cricrate: new cricket ratings website

[viriya]

Always knew that Tendulkar and Bradman couldn't be separated statistically.

This is something that is somewhat deceiving because the rating value is not showing pure mean + confidence interval. There are adjustments based on longevity that has inflated Sachin's overall number. If I were to show just the mean innings rating and corresponding confidence interval Sachin's upper bound would be lower than Bradmans lower bound.

Adding confidence intervals has made me think of reducing the longevity bonus factor. To avoid "overrating" longer careers that will have naturally lower confidence interval ranges.

 

[Spark]

This is something that is somewhat deceiving because the rating value is not showing pure mean + confidence interval. There are adjustments based on longevity that has inflated Sachin's overall number. If I were to show just the mean innings rating and corresponding confidence interval Sachin's upper bound would be lower than Bradmans lower bound.

Adding confidence intervals has made me think of reducing the longevity bonus factor. To avoid "overrating" longer careers that will have naturally lower confidence interval ranges.

This was basically my concern, some of those confidence intervals for early-era batsmen struck me as implausibly large. Len Hutton having a confidence interval of 300 points from end to end after a 80-odd Test career just felt odd.

 

[viriya]

This was basically my concern, some of those confidence intervals for early-era batsmen struck me as implausibly large. Len Hutton having a confidence interval of 300 points from end to end after a 80-odd Test career just felt odd.

+- 150 is actually not high if you look at the others. This is just a function of batsmen having more variability in their scores (sometimes out for a duck, sometimes a ton). This is exacerbated with Bradman who's highs are much higher than others.

If you look at bowling careers, someone like Shaun Pollock has +- 86 - since bowlers tend to have more "consistent" performances.

 

[Bijed]

Added 95% confidence intervals to Test/ODI/T20I Batting/Bowling/All-Round career ratings as a +- range after the existing rating.
eg: cricrate | Career Ratings - Test Batting

If two players have overlapping confidence intervals this suggests that they can't clearly be separated.

If a player's lower CI bound is higher than another's higher CI bound that's a sign that there is a clear difference there.

Nice addition

On a separate note, which has probably been mentioned before tbf, and I don't want the whole debate to happen again, but I noticed that as per convention you don't show 100s and being 50s as well, but 200s show up as 200s and 100s, and 300s and 3/2/100s? It doesn't especially bother me, but just seems a little inconsistent?

Also, and I may just be missing a way to get this anyway, but would it be possible to have an option for each player, instead of just viewing the graph with innings ratings and the current rating over time, have one with normal scores (like on a cricinfo profile) but have career average at that point superimposed on it? I know the whole point is that the ratings are a more meaningful measure of a player, and I'm not arguing against that, but it'd still be a feature which I think people would appreciate.

Anyway, thanks for the continued effort you've put into this!

 

[viriya]

Really, really surprised that Smith's 109 isn't his top rated performance. Although I know the 199 came in a match where no one else did that much.

Yea this is a tough one. 199 is almost double the score and the next highest score was 47. For the 109 the bowling attack was better. The ratings difference between the two is only 15% - so it's really about how much actual runs is valued.

 

[viriya]

Nice addition

On a separate note, which has probably been mentioned before tbf, and I don't want the whole debate to happen again, but I noticed that as per convention you don't show 100s and being 50s as well, but 200s show up as 200s and 100s, and 300s and 3/2/100s? It doesn't especially bother me, but just seems a little inconsistent?

Also, and I may just be missing a way to get this anyway, but would it be possible to have an option for each player, instead of just viewing the graph with innings ratings and the current rating over time, have one with normal scores (like on a cricinfo profile) but have career average at that point superimposed on it? I know the whole point is that the ratings are a more meaningful measure of a player, and I'm not arguing against that, but it'd still be a feature which I think people would appreciate.

Anyway, thanks for the continued effort you've put into this!

Yea I followed convention with the 50-100-200 counting method - mainly to avoid confusion.

Not a bad idea to add the actual runs + average graph for batting in Tests, but it gets messy when you think about bowling and other formats where SR matters almost as much as runs. That's why I decided to just keep to the current/innings ratings which is a better representation of career progression.

 

[Spark]

+- 150 is actually not high if you look at the others. This is just a function of batsmen having more variability in their scores (sometimes out for a duck, sometimes a ton). This is exacerbated with Bradman who's highs are much higher than others.

If you look at bowling careers, someone like Shaun Pollock has +- 86 - since bowlers tend to have more "consistent" performances.

Yeah, this is reasonable. But I think as a statistical measure—i.e. if you had a "distribution" of batsmen with the same underlying quality, then I don't think you'd expect to see that sort of variation over a 80-Test career. I could be wrong, though.

