Level playing field across ages and across events

[sjors]

think you'll find most erg WRs are done by athletes in the 30-40 age group than the 20-30.

Take a look at https://www.rankedworkouts.com
Select your filters and discover it's fifty/fifty.

[faach1]

For anybody who is interested, I have made csv-export's of all best performances for the normal rower.
They are split into: distance/time, weight-class and gender
In each file there are rows for the best performance per age as long as there are more than 5 performances for that age.
You can download it from here: https://www.rankedworkouts.com/share/performances.zip

I’m interested in this Siebe, but the url doesn’t work for me when I click on it. It says it’s not found.

[sjors]

For anybody who is interested, I have made csv-export's of all best performances for the normal rower.
They are split into: distance/time, weight-class and gender
In each file there are rows for the best performance per age as long as there are more than 5 performances for that age.
You can download it from here: https://www.rankedworkouts.com/share/performances.zip

I’m interested in this Siebe, but the url doesn’t work for me when I click on it. It says it’s not found.

O, my mistake it should be
https://www.rankedworkouts.com/docroot/ ... mances.zip

[MudSweatAndYears]

Regarding your formula, I am surprised that distance performances decline after 20 given many world champions are older and older athletes seem to decline a little slower in the longer events. Also the research suggests that there is acceleration in performance lost through aging somewhere around 50 and again around 70 IIRC, your formula has no such discontinuity.

A key element in the formula is the power of two in the dependency on aging. It might see a detail, but it is absolutely essential: age does not enter as ‘years above 20’ but as ‘years above 20 squared’. This adequately addresses the challenges you list above:
1) according to thee equation performance hardly declines in younger athletes (athletes in their twenties), the decline slowly builds up and becomes significant only at much older age.
2) the formula doesn’t need a discontinuity (in fact a discontinuity in performance drop would be a weird artifact in any such formula). In the above equation the performance drop with age accelerates (as data indicates), and results in an ever faster performance drop.

[MudSweatAndYears]

I have been busy for a while and did not read this threat. Very interesting formula's.
Did you find something similar for the shorter work where it seems that age does not have the same impact on performance as does endurance?
As you maybe know I'm trying to solve this issue as well for my site.

The shorter distances render things more complex. In particular the simple power-law distance-time scaling is no longer valid. It seems quite difficult to describe both aerobic and anaerobic performance in a single formula. Perhaps a single formula for 500-2000 m would be feasible, have not looked into that yet.

[Carl Watts]

Interesting concept but you cannot derive a formula to include drugs, that of course nobody is using so your never going to be able to create a truly level playing field with just numbers and no testing. Would work if you had two fields though, all those not using drugs and all those using drugs.

[sjors]

In the above equation the performance drop with age accelerates (as data indicates), and results in an ever faster performance drop.

After studying age-decline in several distances, I have found that the acceleration in the drop of performance is quite subtle

[MudSweatAndYears]

I have looked at the age correction for 2000 m. Using the above methodology I arrive at:

Tcorr = T (1 - ((age-20)/maxyrs)^2)

I fit this equation to the WR data using the single fitting parameter 'maxyrs'. The result is quite interesting: it appears that lightweights age more gently than heavyweights (applies to both males and females):
M, Hwt: maxyrs = 120
F, Hwt: maxyrs = 115
M, Lwt: maxyrs = 130
F, Lwt: maxyrs = 135

[iain]

Interesting discussion. Thanks for the analyses.

MS&Y, how did you determine this parameter? Was it a standard regression and of what datapoints? There is considerable differences between the curves. I would expect some differences due to differing importance placed on training as well as the absolute ability levels of the exceptional athletes that set each of the world records. Have you tried alternative data (90th percentiles perhaps)?

I am still loathe to reject the conclusions by several meta studies that VO2max decline accelerates after 70 and is small for those training under 40 based on these reports. But I am prepared to accept that the relationship is complex and effected by other factors (including weight and age). Essentially you have set yourself a very difficult task and will never produce a perfect answer! Also, the data on decline in those that continue to train vigorously suggests that this is significantly lower although the number of athletes in this category that have been tested are too low to quantify this, especially if sub-divided by age and weight.

[MudSweatAndYears]

MS&Y, how did you determine this parameter? Was it a standard regression and of what datapoints?

The single parameter fit is determined by minimising for each category the variance in the corrected times.

I am still loathe to reject the conclusions by several meta studies that VO2max decline accelerates after 70 and is small for those training under 40 based on these reports. But I am prepared to accept that the relationship is complex and effected by other factors (including weight and age).

The quadratic dependency on age-20 ensures that the age correction is small for under 40 and progressively larger after 70.

Essentially you have set yourself a very difficult task and will never produce a perfect answer!

