Median not Average and exclude 1 point matches
under review
J
John Pollard 🇬🇧 (johnypoll)
Sometimes you can get a very high match PR after making a small mistake as one of very few moves, then being stuck on the bar for the rest of the match. Galaxy use median PR rather than average, which I suggest makes more sense as it excludes outliers. Also removing 1 point matches from the overall PR would be a big improvement, they make the overall PR fairly meaningless.
A
Alfie Kirkpatrick [Developer]
John Pollard 🇬🇧 (johnypoll), am open to making changes in this area if clear what they should be!
Using median rather than average should be straightforward but PR is shown in a few different places...
- The instant challenge dropdown shows mean for the last 25 matches. Simple to change to median
- The Hub ladder shows mean for matches in the last 6 weeks. Again should be simple to change to median
- The player profile page shows the mean of all matches displayed (according to any filter applied). I think here we might want a toggle to show mean/median just to give flexibility
Regarding excluding 1 point matches, this is tricky because some players _only_ play 1 point matches. Also the PR for short matches such as 3 or 5 points also typically display lower PRs. I wish there were some kind of scale factor based on match length but have never seen one, and maybe this would just cause more confusion!
Do you see any issue using median over a small 25 match sample?
J
John Pollard 🇬🇧 (johnypoll)
Alfie Kirkpatrick [Developer] I'm quite excited that this change might go through, thanks for picking it up.
Could the instant challenge and hub ladder just show the same overall PR as your profile page? I think it makes more sense to be consistent. Is there a strong argument to tailor them?
On the profile page if it were me I'd take the easy option and just go to median without a toggle, perhaps with a note for a while to let people know when it changed. Personally I'd find it interesting to have a mean/median toggle, but I'm a bit geeky like that and a cleaner interface without it might be preferable.
You could change the All label to "All (3+)" and perhaps for players that have only played 1 point matches, select the 1 point filter for them when the page displays. One way or another, for serious players wanting to improve/compare PR, I think 1 needs to be separated off.
The difference in PR for 3+ match lengths is very interesting. You might notice another post I made where I was hoping you might query the database to see what the averages actually are at each match length. I suspect there isn't a huge difference and that by excluding 1 pointers, you have a fairly good overall figure. I couldn't find anything online about actual real world PR differences at different match lengths, you could be a pioneer in finding this out!
Do I see an issue with median over a small sample? I can't see an issue, though with only 2 values in a sample you might want to take the mean of them and only switch to median for 3+ in the data set.
B
Backgammon Hub Admin
under review
J
John Pollard 🇬🇧 (johnypoll)
Can I give a nudge on this one, PR is such a key thing for measuring progress. I think the two points above both have merit independently of each other, but both together would be a big step forward in making the PRs more representative. Oh, also implementing the change would be fairly simple in coding terms I believe, so would not sap your resources :)