Can someone pass this on to slugberg

Sachhyam257

I was just browsing Auntm.ai and I noticed something. I think the crit rating numbers and percentages are accurate, but the percentile data is wrong. It shows that elsa is on the 48th percentile for base crit rating, when there is only one champ with a higher base crit rating (shocker). It shows shocker on the 100th percentile correctly, but it is inaccurate for all other characters (they are all below 48th percentile).

I'm not sure if percentile data for other stats have this issue or not, I just noticed the crit rating one.

MCOCHazza

I think it’s just skewed because shocker’s base crit rate is like double elsa’s, so as a percentile it looks a bit funky because of the spread

Sachhyam257

MCOCHazza said:

I think it’s just skewed because shocker’s base crit rate is like double elsa’s, so as a percentile it looks a bit funky because of the spread

Doesn't 48th percentile mean 52% of entries scored higher? How can that be possible when there is only one entry (shocker) that is higher? Is that not what percentiles do?

DNA3000

Sachhyam257 said:

MCOCHazza said:

I think it’s just skewed because shocker’s base crit rate is like double elsa’s, so as a percentile it looks a bit funky because of the spread

Doesn't 48th percentile mean 52% of entries scored higher? How can that be possible when there is only one entry (shocker) that is higher? Is that not what percentiles do?

Generally, yes. However, what Auntm.ai appears to be doing with those percentile bubbles is specify what percentile of the extant range of the stat the champion falls in, not where in the total champion population the champion falls in.

For example, consider 6* R5 Black Widow CV. Her Armor Rating is 563, and it shows as 37 percentile of that stat. The extant range is listed as -428 to 2250. Every 6* rank 5 champion (presumably) falls within that range. BWCV thus falls at the 37% mark relative to that range. The total size of the range is 428 + 2250 = 2678. BWCV is 428+563=991 points higher than the lowest point. 991/2678=0.37. So BWCV is in the bottom 37% of the total range of base armor stats.

This is not a standard use of the term “percentile.” I’m actually not sure, off the top of my head, if there is a standard term for what Auntm.ai is calculating. I would call that “percent of extant range” but then nobody would know what I was talking about. Knowing Slug, I suspect he knows this as well, and chose a term that people would be familiar with over precise but opaque (and very verbose) language as a compromise.

Also, I had no idea people actually looked at those percentile numbers.

Sachhyam257

DNA3000 said:

Sachhyam257 said:

MCOCHazza said:

I think it’s just skewed because shocker’s base crit rate is like double elsa’s, so as a percentile it looks a bit funky because of the spread

Doesn't 48th percentile mean 52% of entries scored higher? How can that be possible when there is only one entry (shocker) that is higher? Is that not what percentiles do?

Generally, yes. However, what Auntm.ai appears to be doing with those percentile bubbles is specify what percentile of the extant range of the stat the champion falls in, not where in the total champion population the champion falls in.

For example, consider 6* R5 Black Widow CV. Her Armor Rating is 563, and it shows as 37 percentile of that stat. The extant range is listed as -428 to 2250. Every 6* rank 5 champion (presumably) falls within that range. BWCV thus falls at the 37% mark relative to that range. The total size of the range is 428 + 2250 = 2678. BWCV is 428+563=991 points higher than the lowest point. 991/2678=0.37. So BWCV is in the bottom 37% of the total range of base armor stats.

This is not a standard use of the term “percentile.” I’m actually not sure, off the top of my head, if there is a standard term for what Auntm.ai is calculating. I would call that “percent of extant range” but then nobody would know what I was talking about. Knowing Slug, I suspect he knows this as well, and chose a term that people would be familiar with over precise but opaque (and very verbose) language as a compromise.

Also, I had no idea people actually looked at those percentile numbers.

Ah, that makes sense. I was looking at it from the view of the standard definition of percentile, so elsa being on the 48th percentile looked wrong. Seeing all the LOLs made me question if I was being stupid hahaha. Thanks for the explanation.

DNA3000

Sachhyam257 said:

DNA3000 said:

Sachhyam257 said:

MCOCHazza said:

I think it’s just skewed because shocker’s base crit rate is like double elsa’s, so as a percentile it looks a bit funky because of the spread

Doesn't 48th percentile mean 52% of entries scored higher? How can that be possible when there is only one entry (shocker) that is higher? Is that not what percentiles do?

Generally, yes. However, what Auntm.ai appears to be doing with those percentile bubbles is specify what percentile of the extant range of the stat the champion falls in, not where in the total champion population the champion falls in.

For example, consider 6* R5 Black Widow CV. Her Armor Rating is 563, and it shows as 37 percentile of that stat. The extant range is listed as -428 to 2250. Every 6* rank 5 champion (presumably) falls within that range. BWCV thus falls at the 37% mark relative to that range. The total size of the range is 428 + 2250 = 2678. BWCV is 428+563=991 points higher than the lowest point. 991/2678=0.37. So BWCV is in the bottom 37% of the total range of base armor stats.

