabilities for the top five teams are listed below.

Team	Chance  of Winning
Cutters	35%
Delta Tau Delta	15%
Sigma Chi	13%
Phi Delta Theta	12%
Beta Theta Pi	8%

How it works:

A regression of historical data reveals that 68% of the race outcome can be explained by ITT times, team pursuit times, and qualification times. The ITT time of the team’s fastest rider and the team pursuit time were the most significant variables to explain race outcome. The ITT time of the third rider was insignificant and not included in the regression model. The model can predict a team’s race time to within 161 seconds. The expected race times are assumed to be normally distributed. This assumptions allows for simulation of 1000 iterations using the standard error of 161. This revealed each team’s chances of winning as listed above.

The model assumes that a team’s four fastest riders ride in the race. This assumption is, of course, imperfect but it provides the only simple way to regress ITT times to race outcome. All times are normalized to their respective yearly averages. This accounts for anomalies in weather, track conditions, and excessive yellow flags.

Can you beat the computer?

We all know the Little 500 is mostly unpredictable. And even I know that a purely stastitical approach may not be the best way to make predictions. It is for this reason I would like to know: can you beat the computer?

If you think you can, email me your predictions by noon on Wednesday. Send me an excel spreadsheet with a list of men’s teams and finishing positions. Email files to eandreol@indiana.edu with subject “Prediction Contest”.  Predictions must be made for the entire field of 33 teams. After the race I will compare your predictions to my computer predictions by caclulating the correlation coeficient.

As an added bonus the submitter with the highest correlation coefficient will recieve a free Little 5 themed Kilroy’s poster courtesy of Kilroy’s Was Here. Winners will be announced after the race.

Best of luck. Look for my race predictions Wednesday at noon!

Why we run the race

Saturday’s team pursuit results prove that all Little 500 event outcomes can be unpredictable. My excel model, which was based on ITTs and quals, had Cutters winning by 12 seconds. Instead, Cutters finished third and Delts secured the win, despite the computer only giving Delts a 9.3% chance of winning.

Here is a look at each team’s actual versus predicted results.

Team	Actual	Predicted
Delta Tau Delta	1	4
Sigma Chi	2	7
Cutters	3	1
Phi Delta Theta	4	2
Beta Theta Pi	5	3
Black Key Bulls	6	5
Phi Gamma Delta	7	9
Hoosier Climber?	8	12
Gray Goat Cycling	9	16
Cru Cycling	10	11
Delta Chi	11	10
#JungleExpress	12	27
Phi Kappa Psi	13	8
Kappa Sigma	14	13
Wright Cycling	15	29
Sigma Nu	16	6
Theta Chi	17	14
Delta Upsilon	18	18
Sigma Pi	19
Emanon	20	21
CSF Cycling	21
Acacia	22	17
Sigma Phi Epsilon	23	22
Achtung	24	28
Cru Cycling B	25
Air Force Cycling	26	15
Dodds House	27	19
Sigma Alpha Mu	28	20
Phi Kappa Sigma	29	26
LAMP	30	23
Sigma Alpha Epsilon	31	31
Delta Sigma Pi	32	25

 

The predictions were not as accurate as last year’s race predictions, but still performed better than expected with a .78 correlation coefficient. Twenty teams were predicted within 5 spots of their actual performance.

Look for race predictions to be posted by Wednesday.

Team Pursuit Predictions

With team pursuit underway, I wanted to make some predictions. It should be noted that the prediction model assumes that all qualifying teams (except CSF Cycling and Sigma Pi) will be riding in TP. This is already not accurate, as it seems like a few teams have not shown up.

The predictions are based on multivariate regression of historic qualification and ITT times. Significant variables were found to be a team’s fastest rider, a team’s third fastest rider, and a team’s average adjusted qualification time. These variables combine to explain approximately 57% of TP results. The model is not as strong as the race-day model, and should not be taken too seriously.

Here are the predicted results:

1	Cutters
2	Phi Delta Theta
3	Beta Theta Pi
4	Delta Tau Delta
5	Black Key Bulls
6	Sigma Nu
7	Sigma Chi
8	Phi Kappa Psi
9	Phi Gamma Delta
10	Delta Chi
11	Cru Cycling
12	Hoosier Climber?
13	Kappa Sigma
14	Theta Chi
15	Air Force Cycling
16	Gray Goat
17	Acacia
18	Delta Upsilon
19	Dodds House
20	Sigma Alpha Mu
21	Emanon
22	Sigma Phi Epsilon
23	LAMP
24	Pi Kappa Alpha
25	Delta Sigma Pi
26	Phi Kappa Sigma
27	#JungleExpress
28	Achtung
29	Wright Cycling
30	Evans Scholars
31	Sigma Alpha Epsilon

 

The model predicts the Cutters will beat Phi Delt by as much as 12 seconds, with the next four teams all being within 5 seconds of each other.

I will follow up tomorrow to compare the actual results.

