What is the grade-averaging algorithm?!

Look at the MtnPrjct "statistics" for about any climb and you'll probably find something like this:

100 climbers "ticked" it, of these..

about 10-20 bothered to give it a "star" rating but did not voice their opinion of the grade,.... of the 10-20 who "star-ed" it,

about half (5 or 10) also gave it a suggested grade rating, and

80-90% ( 4 of 5,  to about 9 of 10) of those stayed with the climb's "given" grade rating.

So whatever the algorithm is, it's unlikely the grade will change.    
 

Robert Hall wrote: Look at the MtnPrjct "statistics" for about any climb and you'll probably find something like this:

100 climbers "ticked" it, of these..

about 10-20 bothered to give it a "star" rating but did not voice their opinion of the grade,.... of the 10-20 who "star-ed" it,

about half (5 or 10) also gave it a suggested grade rating, and

80-90% ( 4 of 5,  to about 9 of 10) of those stayed with the climb's "given" grade rating.

So whatever the algorithm is, it's unlikely the grade will change.    
 

Certainly not based on the info you've posted.

I think the issue is more along the lines of, when someone enters a route in the database, and one person rates it different, then the rating changes.

From what Nick has said/told us it would have to be a BIG difference.  Ah.. but wait, didn't he say it actually returned the MEAN (not the AVERAGE); if so then one different could affect it. ??

Nick...what's the story here?

We use the MEDIAN grade as the consensus.  So when there are very few submissions (they start with one, from the OP), the consensus is very sensitive.  As the route gets more ratings, an outlying rating will have no affect.

Nick Wilder wrote: We use the MEDIAN grade as the consensus.  So when there are very few submissions (they start with one, from the OP), the consensus is very sensitive.  As the route gets more ratings, an outlying rating will have no affect.

I think this makes eminent sense.  Especially since you can get really out-of-line ratings (like 5.9+ suggested for a consensus 5.12a for Days of Future Passed, https://www.mountainproject.com/route/105758080/days-of-future-passed-aka-our-future-has-passed).

The problem is easily solved.
Go back to the F rating system used at the Monument back in the 70s, also known as the, NCCS. ( National Climbing Classification System)

It is messy.  But most of the time gives you a ball park idea.  The more folks comment on a route's grade, then usually the consensus will become closer to reality.  I too like the letters over the +s & -s.
I set a few routes and find that sometimes certain moves on said route have height issues.  Most of the time it sucks for the shorter climber, but not always.  So here is where I like to give a rating like 11a/b.  That is to say, a bit harder for the shorter climber.  The 5.11 is IMO the 11b/c.  The + seems more often than not to mean a kick ass 11 and the - an easier 11, but the letters tell more.  

Cosmiccragsman AKA Dwain wrote: The problem is easily solved.
Go back to the F rating system used at the Monument back in the 70s, also known as the, NCCS. ( National Climbing Classification System)

A more coarse grading system would make for less grade averaging, as, there's less grades divided amongst climbing routes.

I think most folks like a bit more refined grade to argue about (ha ha).

We could convert to the French system like most of the rest of the climbing world...

Confusing is the British system.  I'm guessing it makes perfect sense to those accustomed to its intricacies.

I think we should convert to a numbers-only grading system (a la Australian or the bouldering V-scale).  Because with 11d and 12a, a lot of people will prefer to climb a "5.12" because it makes them feel so much more accomplished than a "5.11".   Kind of like how items are priced $9.99 in stores to make them seem cheaper.

And first ascentionists know this, so they grade their route 5.12a hoping it will stick because they want people to climb the route (and 11d's get skipped over a lot).

If you have the Aussie scale, it's just 20, 21, 22, ... no letters and no psychology.

A question more in line with this thread: if I go to a climb that has two ratings of (say) 5.11b and 5.11, the consensus is given as "5.11b".  So if I then add my own rating of 5.11, the median should change to 5.11, right?  But it doesn't... it just stays at 5.11b.  Same bug as described up-thread?

John RB wrote: A question more in line with this thread: if I go to a climb that has two ratings of (say) 5.11b and 5.11, the consensus is given as "5.11b".  So if I then add my own rating of 5.11, the median should change to 5.11, right?  But it doesn't... it just stays at 5.11b.  Same bug as described up-thread?

It would be nice if there was a handy explanation, or link to an explanation, next to the grade. 

The latest algorithm I remember is this:

1) The person who posts the route has his grade counted twice.
2) The other ratings are added to the list, low to high.
3) The median grade is listed. In case there is an even number in the list, the lower of the two medians is used.

Jon Nelson wrote:

It would be nice if there was a handy explanation, or link to an explanation, next to the grade. 

The latest algorithm I remember is this:

1) The person who posts the route has his grade counted twice.
2) The other ratings are added to the list, low to high.
3) The median grade is listed. In case there is an even number in the list, the lower of the two medians is used.

Thanks for posting this Jon.  That makes a lot of sense: give the OP more weight (and double weight feels about right), and then take the lower of two in an even list.

The "normal" way to compute a median of an even-numbered list is to take the mean of the middle two items, but that's a mess if the middle two are (say) 11c/d and 11d.  You can't put 11c/d+ right? :)

So yeah, sounds like what you guys are doing makes a lot of sense. :)

Is the OP's grade visible anywhere?

John

Admins: I think this might be a bug.

To test the algorithm, I used Problem Child (at Primo), rated 13b/c.  ( mountainproject.com/route/1…) Rated 13b/c on MP.  The submitter was Luke Childers.

Luke Childers was the submitter, so his counts twice.  That means we have
13b
13b
13b
13b
13b
13a/b
13a/b
13-
13a
12d

Which is an even numbered list, so the route was given 13a/b (the lower grade) just as Jon Nelson said!  

To test, I added a rating of 13b, and sure enough the overall grade rose to 13b!  But then I deleted my rating of 13b and MP did not revert it to 13a/b.  Bug?

edit: I then added another rating of 13a, which DID move the overall back to 13a/b, then I deleted my rating again and it stayed at 13a/b.  I think I'll stop testing now. :)  But something weird is happening here.

John (and all), route pages are cached for up to a week, and I don't believe that adding or changing your rating invalidates the cache, so you are unlikely to see changes in real time.  If you are the OP of a route, you can edit/save it (you don't need to make any actual changes) and that will force the page to be rebuilt fresh.

What do you do with routes that have variations and people rate them both? The consensus then end up being the average. 

Nick Wilder wrote: John (and all), route pages are cached for up to a week, and I don't believe that adding or changing your rating invalidates the cache, so you are unlikely to see changes in real time.  If you are the OP of a route, you can edit/save it (you don't need to make any actual changes) and that will force the page to be rebuilt fresh.

Thanks Nick.  But I was able to make real time changes by adding a rating (and then refreshing the page).  But removing the rating didn't revert the change (strangely).  Then I added a lower rating and refreshed the page, causing another change downward.  And then removed the rating and it stayed.

Anyway, I'm sure if I saw the code it would all make sense, but looking from the outside it seems kind of perplexing.  In any case, this is a corner case and I doubt matters much.  Thanks!