I agree THE perfect answer does not exist. But that doesn’t mean improvements upon existing algorithms are not possible. The key lesson here is that with more robust few-parameter fitting procedures one can do way better than the complex age correction algorithm applied in the Nonathlon competition.

[iain]

Hi MS&T, thanks for your response. I was not completely clear on your methodology, is this using all data in the rankings with equal weighting? i would have thought that this would still adjust for changes in training with age and would be looking to try and make the average person who logs distances equal. whether this is appropriate for a competition will depend upon how the data is spread for different categories. Without an analysis I would be worried that:

1) younger performers (or perhaps those around their peak) are likely to be more self-concious about ranking and therefoe for their to be a reduction in weaker performances ranked

2) Much reduced ranking from those most competitive in their prime (not wanting to give info away to international competitors and possibly domestic competition for boat seats).

3) Bias in open weight category from a significant number of rankers at a significantly reduced level of fitness who are erging to lose excessive weight, particularly compared to lightweights where the majority need to be in shape to make weight in the first instance.

As for comparison with Nonathlon, the results will be different. This explicitly sets out to standardise the average scores of the fastest at each age for each category, this ignores all data from slower rowers in setting the standard. Studies have shown that there is a significant decline in the performance of older athletes that can be mitigated by training. As such I would expect that the average relative performance actually falls. That said, the self-selection for Nonathlon will favour those maintaining fitness for at least part of the declining time. This is particularly so as to be competitive requires the performance of a maximal HM. I suggest that the proportion of erg users prepared to face this will decline in older age groups and so the average score may be reflective of different entrants as much as biased results. Further, the Nonathlon is the only major individual erging competition I am aware of (at least with wide entry) where older athletes can compete across the spectrum and so there may be more concentration on this rather than races etc where they would normally find few competitors. Conversely prime athletes will see little prestige in maximising the Nonathlon compared t competing for Hammers or at other prestigious events or importantly OTW where there is a prestigious olympic sport taht is the focus of many of the most talented candidates.

Finally, with improved longevity, baby boomers entering older categories and increased knowledge of training needs for this cohort performances have improved faster in older age groups than the younger. I would expect this to be reflected in more older winners.

As for your quadratic function, I note that this will give greater weighting to older athletes, but in a particular weighting that may or may not be appropriate. Previous suggestions have been for linear decline at 2 rates, this is almost certainly not the best fit, but it will be significantly different to a quadratic formula that will aid those approaching the first age of decline and give more support to the very old. This may be appropriate, but considerable data would be required before this could be established.

However, these are superficial considerations and the project remains a significant step forward in evaluating the performances of different categories. Personally I think both this and the nonathlon are huge steps forward to the American age handicapping system. It would be interesting to see a side by side comparison of your ratings with Nonathlon results so people could evaluate which they see as the better assessment.

[MudSweatAndYears]

Hi MS&T, thanks for your response. I was not completely clear on your methodology, is this using all data in the rankings with equal weighting?

I just fitted to the WR data themselves. So in terms of data similar to the Nonathlon approach. I can use other data (e.g. the 90% percentile in the ranking data, I believe you made this suggestion earlier?).

Without an analysis I would be worried that:

1) younger performers (or perhaps those around their peak) are likely to be more self-concious about ranking and therefoe for their to be a reduction in weaker performances ranked

2) Much reduced ranking from those most competitive in their prime (not wanting to give info away to international competitors and possibly domestic competition for boat seats).

3) Bias in open weight category from a significant number of rankers at a significantly reduced level of fitness who are erging to lose excessive weight, particularly compared to lightweights where the majority need to be in shape to make weight in the first instance.

Yes, these are all valid concerns. However, by basing the fit on WR data, I think most of these concerns diminish in importance.
Obviously, the disadvantage of using WR data is that there is much less richness in that data set (compared to all C2 ranking data), so there is always the danger that a single outlier can influence the age-correction relationship significantly. Fortunately, a single-parameter fitting procedure is much less sensitive to such outliers.

As for your quadratic function, I note that this will give greater weighting to older athletes, but in a particular weighting that may or may not be appropriate. Previous suggestions have been for linear decline at 2 rates, this is almost certainly not the best fit, but it will be significantly different to a quadratic formula that will aid those approaching the first age of decline and give more support to the very old. This may be appropriate, but considerable data would be required before this could be established.

I don't think more data is needed to establish the non-linear decline with age. A quick look at the curved trends on the left-hand side of above plot for 2k WR times as function of age establishes the non-linear decline.

It would be interesting to see a side by side comparison of your ratings with Nonathlon results so people could evaluate which they see as the better assessment.

Good idea. I could do this. But we need to establish what are the criteria for identifying the better assesment. I personally think an approach that would result in more diversity (age diversity + gender diversity + weight diversity) amongst the annual winners will render motivation to compete to a larger groep, and therefore is 'better'. But not everyone will agree with that criterion.