Barnett evaluates Games competitors by the numbers

2010 CrossFit Games Finals

Jeff Barnett's Statistical Analysis of the 2010 Games

One Man's Look into the Data

The CrossFit Games are the proving grounds for effective fitness methodology. From workout programming to recovery techniques, if something makes you fitter it can be validated by the results of the Games. Unfortunately, as with all statistics, the data from the Games does not always immediately lend themselves to conclusions. CrossFit competition is still in the early stages of development, and we do not have full access to the methods used by the competitors. Nonetheless, it is possible to glean some insight from the results. 

Jeff Barnett, CrossFit Impulse, has taken one shot at interpreting the data we have from the CrossFit Games. Barnett decided to compare the competitors' self-reported workout performances from before the Games with their results at the 2010 CrossFit Games. While we have not yet vetted Barnett's statistics, he reaches some interesting conclusions about the interplay between athletes' training results and their performance at the Games.

Download Barnett's full article as MS Word HERE.

Download Barnett's full data set as XL HERE.

Download Barnett's analysis as XL HERE.

Rogue Fitness Again Faster Equipment CrossFit Journal

Proudly supporting the 2010 CrossFit Games

Follow crossfitgames on Twitter

Follow the CrossFit Games on Twitter

The CrossFit/USA Weightlifting Open

The CrossFit/USA Weightlifting Open

13 comments on this entry

1. Hari wrote...

I am extremely impressed with the analysis. Thank you!

I am curious whether the conclusions hold up when you separate the athletes into two groups, those who were cut and those who were not.

We can look at Games as two separate competitions, the part that included all the athletes, and the part that included only those athletes who survived the cut.

To make a analogy, there are certain skills that might be critical for getting into college but not for getting through college. That is, those who make it to the next level all share a common set of attributes that are not necessarily as common in the larger group applying for admission.

2. Jeff Barnett wrote...

Great point. I agree that the results might be very different if I only included athletes that made the cut. However, the conclusions would be even more fast and loose because of the limited data set. 43-45 data points carry some statistical significance. 24 data points--much less. 16 data points--very little.

I suppose my assumption was that the regional events that fed the final competition were also thorough tests of overall fitness. I am sure that varied from region to region, and probably introduced some variance into the results.

Thanks for the feedback!

3. Hari wrote...

One possibility that uses all the data would be to run the analysis using the standings prior to the cuts.

My guess is that this would attenuate or eliminate the unexpected inverse relationship between endurance and standings among the men.

4. Jeff Barnett wrote...

Oooh, now that's an idea! I'll take a look at how much effort is involved in reconstructing the standings before the cuts. However, I'd also want to examine the difference in standings from pre-cut to final. If they are significantly different then we've just traded one bad assumption for another.

5. Hari wrote...

The relative standings change less and less after each subsequent event. (This is particularly true when the number of people competing in the final events is significantly fewer than those competing in the earlier events.)

So, perhaps it makes sense to determine the correlations with the totals used to determine the standings as opposed to the standings themselves.

For example, a hypothetical athlete who was a distant 16, far behind number 15, might turn in a very strong performance on the final day, but move up only slightly.

(As an aside, this is an argument to assign points based not on absolute standing but on relative standings, something you can now do ex post. So, if there were forty athletes in the earlier events, they would be rated 1, 2, 3, . . . 40, for each WOD. Then, if there are 16 athletes in the final events, they would be rated 2,5, 5, 7,5, . . . 40. As a practical example, this would draw appropriate focus on a top athlete who comes in last in a WOD on the second day.)

Finally, I think that the test you describe would be best run by comparing the standings of the final 16 before the cuts to their relative standings based on only the subsequent events.

6. Jeff Barnett wrote...

Hari, I ran the numbers on the men comparing pre-cut placement and final placement. The two placements are almost identical. Then I ran a correlation analysis using the pre-cut placement. All of the variables were very similar, including the inverse correlation with endurance. Looks like that was a dead end.

Your further explanation might yield something interesting if implemented, but I'll leave that one to you.

7. Hari wrote...

Not a dead end at all. It further validates your conclusions.

Nice work!

8. Catlyn wrote...

Hey, I'm not sure if anyone else is having this problem. I'm on a Vista machine with IE 8 and when I hover my mouse cursor over the hyperlinks, I can see URLs for the Word and XLS files. However, when I click to download, I get sent ZIP archives that include numerous XML and other files that I can't open. Help! I'd love to see the data.

9. Jeff Barnett wrote...

Catlyn, try hovering over and then right-click, save as (or save target as).

10. Cliff wrote...

Jeff, Very interested to read your analysis, but I can't get it either (tried method you described). Help?

11. Justin Smith wrote...

How does the order of the exercises come into play? For example, can one use these results to predict outcomes of a contest that uses a different exercise order?


12. Jeff Barnett wrote...

Cliff and Catlyn,
To make sure people can access the article and that it's searchable via Google, I have reproduced the article on my affiliate's site here:

Justin, the assumptions of this analysis don't allow for order of exercises to be examined. I can't think of a way you could analyze that statistically without volumes and volumes of data (many years of the CrossFit Games).

13. Luke Mullen wrote...

Suggestions for Crossfit Games (CFG) 2011:
(1) no athlete can compete in CFG2011 if they have not reported their scores for a TBD list of named WODs. No score reported means disqualified at start of CFG2011.
(2) first event of CFG 2011 is Crossfit Total (CFT)
(3) last event CFG 2011 is CFT

The goal being to have measure by which to gauge/predict fitness. It would also fill in the various holes in the analysis you presented; although, maybe those who did not report a named WOD score never had a chance in the first place?


The 2010 CrossFit Games Official Sponsors

For more information on how to sponsor the CrossFit Games, see the sponsorship page.

  • Under Armour
  • Rogue Fitness
  • Again Faster Equipment
  • MiR Pro Weighted Vest
  • Concept 2
  • Vibram Five Fingers
  • OMG*Omega3
  • Stud Bar
  • Regupol America
  • Mindbody Online
  • Forged Clothing
  • GarageGym
  • Gymboss Interval Timer
  • Inov-8
  • At Large Nutrition
  • LifeAsRx
  • LifeAsRx
  • Undefeated Sports Nutrition
  • Rage Fitness Supply: Leader in Functional Fitness Equipment
  • Stronger Faster Healthier
  • Watermans Applied Science
  • Rocktape
  • DeFeet
  • AquaHydrate