Playing with statistics part 3

After a general rant in part 1 and part 2 I am about to start actually doing some stats.

Ok, the first 30-15IFT test we did was in January and here are the scores.

		TEST
		January 14th, 2012.
Player		v30-15IFT	Z-Score
1		19,5	-0,89
2		19	-1,57
3		19,5	-0,89
4		20,5	0,45
5		19,5	-0,89
6		20	-0,22
7		21	1,12
8		21,5	1,79
9		21	1,12
10		19,5	-0,89
11		19,5	-0,89
12		19,5	-0,89
13		20,5	0,45
14		20	-0,22
15		21,5	1,79
16		20	-0,22
17		21	1,12
18		20	-0,22
Mean		20,17	0,00
SD		0,75	1,00
Min		19,00	-1,57
Max		21,50	1,79

As you can see from the table I have calculated Mean and Standard Deviation. Z-score (or standard score) for each athlete is the number that shows how many standard deviations players score is above/below mean value. Having 0 in z-score is equal to mean value of the group. Standard score is great because it take variability of the group scores into account, yet again this might present a problem since some outliers can skew the score by shifting the distribution. This skewness stuff can be calculated too and it deals with normal distribution. Again I am not an expert on this. Anyway, here are the histogram and scatter gram of the scores. 

Standard score is great for comparing the athletes and creating the rankings. Here is the graph of Z-scores of the player.

 

Now it is easier to identify outliers or guys who are above/below 1SD or 2SD or whatever (which one to choose is beyond my statistic knowledge at the moment). This might guide training prescriptions for certain athletes. For example guys with Z-Score below -1 can/should do more conditioning volume (and less intensive due their lower v30-15 score which I use to determine running intensity in the intervals). After some time we can see how they respond to training (this will be covered later) and identify responders vs. non-responders and thus give some feedback to planning/programming process and individualization of training loads in general. 

Basically we can group those players in four major groups: 

Low initial score, low training response	Low initial score, high training response
High initial score, low training response	High initial score, high training response

Based on this we can judge on talent of certain players regarding certain quality. But we still lack one parameter that it is hard to measure and that is the ceiling, but this is beyond this article. 

It would be interesting to see the distribution of the athletes in these 4 quadrants over time. Again this might help with the training prescription and individualization within team sport.

If we create z-scores for more than one physical quality (like sprint time, broad jump, % of body fat, etc) we can create spider graph for each player. Here is an example of spider chart from Marco Cardinale’s blog.

The spider graph can be based on Z-score if we want to compare the player to the team average, or it could be based on absolute values if we want to compare it to a certain model of player or certain test standards (make sure to check BioForce by Joel Jamieson). Again, this can guide us with training prescription and individualization.

Another interesting graph could be equalizer. I got this idea from Marco’s blog.

 

To summarize. In this part I covered some basic descriptive statistics of static score (one test). In the next installment I will cover re-test and the statistical analysis of the change and will probably create a 4 quadrant’s graph. Until then check the links posted. Stay tuned….

Playing with statistics part 2

After a general rant in part 1 I will now try to be more pragmatic. Please note that my knowledge of statistics is basic so don’t be harsh with the comments. And if you have any good tip, suggestion or resource please be free to post.  Besides, please note that with these posts I am not trying to impress lab coats, but rather provide an example for my fellow coaches who deal with testing and assessment. Nothing fancy, just down to earth analysis. 

For an example I will use 30-15IFT test and re-test we did during the pre-season. For more info about this test use the search function on the blog. I really like this test, but somehow I think it could improve its sensitivity by utilizing more ‘capacity’ component. What I do mean is that for example athletes improve vVO2max for certain degree, but what they also improve is the ability to maintain vVO2max pace for longer time (tlim) and I think that this is even more trainable than ‘power’ component. Check these articles: Billat and Heubert. For example if your vVO2max is  18km/h and you improve it to 19km/h with training that’s equals to 5,5% improvement, but if your tLim at 18km/h was 300sec and after that same training protocol you improve it to 500sec, that is 66% improvement. Someone correct me if I am wrong. Again, I am referring to tLim at constant speed and not % of vVO2max (because that might decrease as you improve vVO2max). 

30-15IFT more than YoYo IRT test  assess this ‘power’ ability (intermittent endurance), thus it is less sensitive to training changes. In YoYo test athlete accumulate distance by performing reps at certain pace (thus more ‘capacity’ measure), but in 30-15IFT they only get to repeat one 30sec interval at certain pace that increments for 0,5km/h every rep. This is why the 30-15IFT is less sensitive to training.  In more statistical term, if one improves 30-15IFT from 20km/h to 21/km that is 5% improvement. On the flip side if someone improve score in YoYo test from 2,300m to 2,700m that is 17%. I don’t know how to calculate statistical significance at the moment and other fancy statistical measures, but hopefully I will know in a month or two. Till then bear with my basic statistical knowledge. 