This is not a standard use of the term “percentile.” I’m actually not sure, off the top of my head, if there is a standard term for what Auntm.ai is calculating. I would call that “percent of extant range” but then nobody would know what I was talking about. Knowing Slug, I suspect he knows this as well, and chose a term that people would be familiar with over precise but opaque (and very verbose) language as a compromise.

Also, I had no idea people actually looked at those percentile numbers.

Ah, that makes sense. I was looking at it from the view of the standard definition of percentile, so elsa being on the 48th percentile looked wrong. Seeing all the LOLs made me question if I was being stupid hahaha. Thanks for the explanation.

It is entirely possible that many people reading this thread aren’t aware of what is being discussed, because I don’t think it is a common thing that all Auntm.ai users are aware of.

Sachhyam257

DNA3000 said:

Sachhyam257 said:

DNA3000 said:

Sachhyam257 said:

MCOCHazza said:

I think it’s just skewed because shocker’s base crit rate is like double elsa’s, so as a percentile it looks a bit funky because of the spread

Doesn't 48th percentile mean 52% of entries scored higher? How can that be possible when there is only one entry (shocker) that is higher? Is that not what percentiles do?

Generally, yes. However, what Auntm.ai appears to be doing with those percentile bubbles is specify what percentile of the extant range of the stat the champion falls in, not where in the total champion population the champion falls in.

For example, consider 6* R5 Black Widow CV. Her Armor Rating is 563, and it shows as 37 percentile of that stat. The extant range is listed as -428 to 2250. Every 6* rank 5 champion (presumably) falls within that range. BWCV thus falls at the 37% mark relative to that range. The total size of the range is 428 + 2250 = 2678. BWCV is 428+563=991 points higher than the lowest point. 991/2678=0.37. So BWCV is in the bottom 37% of the total range of base armor stats.

This is not a standard use of the term “percentile.” I’m actually not sure, off the top of my head, if there is a standard term for what Auntm.ai is calculating. I would call that “percent of extant range” but then nobody would know what I was talking about. Knowing Slug, I suspect he knows this as well, and chose a term that people would be familiar with over precise but opaque (and very verbose) language as a compromise.

Also, I had no idea people actually looked at those percentile numbers.

Ah, that makes sense. I was looking at it from the view of the standard definition of percentile, so elsa being on the 48th percentile looked wrong. Seeing all the LOLs made me question if I was being stupid hahaha. Thanks for the explanation.

It is entirely possible that many people reading this thread aren’t aware of what is being discussed, because I don’t think it is a common thing that all Auntm.ai users are aware of.

Is there an upside to using percent of extant range over population based percentile?

Shocker is an outlier where his base crit rating is way higher than any other champ, which basically skews the percentile displayed for all champs. I was previously under the assumption that the percentile data on stats used the standard definition, so anytime I checked the stats, I would check the percentile color (red/yellow/green) and judge accordingly.

However, because of shocker's wildly blown out crit rating (when compared to other champs) it makes every other champ's crit rating stat appear yellow/red. As a regular user unaware of this, it affects the way I judge characters (because in the common logical sense I viewed it with; red = bad, yellow = mediocre, and green = good).

RichardSlugberg

Sachhyam257 said:

I was just browsing Auntm.ai and I noticed something.

Hi! Love seeing the questions, always helpful to understand how something on AuntM.ai is received, I have some pretty strong blinders geared towards my perspective and the feedback helps a lot.

@DNA3000 covered this pretty well, thanks DNA! One thing that happens consistently when you start exposing game stats is the separation from practice - when you look at something and it tells you a champion is the 99th percentile and another champion is 98th, it's very easy to assume they are both close in value and close to the top. This is a perfect example where that would be misleading.

Unfortunately this compromise might make you think "48th percentile is not very good". This is counteracted by being able to go the prestige grid, enable stats and sort by crit rating (ugh, that's not intuitive at all!) and see her position relative to other champions.



I do think there is likely a better representation, something like deviation from the norm, where it's not a fixed 1-100 percentile. Hopefully the current number works for now, and this discussion has given me some food for thought for when the champion page is revisited.

o_o

RichardSlugberg said:

Sachhyam257 said:

I was just browsing Auntm.ai and I noticed something.

Hi! Love seeing the questions, always helpful to understand how something on AuntM.ai is received, I have some pretty strong blinders geared towards my perspective and the feedback helps a lot.

@DNA3000 covered this pretty well, thanks DNA! One thing that happens consistently when you start exposing game stats is the separation from practice - when you look at something and it tells you a champion is the 99th percentile and another champion is 98th, it's very easy to assume they are both close in value and close to the top. This is a perfect example where that would be misleading.

Unfortunately this compromise might make you think "48th percentile is not very good". This is counteracted by being able to go the prestige grid, enable stats and sort by crit rating (ugh, that's not intuitive at all!) and see her position relative to other champions.

I do think there is likely a better representation, something like deviation from the norm, where it's not a fixed 1-100 percentile. Hopefully the current number works for now, and this discussion has given me some food for thought for when the champion page is revisited.

Auntmai is an amazing resource... really appreciate your work!!