 

How much does team pursuit matter?

Earlier this week, I posted a riveting analysis on the usefulness of quals as a predictor of race outcome. Scroll down or click here to read to the full scoop.

With team pursuit coming up in a few days, I think it is worthwhile to perform the same analysis with TP results.

Using the same statistical techniques, it can be concluded that a team’s TP position can explain 68% of a team’s race day outcome. Thus TP performs much better than quals as a sole predictor of race day performance.

While a team only has an 8.5% chance of finishing the race in its exact TP position, it is important to note that 70% of teams finish within 5 spots of their TP position. 96% of teams finish within 10 spots and no team has ever finished the race more than 16 spots away from its TP position.

Here is a look at TP winners and their following race performance. (TP was cancelled in 2006 and 2002)

 

Year	TP Winner	Race Finish
2010	Phi Delta Theta	2
2009	Black Key Bulls	5
2008	Cutters	1
2007	Cutters	1
2005	Phi Gamma Delta	2
2004	Cutters	1
2003	Gafombi	1
2001	Cutters	7
2000	Sigma Phi Epsilon	8

 

 

For the nerds: Here is the scatterplot. Notice how there is less scatter as compared to the quals plot.

A note on ITTs

After looking through men’s ITT data, I was certainly impressed by the 4 second difference between first and second place. In fact, Young receives the honor of enjoying the largest margin of victory in this decade.

A more intriguing question, however, is how Young’s time compares to the overall field. By analyzing each year’s top 50 average, we can determine how the winning time compares to the 50 fastest riders.

Year	Winning Rider	Winning Time	Top 50 Average	Difference
2011	Eric Young	142.00	151.94	9.94
2010	Eric Young	142.09	150.41	8.32
2009	Eric Young	138.25	147.20	8.95
2008	Issac Neff	139.75	147.67	7.92
2007	Sasha Land	138.94	146.54	7.60
2006	Hans Arnesen	137.68	146.80	9.12
2005	Hans Arnesen	135.70	146.81	11.11
2004	Chris Vargo	141.49	150.03	8.54
2003	John Grant	143.38	150.98	7.60
2002	Luke Isenbarger	145.00	151.60	6.60
2001	Josh Beatty	149.87	159.93	10.06
2000	Chris Wojtowich	144.28	154.38	10.10

Interestingly, Young does not hold the fastest average adjusted time. This honor goes to Hans Arnesen, whose blazing 135.70 was 11 seconds faster than the top 50 average.

This suggest that while Young was undoubtedly the fastest rider last Wednesday, the gap between the fastest rider and the field is not as large as we saw in 2005.

How much does quals matter?

How much of a stastical impact does a team’s starting position have on race outcome? The nerdiest way to answer this question is to determine the correlation between qual position and race outcome. Using men’s data from 2000 to 2010, I performed a single variable regression using qual position as the independent variable and final race position as the dependent variable.

This analysis reveals that 49% of a team’s race day performance can be explained by the qual position. This certainly suggests that while qualifications are very important, a team’s starting position can not be used as a sole predictor of race outcome.

A team only has an 8% chance of finishing in the exact same spot it qualified. About half (52%) of all teams finish within 5 spots of their starting position, while 74% of all teams finish within 10 spots. Teams very rarely finish more than 15 places away from their starting position.

The 2008 ATO team receives recognition for qualifying in 26th and finishing the race in 3rd. Since 2000, this was the biggest difference between quals and race day performance.

Here is a look at each of the pole teams since 2000.

Year	Pole Team	Race Finish
2010	Cutters	1
2009	Phi Delta Theta	15
2008	Sigma Alpha Mu	14
2007	Phi Kappa Psi	2
2006	Cutters	5
2005	Phi Kappa Psi	6
2004	Team Major Taylor	4
2003	Phi Gamma Delta	9
2002	Phi Delta Theta	11
2001	Phi Gamma Delta	13
2000	Delta Chi	4

For the nerds: The scatterplot showing the relationship and trend line. Note the large scatter.

Last Year’s Results

Welcome to Project Left Turn, a statistical look at the Little 500. Over the next two weeks, I hope to draw excitement for race day in the nerdiest way possible: by analyzing data and making predictions. Before we get into this years data, let’s review what I predicted last year.

Below is a chart of my predictions versus the actual results. Teams in green performed better than prediction while teams in red performed worse.

So how well did I do? A high correlation suggests that there is significantly strong relationship between the predictions and actual results. The model, however, did not do a good job of predicting the exact place a team would finish. In fact, Phi Sigma Kappa was the only team to be accurately predicted.

However, 11 teams were predicted within one spot of their actual finishing position and 17 teams were within two spots. This suggests the model does a great job of predicting the general area a team will finish.

Below is a chart showing the number of teams who were within a given difference between predicted and actual finishing position.

Special props go to Dodds house for finishing 10 spots better than predicted. But what happened to Acacia? They finished 14 spots below the prediction.