Freezing Saddles 2015

[dkel]

@rcannon100 100874 wrote:

And how do you account for a top twenty rider who did not participate last year?

Can we go back to talking about me?

[dasgeh]

@rcannon100 100862 wrote:

Any further thoughts? Again, I like Steve’s modified random – but the problem is, I think, that it cannot account for new ringers who would through the statistics. If a mythical Tom Kellie were to be assigned to one of the teams and if Tom Kellie had not played last year – this mythical uber rider would throw the game.

Does that bring us back to straight up handicaps? And if so, based on what period? September through November?

We need to move to consensus. Which is the right option?

Why not have everyone enter handicaps, but only use them to find the top 20 riders, then randomize that group, and randomize the next group? Solves the new riders problem AND solves the issue of “what if we need more than 10 teams”

[ShawnoftheDread]

@dasgeh 100877 wrote:

Why not have everyone enter handicaps, but only use them to find the top 20 riders, then randomize that group, and randomize the next group? Solves the new riders problem AND solves the issue of “what if we need more than 10 teams”

I don’t think the top 20 should be randomized. It defeats the purpose if identifying the top 20 to begin with, which is to make sure they don’t end up on the same team.

[dasgeh]

@ShawnoftheDread 100880 wrote:

I don’t think the top 20 should be randomized. It defeats the purpose if identifying the top 20 to begin with, which is to make sure they don’t end up on the same team.

No, the top 20 would be randomized so 2 are on each team. Then the rest randomized to fill out the teams. And if we have, eg, 12 teams, we take the top 24 and randomly put them 2 per team.

[ShawnoftheDread]

@dasgeh 100881 wrote:

No, the top 20 would be randomized so 2 are on each team. Then the rest randomized to fill out the teams. And if we have, eg, 12 teams, we take the top 24 and randomly put them 2 per team.

But if they’re randomized, you could end up with the top 2 on the same team and 19 and 20 on a team. Lame. The top 20 system described earlier already accounts for the possible need of more than 10 teams – you just take the first 11 or 12 or whatever it is and then the next 11 or 12.

[dasgeh]

@ShawnoftheDread 100883 wrote:

But if they’re randomized, you could end up with the top 2 on the same team and 19 and 20 on a team. Lame. The top 20 system described earlier already accounts for the possible need of more than 10 teams – you just take the first 11 or 12 or whatever it is and then the next 11 or 12.

Either you go with last year’s miles , which assumes a lot and doesn’t work for newbies or you go with people’s handicaps, which is out of their heads.

I think you have to go with handicaps

Either way, you’ll have some variance between reported and actual, such that even report #1 and reported #2 end up on the same team, that doesn’t mean that those will be actual #1 and #2 at the end.

But it’s two separate decisions: 1) what numbers do you base the system on (i think you have to use handicaps) and 2) what to do with the top tier ( i feel less strongly that randomized is best there)

[Steve O]

Points:

– We used handicaps last year and the results were no better than if the riders had been randomly assigned.
– Modified random significantly improves on randomly assigned based on analysis of actual data.
– Modified random improves on the handicapping system used last year 85% – 90% of the time.
– Handicaps for newbies are not reliable–no better than random.
– Handicaps based on past performance are better.

Therefore using handicaps on known riders will improve competitiveness. Using handicaps on new riders is a waste of effort.

That is the reasoning behind my recommendation of modified random.

Additional points:
1) Modified random mostly indemnifies the team assigner’s system from being second guessed
2) Modified random cannot be tested at the end of FS to see how it did against the handicapping system, but not vice versa. That may be good for data analysis, but invalidates point #1 if it turns out MR would have been more competitive (which I will bet you lunch it will be).
3) MR is easier to implement

[vvill]

I’d think assigning the top 20 as proposed by Steve O (assuming 10 teams) would work pretty well.

I think after that it’s still worth considering doing buckets/pools (and then randomly picking an equal number from each). If you look at the points from last season, it’s not as if the points suddenly drop off after the top 20.

So… assuming say 10 teams and 100 riders:
Top 20
Pool A next 20
Pool B next 20
Pool C bottom 40

For newbies, they can self-nominate to be in either A/B/C. You can’t predict their points, so don’t bother trying too hard.

[Rod Smith]

I’m surprised that anyone gave themselves a handicap that was LOWER than the number asked for (miles and days ridden in Nov. and Dec.) I gave myself a handicap greater than the number asked for because using my Nov. and Dec. miles would have been unfair. I can’t believe a significant number of us intentionally deceived.

If some of us cheated to our advantage (or to our disadvantage), I assume most of us did neither. I’m sure some riders didn’t have a record of the miles they rode in Nov. and Dec. so they estimated and I assume the estimates were almost all in good faith, as accurate as possible. I feel handicaps, based on real numbers or estimates are going to be mostly close enough for our purposes.

So if Chris Watson or other big mile rider joins the forum and the challenge, we should use her handicap and if that puts her in the top twenty, assign her to a team based on that, not add her to the pool of randomly assigned participants. Chris was fourth placed rider on our National Bike Challenge team, ahead of Jeff Lemieux, Greg Garrett, Chris Randall, Dickie Sewell.

[Rod Smith]

Also, pairing #1 with #20, #2 with #19, #3 with #18 etc. and assigning the rest randomly will give my team a big advantage (almost 500 points) over the other teams. Not that there’s anything wrong with that. #800poundgorilla

On the other hand, 500 points is less than 3% of the winning team’s total last year so perhaps not worth worrying about.

[dasgeh]

@Steve O 100887 wrote:

Points:

– We used handicaps last year and the results were no better than if the riders had been randomly assigned.
– Modified random significantly improves on randomly assigned based on analysis of actual data.
– Modified random improves on the handicapping system used last year 85% – 90% of the time.
– Handicaps for newbies are not reliable–no better than random.
– Handicaps based on past performance are better.

Some of these are based on your one analysis of one year of data, and I don’t think you looked at people’s reported handicap last year in your analysis (at least you didn’t discuss it in the write up). Others of these are pure conjecture.
So why not put an entry on registration for handicap*? Then when we know the number of registrants, we can tier into buckets based on reported handicap.

* we have to define handicap which could be A) avg weekly mileage from bafs last year, avg weekly mileage from Oct – Dec, C) estimated weekly mileage for 2015 bafs or D) something else. All of those options have their issues.

[peterw_diy]

Random random random. Easy, no sandbagging, no hard feelings.

Happy Thanksgiving, everyone!

[dcv]

seunpu

[hozn]

I say do what the data says makes sense. We have handicap info from last year, though it is “massaged” (corrected, when it underreported wrt strava) data so not sure if that helps this analysis really.

I know some of you have asked for specific datasets. Sorry, I have little time to give that now, but am happy to provide db access to anyone that wants to build/explore the data directly.

Or jrenaut may have more time to support right now. (Or others that already have db access.)

[ShawnoftheDread]

@Rod Smith 100891 wrote:

Also, pairing #1 with #20, #2 with #19, #3 with #18 etc. and assigning the rest randomly will give my team a big advantage (almost 500 points) over the other teams. Not that there’s anything wrong with that. #800poundgorilla

On the other hand, 500 points is less than 3% of the winning team’s total last year so perhaps not worth worrying about.

What would give you a bigger advantage: being purposely placed with #20 or being randomly placed with #4 and #5?

I think completely random shouldn’t be used because of the possibility or the latter. As for modified random or handicaps, good arguments have been made for both. But I think modified random would be easier to implement, as I’m sure Bob spent a lot of time creating the teams the past two years.

[Author]

Posts