Chess.com cannot do arithmetic

I don't think that rank and percentile have any meaning anymore.  About a month ago, I noticed that my percentile dropped dramatically.  Now I find that if my rating is in the 1500s, my rank is 64K and my percentile is 85.  This is true whether my rating is 1500 or 1599.  When my rating drops below 1500, my rank is 103K and my percentile is 76.   Again, it doesn't make any difference whether my rating is 1400 or 1499.

Before that, the rank and percentile were calculated once a day based on my highest rating for the day.  As I remember, the calculation was 1 - rank/active members. 

While I don't have access to the calculations the normal formula for calculating percentiles in a distribution curve is given in the link I provided. It isn't something we can calculate with the data members have access to.

Martin

The concept and calculation of percentile is rather simple:  the number of people whose rating is below mine divided by the total number of people in the pool.

http://www.dummies.com/education/math/statistics/how-to-calculate-percentiles-in-statistics/

That's why some of us are having trouble understanding the Chess.com calculation.

 I'll look at that in more detail later but I think you are looking at a different calculation. You want percentile rank which is base on a different calculation and incorporates the frequency of the rating in the calculation.

Oops, sorry, I looked at woton's link. Now I see the link you posted.

 OK, I think that I found the answer.  There's a concept called Group Frequency Distributions and the formula is somewhat complex.  If you're a statistics junkie, here's the link:

http://harding.edu/sbreezeel/460%20files/statbook/chapter5.pdf

Martin

Forget about percentile vs percentile rank Do the following two data points make sense to you?

Rating  1499  Rank  103K  Percentile  .76

Rating   1506  Rank  64K   Percentile   .85

It's hard for me to believe that a seven point rating increase would move me up 39,000 places on the list.

Yes, it does make sense if you are thinking about splitting up the distribution graph into percentile bands. My statistics experience is minimal but it appears to be a fairly standard practice to split up a distribution graph in a number of distinct bands and lump a set of statistics into those bands for comparison.

So, a range of ratings can fall in the same percentile rank band.

Just look at the image in the wiki article.

I don't know exactly how they are doing the Rank value but it is very likely not calculated live and may be once a day or even less, since the number listed may or may not actually match one's place in the list.