cancel
Showing results for 
Search instead for 
Did you mean: 
AdamVD
Pico de Orizaba
Status: Archived

It appears Strava has changed the algorithm for segment times. Looking into the data, it may be that average speed calculated over a segment is now more critical in final calculation of segment time rather than position data of rider moving over the segment with (time_at_end - time_at_start).

This appears to have changed about the same time that new segments of less than 500m were no longer able to be created. So, my guess is that position GPS position data is, for whatever reason, no longer a viable option for segment times. Because of this, it is clear in segment analysis that older times have an advantage over newer times. If segments are analyzed comparing 2013 - 2018 to 2019-2023, it is clear, it was somehow easier to set a fast time in the past.

Options are:

- Re-calculate old segment times using the new algorithm so leaderboards are fair again. Some may loose their KOMs, but based on the current algorithm, they never should have had them in the first place.

 - Change the algorithm back

- Give newly set times an automatic handicap to be X% faster so they are comparable to older data

 - Split leaderboards in "Pre-2018" and "Post-2018" (or whenever analysis indicates the change in algorithm made getting a top 10 much harder)

 - Delete old times

 

Thanks in advance. I think the KOM is sacred and having a KOM from 2016 stand just because the algorithm has changed does not respect that. I like Strava looking to make segment times more accurate, but I don't like that old times have an advantage in the leaderboards now.

8 Comments
Jan_Mantau
Denali

Do you have an example where (time_at_end - time_at_start) isn't used for the segment time?

AdamVD
Pico de Orizaba

Sorry, I tried to dig up an example but it is really tricky to define because the data is the data if that makes sense. I think statistical analysis would be more appropriate such as "where the KOM is set before 2016, what percentile of that KOM are the times, on average, for any leaderboard time set after 2016". For example, if the KOM is 1:00 in 2016 and the rest of the top 5 is 1:07, 1:10, 1:11, 1:13. With this, the KOM time is both an outlier in terms of when it was set and in terms of speed because all other top 5 results are very close. Because all other times are very close, assuming they were not all set on a group ride (which is rare), it implies they are all relatively close to a maximum effort. If the top 5 were set more than 5 years after the KOM, it would also suggest this because there has been a long time period for people to get close to the KOM. However, in such instances, the KOM has stood. I've noticed this trend on more than one occasion but don't have the statistical analysis capability to prove the trend. I hope Strava can look into it. 

AdamVD
Pico de Orizaba

I found 1 example: https://www.strava.com/segments/8378427?%243. It should be the case that a pre 2018 time remains in the top 5 shouldn't it? With KOM inflation, only the best times remain at the top, so through time, the older times should slip further and further down; however, it seems with some segments, this is not the case and only relatively recently is the KOM stolen. It does seem that there is some underlying advantage given to times set pre 2018. I would expect, with competitive riders, it is unlikely that everyone was giving it full beans for some reason before 2018 then stopped trying. It seems more likely the GPS algorithm must have changed somehow. Another analysis strava could do is look at riders and their KOMS. For riders who have most of their KOMs before 2018 but continue to be as active, why did their performance generate KOMs in 2018 but no longer does? You'd think, if it were a level playing field, their old KOMs would fall if they no longer can accumulate KOMs at a similar rate. Anyway, as I say, I just see the trend but lack the statistical analysis tools to understand why its there. Machine learning could be a good avenue to try and understand what is going on. What I know for sure is, if I'm going out riding and try to take a KOM set before 2018, I know it's going to be much harder to beat than something set post 2020

anchskier
Denali

While I can't rule out that something may have changed in their segment time calculations, I don't feel like what you have stated is conclusive in proving that a change was made.  There are a number of factors that can impact segment times from the past compared to those of more recent times.  I will use the area I live as an example.  5+ years ago, we had a very serious and competitive field of racers in their mid 20s to late 30's.  A bunch of us got pretty into Strava and there was a lot of KOM hunting going on and a lot of strong competition between each other for those KOM times.  In recent years, a lot of those people have either stopped competing as much or just gotten older and can't compete with their previous times on most segments.  We also don't have many young kids coming into the scene, and those that do are not hooked on racing for KOMs.  Many of our leaderboards look just like what you found as an example.  We put down some really fast times 5-10 years ago that nobody has touched in years.  Mostly because nobody is really trying, but also because the focus is not on fast segment times or just that we are older and slower.  Yes, I still go out and chase an old segment from time to time, and get them as well, but NOTHING like I used to.  It used to be a daily thing, not it is something I might do a couple times a month at best.  The others around me are the same.  The level of excitement over segment times just isn't there like it used to be, thus less push to beat the old times.  

Again, they may have changed something in their calculations, but I am not convinced.  

AdamVD
Pico de Orizaba

Yeah, a very reasoned response. Thank you. I suppose the "This year" board exists so we can always look there to see something compared but maybe this would be better as a board for time set in the last 365 days rather than Jan-Dec. If it was rolling 365 days, there would effectively give a leaderboard of KOMs with a 365 day expiration date. That could be very fun!

Now I've looked into it a little more, yeah, aside from the idea of a rolling board, I am included to agree that maybe no change is needed but the statistical analysis would be very interesting - especially if Strava cares about if KOM appetite is a metric on app popularity/use. Lots of marketing potential in getting the best time

AdamVD
Pico de Orizaba

For this segment, there is proof the algorithm is taking finishing speed into account. I finished with exactly the same time as my previous KOM, and was ahead at the line but my speed at the line was slower, so my old KOM stood. https://www.strava.com/segments/19356245

AdamVD_1-1689159168338.png

 

 

 

Jan_Mantau
Denali

That's nothing new, you never get a PR or KOM for repeating the same time.

Status changed to: Archived
Jane
Moderator Moderator
Moderator

Hello!

Thanks for sharing this suggestion to improve accuracy of segment leaderboards.  We have a suggestion here that has a similar goal.  

We will be archiving this suggestion to keep the forums organized, ensure that suggestions are easy to find, and prevent votes from being divided across duplicate posts. Don’t forget to show your support by clicking on the kudos button and adding a comment to the thread.


Jane (she/her)
STRAVA | Community Hub Team