Tema za raznorazne RB statistike ExYu usera
Evo najjednostavnije - preko 1000 rejtova
Znači, stižeš me za godinu dana?
Ovim tempom tek za dvije. Ali gotovo sam siguran da ću ja usporiti.
OK, time for some stats crunch. I’ll write in English since some other users may be interested in this.
For some time now I’ve had an ongoing discussion with Marko about rating criteria he is using. There was a thread about it here, someone created MRI (Miserable raters index) which puts in perspective number of 3.8/4+ rates and total number of rates. However Marko made a valid point that MRI doesn’t take in account avg rate of beers you’ve tried. So I’ve decided to put that into perspective. I chose top exyu raters and some foreign Ratebeerians so the thing goes beyond Marko. Ideally it would be nice to see data for more users, but since that requires significant amount of time which I don’t have.
Basically each range of scores is compared to total beers and to the number of avg 3.8+ or 4+ beers tried. Color scales are horizontal, comparing the scores with other users. I wanted to see if ogivlado and Marko adjusted over the years so made a special columns for their rates in last three years.
So my premises are:
- most raters have a certain upper limit score after which they are reluctant to give ratings, as a result some users are not using good portion of the scoring scale
- I tend to give really bad rates to bad beers and really good to good ones
- Marko does just the opposite - good beers and bad beers are scored closer to the average which results in his avg beer rated and avg score given to be almost the same although he underscores good beers
Comments are welcome.
Disclamer: data was extracted manually and errors are possible. If you see one let me know.
I have to point out that beer total averages I was able to acquire were weighted averages and not mean. That means if someone rated more good beers with small amount of rates his pct avg 3.8+ and pct avg 4+ will be higher.
OK let me see if I can make some points out of this data. I’ll comment mostly relative (coloured) stats:
- Marko seems to have upper scoring limit relatively low, but higher than I anticipated - it’s around 4.2 rather than 4
- also he seems to be rating lower in last 3 years for 4.2-4.5+ rates. I find it very interesting that his 3.8+ rates decreased from 186% to 97%, but in the same time 4+ increased from 183% to 215%. I wonder if that had something to do with me bugging him about 4+ rates
- ogivlado in total rates good beers lower than me, but for the last three years his good rates dropped enormously, even lower than Marko’s. He might even challenge @FatPhil for the most miserable rater title
- makroijelena rates good beers bit lower than others and his upper limit is somewhere around 4.3+ with a great drop on 4.5+
- heavy rates similar to ogivlado, wonder if his number dropped over the years
- both @MarkoNm1 and Kita rate very high, curiously their avg rated and avg given difference is quite low?
- klm1 drank most good beers but rates quite low. Maybe your criteria gets different when you drink only good beers. Also I wonder if his numbers could be interpreted like opposite of Kita’s who scores highly
@kraddel has lower number of scores in the 3.8+ and 4+ range, but his 4.3+ and 4.5+ are among higher ones
@Travlr seems to have lots of 3.8+ and 4+ scores, 4.5 seems to be upper limit for him
@FatPhil scores tho lowest, I guess as expected if you look at the difference between avg given and rated
@jtclockwork has similar numbers as me, just a bit lower one the higher scores
@cgarvieuk starts similar as @jtclockwork but starts dropping after 4.3+ and dives after 4.5+
I’m still not sure if this is the best way to show the data I was looking for but I’m satisfied with this little survey. It would be interesting to see this kind of data for whole countries to put some number behind theories there are miserable or hyped countries when it comes to beer rating.
For reference, @FatPhil views 2.75 as the middle of a bell-curve that can range from 0.5 to 5.0. This puts me at odds with many other raters on the site. I’m not stingy - I’m just a mathematician!
I’m super-surprised to be near 100% (slightly below at ~90%) for the 3.8 and 4.0 cut-offs. I think that proves that I am not too much of a low rater. However, that “OMG wow factor” area above 4.2 is clearly one that I do shy away from, for good or for ill. Very few beers really wow me.
Mega-cool stat - thanks for posting it (and in particular for including me)!
Do you have the standard deviation for each rater too? If you have the counts of each score, which I presume you do in order to generate some of the rows, then you can work out the SD quite easily.
