Rating inflation?

There is severe rating inflation on this website...

here is a list of 30 people that i played back in april/may (i omitted the names for privacy concerns), the first column is their rating back then, and the second column is their current rating (so their last login is after Nov. 1), as you can see, the average difference is a whopping 63 points!

I noticed that my own rating increased from around 1950ish to 2100ish, but i don't feel that my chess understanding has improved by that much, maybe by a tiny bit, but not by 100+ points.

oh, by the way, i play 5/3 and 2/12 games.

April/May November/December Delta Average ∆
1955 2117 162 63.63333333
1952 2057 105
1891 1922 31
1848 1823 -25
1906 2003 97
1928 2005 77
1880 1880 0
1989 1865 -124
1918 1955 37
1867 2002 135
2037 2029 -8
1892 1977 85
1913 2150 237
1906 1947 41
1908 1979 71
2106 2118 12
1953 2107 154
1902 1973 71
1900 2001 101
1855 1840 -15
1968 2033 65
1893 2006 113
1882 2109 227
1909 1925 16
1974 1959 -15
1920 2000 80
1822 1881 59
2014 2062 48
1921 1958 37
2000 2035 35

30 members are still all around after so long a time? curious

anyway, established members will tend to have an increase in their rating (as they get used to the site, perhaps they even play a little better)

your sample size of 30 is tiny. the site claims a million members. so, that's 30 / 1,000,000

come back when you've compared a 1,000 members' ratings and the result would be more meaningful imo

besides, the sample is tainted and not a random sampling -- they've all played against you (maybe, something's rubbing off from your games with them, perhaps you're teaching them a new trick or two)

30 members are still all around after so long a time? curious

anyway, established members will tend to have an increase in their rating (as they get used to the site, perhaps they even play a little better)

your sample size of 30 is tiny. the site claims a million members. so, that's 30 / 1,000,000

come back when you've compared a 1,000 members' ratings and the result would be more meaningful imo

besides, the sample is tainted and not a random sampling -- they've all played against you (maybe, something's rubbing off from your games with them, perhaps you're teaching them a new trick or two)

well, some of my opponents became inactive, so i only picked the ones that are still active.

for people rated 1850+, i would think that they would have gotten used to the site before that.

a little better, sure, but 63 points on average is a bit too much to happen without rating inflation taken place.

sure 30 is a small sample size, but another observation also supports my claim:

i was 1950ish back then, i would go into Cube Cafe and count the number of people that are rated above me (i can only see 100 players), i did it many many times, the average was around 25.

now i am 2100, i do the same thing, the average is still around 25.

so in other words, even tho my rating increased by 100+, my percentile ranking remains roughly the same, which coincides with the fact that i don't feel that i, currently at 2150, play significant better than when i was 1950ish.

tried to edit the original post but couldn't do it.

i increased the sample size to 53. the new delta is 49, which is still quite large and suggests rating inflation to some extent.

49.13207547

1955 2117 162
1952 2057 105
1891 1922 31
1848 1823 -25
1906 2003 97
1928 2005 77
1880 1880 0
1989 1865 -124
1918 1955 37
1867 2002 135
2037 2029 -8
1892 1977 85
1913 2150 237
1906 1947 41
1908 1979 71
2106 2118 12
1953 2107 154
1902 1973 71
1900 2001 101
1855 1840 -15
1968 2033 65
1893 2006 113
1882 2109 227
1909 1925 16
1974 1959 -15
1920 2000 80
1822 1881 59
2014 2062 48
1921 1958 37
2000 2035 35
1887 1966 79
1967 1980 13
1909 1928 19
1907 2062 155
1869 1853 -16
1922 1928 6
1925 1938 13
1914 1828 -86
1901 1844 -57
1838 1971 133
1839 1847 8
1884 1808 -76
1906 2006 100
1829 2065 236
1866 1791 -75
1946 1964 18
1871 1877 6
1824 1773 -51
1930 1952 22
1901 1970 69
1886 1834 -52
1836 1852 16
1812 2027 215

a 50% increase in sample size dropped the delta from 63 to 49

sounds like the beginning of a trend, so I predict the larger the sample size, the closer to 0 will be the result

(really, for 1,000,000 members, a 1,000 or at least a 200 sized sample would seem minimal)

a 50% increase in sample size dropped the delta from 63 to 49

sounds like the beginning of a trend, so I predict the larger the sample size, the closer to 0 will be the result

(really, for 1,000,000 members, a 1,000 or at least a 200 sized sample would seem minimal)

first of all, i am only interested in players whose rating is over 1800, so the total is way smaller than one million.

also keep in mind that i am only keeping track of people that i played back in april/may whose last log in is after Nov 1st, so my sample size is limited.

in fact, i just went through all of them, 71 players. average rating change is 51. with a sample size that's reasonably big, i believe there is indeed rating inflation to some extent, which agrees with how i feel about my own rating increase.

Average ∆
51.70422535

April/May November/December Delta
1955 2117 162
1952 2057 105
1891 1922 31
1848 1823 -25
1906 2003 97
1928 2005 77
1880 1880 0
1989 1865 -124
1918 1955 37
1867 2002 135
2037 2029 -8
1892 1977 85
1913 2150 237
1906 1947 41
1908 1979 71
2106 2118 12
1953 2107 154
1902 1973 71
1900 2001 101
1855 1840 -15
1968 2033 65
1893 2006 113
1882 2109 227
1909 1925 16
1974 1959 -15
1920 2000 80
1822 1881 59
2014 2062 48
1921 1958 37
2000 2035 35
1887 1966 79
1967 1980 13
1909 1928 19
1907 2062 155
1869 1853 -16
1922 1928 6
1925 1938 13
1914 1828 -86
1901 1844 -57
1838 1971 133
1839 1847 8
1884 1808 -76
1906 2006 100
1829 2065 236
1866 1791 -75
1946 1964 18
1871 1877 6
1824 1773 -51
1930 1952 22
1901 1970 69
1886 1834 -52
1836 1852 16
1812 2027 215
1871 1897 26
1855 1880 25
1802 2052 250
1803 1779 -24
1849 1941 92
1855 1887 32
1865 1900 35
1830 1963 133
1840 1873 33
1856 1871 15
1800 1838 38
1919 2017 98
1882 2122 240
1940 2007 67
1911 1823 -88
1870 1870 0
1988 2099 111
1816 1800 -16

Did you consider the possibility that people whose ratings dipped a lot, drop out?

that is actually a very good point: players whose ratings dip tend to quit. but chesscube wants to keep their players happy. players whose ratings improve are happy, including myself.

therefore, i have a strong feeling that this inflation is done on purpose to some extent to keep players.

would be nice if a chesscube employee posted some relevant data, for example, average rating, percentage of players above 1800/1900/2000/2100/2200/... over the past 6 months, etc. to confirm/deny my hypothesis.

looking at the view count of this thread, i have a feeling that many were searching for "chesscube inflation" and were directed here.