Please note that power~capacity is always involved in every test in different proportions and that the scores (both performance wise and physiological wise, like bLA) are always task dependent. Also, improving one aspect of performance might not mean that the performance itself improved. Think about it. And check this article by Steve Magness. 

I have some ideas how to modify this test, or maybe supplement it with MAS test or even intermittent endurance test that demands repetition of certain pace until exhaustion (e.g.  repeating 30sec @20km/h with 15sec rest until exhaustion), but that would be a whole another article. Until then check Dupont article.

This is getting long. I will cut here and actually come back with test and re-test scores in the next post.

Playing with statistics part 1

I must admit that my recent obsession is statistics and variability. Statistics got my attention when I started playing/working in Excel to create monitoring system for my current club utilizing data from Polar Team2. 

 

Recently I have seen EXCELLENT presentation from Stuart Cormack regarding monitoring in rugby. They utilize GPS, biochemical monitoring, sRPE and Neuromuscular fatigue testing and create light system that red-flags players that need more attention and are away from their usual variability, both to their individual baseline and team baseline (they utilize standard difference score and rolling averages over 4 weeks). 

Interestingly enough they DON’T utilize HR monitoring. I agree with that, but at the moment I don’t have resources to track GPS, biochemical monitoring and jumping/sprint performance, thus I need to utilize what I do have. Anyway, here is the first part of the presentation.

If I ever plan getting licensed besides my B.S. Degree and Tudor Bompa International, I might get it from ASCA. I am somehow in love with their pragmatical work.  And besides they have Dan Baker as the President. 

When it comes to variability (which is ultimately related to mentioned standard difference score) I came across it years ago, but recently it knocked on my doors in the form of HRV and Robert Sapolsky’s lecture. These lectures are a MUST watch and they are new addition to Biology of Behavior lectures I already referenced here.

The impact that variability has on the measurement and evaluation can be seen in Heart Rate Variability research and monitoring systems. I started playing with iThlete and BioForce HRV and monitor my HRV every morning. I am also communicating with Joel Jamieson about getting more HRV sensors for some of my athletes and trying to put some correlations between the Polar Team2 workload and HRV reaction (it would be nice to have measured scores like jump and sprint performance as well and GPS, but I don’t have it at the moment). For more info about HRV I highly suggest checking the blog by Andrew Flatt and the forthcoming book on HRV by Joel Jamieson at 8WeeksOut.

 

If you want more info on variability in general and how it affects measurement, evaluation and everything you can check the book by prof.Keith Davids. I don't have this book, but I plan getting it soon. I must admit that it is expensive though.

Back to statistics. Since I am basically new to this area, I know only the basic stuff and not especially in detail. So, I was thinking about spending some time reading New View of the Statistics by Will G Hopkins. I know that one of my professors recommend this more than any book on the subject. 

 

Speaking on all of this (monitoring, statistics, blah, blah) I am thinking about going back to MATLAB (see the short video here). I was pretty fluent with MATLAB during 2004-2005, but then I quit working with it. I started using MATLAB to experiment with one motor control model (Lambda Model) during that time. If you are interested here is the paper in Serbian. I remember that back then I discovered Euler and Runge-Kutta methods of simulation on my own and then lately I found out about them. Funny. I should have been an engineer, but then I wouldn't need to deal with soccer players :).  Also, here is the link to one of my applications I develop during that time. 

Anyway, I am thinking about rehearsing my MATLAB skills and using it to collect, analyze and visualize the data, maybe even create a system for monitoring. The problem is that this would take away my time from developing coaching skills and put me more in the lab coat category/extreme. Besides I am also interested into performance analysis tools like   SportsCode. Soccer is way behind baseball (volleyball as well) in terms of stats and analysis. And yes I am being influenced by Moneyball. 

 

In the next part I will try to be more practical and do an analysis of test and re-test of 30-15IFT by Martin Buchheit. Stay tuned…

Good reads

I just wanted to give you a heads up for the articles written by Howard Gray at his blog. His paper/poster on sRPE/TL (Polar Team2) influenced me to create my own monitoring system. This is my way to express my gratitude to Howard. Enjoy the links!

Preseason: Low volume & always avoiding fatigue?

Scientific Poster: Pilot Study - Tracking Session RPE TRIMP During an NCAA D1 Men's Soccer Season with Special Emphasis on Practical Application

Sport Science Poster: Between the Polar Team2 Training Load and Session-RPE Training Load as Methods of Monitoring Training and Games in NCAA Division 1 Male Soccer Players

The effectiveness of the strength and conditioning coach

Monitoring training part I - can't see the wood for the trees

Monitoring training part II - using RPEs to track intensity & volume

Monitoring training part III - calculating & analyzing load, monotony, and strain

 The other side of monitoring - keeping track of outside stressors

  

I am a little bit sick, so I was staying home for the last two days and I have been reading Science of Winning by Jan Olbrecht. Some interesting info, but I am really confused by his capacity/power definition.  Hell, going back to reading....