If so, then how about +1-sigma, +2-sigma, and +3-sigma ratios? Of course, that fails to take skew into account, but I’m not sure how you could handle that. (Hmmm, having said that, perhaps there is a hackish way - where do you get your raw data from, perhaps I could have a play with some ideas… ?)
Well my average beer would be around 3.4. So that’s where the difference in avg given is.
Unfortunately I don’t have lots of data. I went to each users profile > sorted by Score or Avg and found the last beer rated 4 or 3.8. To get last three years from Marko and Ogivlado I had to count the rates manually. So I don’t think calculating standard deviation for each user would be possible (if I am understanding what you’re suggesting correctly).
Looks like I can calculate the stats and standard deviations quite easily.
I’m writing a little script…
Here are my stats, not accounting for skew.
n=17259 Sx=49802.3 Sxx=150522.01
mean=2.89 variance=0.395 PopSD=0.628
+1-sigma=3.51 +2-sigma=4.14 +3-sigma=4.77
>1-sigma=2051 >2-sigma=31 >3-sigma=0
>1-sigma=11.88% >2-sigma=0.17% >3-sigma=0%
I think that the skew is such an important issue, I’ll try to fix that before running it for anyone else. No time for that tonight, maybe tomorrow.
This is probably terrible maths, but good does-the-job hand waving. I find the mode, slice the bell curve in two, throw away the bottom half, and flip the top half down in a mirror fashion - tada! instant lack of skew:
Approximate mode at 3.2,1387 (4155 nearby)
Assuming high skew, only looking at the top half of the bell curve:
Restricting only to half of the bell-curve, with 6326 beers
n=6326 mean=3.2 Var=0.117 popSD=0.342
+1-sigma=3.542 +2-sigma=3.885 +3-sigma=4.227
>1-sigma=2051 >2-sigma=321 >3-sigma=8
(naive percentages don’t make much sense, as I should still be dividing by 17k, not 6k, and my code doesn’t take that into account yet)
And this is the equivalent for Iznogud:
n=4597 mean=3.4 Var=0.493 popSD=0.702
+1-sigma=4.080 +2-sigma=4.782 +3-sigma=5.484
>1-sigma=594 >2-sigma=8 >3-sigma=0
>1-sigma=12.920% >2-sigma=0.170% >3-sigma=0.000%
Approximate mode at 3.7,398 (1147 nearby)
Assuming high skew, only looking at the top half of the bell curve:
Restricting only to half of the bell-curve, with 1645 beers
n=1645 mean=3.7 Var=0.140 popSD=0.374
+1-sigma=4.074 +2-sigma=4.448 +3-sigma=4.821
>1-sigma=594 >2-sigma=71 >3-sigma=5
Doing other people is trivial, all I need is userids.
I understand the calculation, but not sure how to interpret the numbers.
Yeah, I think the two most interesting stats from the above are
- How skewed you are, what ratio of your ratings are below/above your mode:
- Me: 6326/17259 = 36.65% above mode
- You: 1645/4597 = 35.78% above mode
- How much you spread your good beers out, so the fake StdDev value:
- Me: with the mode at 3.2 each SD is 0.342, so good, damn good, and exceptional are at 3.542, 3.885, 4.227
- You: with the mode at 3.7, each SD is 0.374, so good, damn good, and exceptional are at 4.074, 4.448, 4.821
The actual number of ratings you give at or above these levels is kinda meaningless, as they are derived from the number of beers you give at those levels so they would be self-balancing, we should all be similar. It would I guess be a measure of kurtosis, but pffft, who cares about kurtosis?
It’s basically intended to be a way to help interpret different raters’ scores. I’m a [3.2, 3.542, 3.885, 4.227] kinda guy, you’re a [3.7, 4.074, 4.448, 4.821] kinda guy.
Or, in simple terms, if a brewer - and I’m not going to mention any names, this is purely hypothetical, but he may be in Belgium, but maybe he’s not - sees me giving one of his hypothetical beers a 4.0, and you giving it 4.6, he should be equally flattered, and shouldn’t send me hate mail claiming that I’m the sperm of satan. (That’s not exactly what he may have hypothetically called me, it’s merely a tribute to Paula Yates.)