 

[viriya]

Yeah, this is reasonable. But I think as a statistical measure—i.e. if you had a "distribution" of batsmen with the same underlying quality, then I don't think you'd expect to see that sort of variation over a 80-Test career. I could be wrong, though.

Think of it as a measure of how good a random innings from a player would be. He might get out for a duck or ton up or whatever so the distribution is all over the place but the more consistent performers will have a lower variation. The best performers will have a high mean and low variation.

 

[Spark]

Think of it as a measure of how good a random innings from a player would be. He might get out for a duck or ton up or whatever so the distribution is all over the place but the more consistent performers will have a lower variation. The best performers will have a high mean and low variation.

Ah, that makes more sense. I thought it was more like a MOE on the rating for a batsman's overall career.

 

[viriya]

Say you compare Sachin and Tendulkar. Both have similar means (Sachin 1133 Lara 1123), but Sachin's range is tighter (+-112) than Lara's (+-146). This is clear since Lara tended to make big scores when he was in, but Sachin used to more consistently get a good score. Who was better? It's a matter of preference.

 

[viriya]

Ah, that makes more sense. I thought it was more like a MOE on the rating for a batsman's overall career.

I mean it's kind of both since there is an additional longevity bonus that complicates matters, but when thinking about the confidence interval you can think of it as both:
- most likely performance by player
- accuracy of career ratings measure

 

[viriya]

Test update:

- Kane moves up to #2, De Kock drops to #5: cricrate | Current Ratings - Test Batting
- Herath back at #2: cricrate | Current Ratings - Test Bowling

 

[viriya]

Seems a bit fickle, to say the least.

Changed the Test Team ratings to be 50% less fickle (literally):
cricrate | Current Ratings - Test Team

 

[viriya]

Yep good call.

Also viriya a minor thing, but is there any chance you can allow an option to view more per page. If readers could select 50, 100 or all on the page would be just a little easier to navigate instead of clicking next per 15.

The rankings are starting to take shape, good to see.

This is done.

 

[viriya]

Test update:

- SL drops to #7, Ban moves to #8: cricrate | Current Ratings - Test Team
- De Kock moves to #2, Pujara to #6, Kohli drops out of top 10: cricrate | Current Ratings - Test Batting
- Jadeja moves up to #5: cricrate | Current Ratings - Test Bowling

 

[viriya]

Is there a dead rubber penalty at all? Most people would automatically discount any performance in a losing dead rubber as not worth that much. EDIT: Ah, I see it isn't. I'd find a way to add series result in to be honest, it feels weird that dead rubber performances are equally weighted with the rest.

After some painful data work this is finally done. Added a new minor factor "Series status" for Tests (will add for ODIs soon) where performances in dead rubbers have a negative adjustment and deciding games have a positive adjustment. I scraped all historical Test series and checked the series status at each Test.

Some other minor changes: increased the importance of the home/away factor in Tests and reduced the bonus given to longevity in the career ratings.

All these changes shuffled things around a lot. Some of the biggest changes:

Batting performances:
cricrate | Performance Ratings - Test Batting

- Brian Lara's 375 drops 35 places to #41
- Dean Jones's 184* drops 38 places to #49
- Kumar Sangakkara's 156* jumps 15 places to #2
- Mark Burgess's 119* jumps 39 places to #26
- Brian Lara's 226* drops 13 places to #16
- Mark Butcher's 173* drops 10 places to #12
- Faoud Bacchus's 250* jumps 41 places to #34
- Lance Klusener's 118* jumps 34 places to #25
- Younis Khan's 175* drops 19 places to #27
- Hashim Amla's 139* jumps 46 places to #47
- Brian Lara's 221* drops 21 places to #31
- Gundappa Viswanath's 114* jumps 41 places to #46
- Rahul Dravid's 270* jumps 12 places to #4
- Clem Hill's 188* jumps 6 places to #1

Bowling performances:
cricrate | Performance Ratings - Test Bowling

- Max Walker's 6/15 drops 41 places to #75
- Michael Clarke's 6/9 drops 37 places to #77
- Hugh Trumble's 7/28 drops 26 places to #50
- Morne Morkel's 6/23 jumps 34 places to #45

Batting career:
cricrate | Career Ratings - Test Batting

- Jack Hobbs drops 12 places to #26
- Ross Taylor jumps 22 places to #39
- Azhar Ali jumps 22 places to #41
- Herbie Taylor drops 24 places to #72

Bowling career:
cricrate | Career Ratings - Test Bowling

- Ravichandran Ashwin drops 5 places to #12
- Dale Steyn jumps 2 places to #3
- Anil Kumble jumps 5 places to #8
- Curtly Ambrose jumps 4 places to #7
- Saeed Ajmal jumps 23 places to #41
- Bill O'Reilly drops 4 places to #13
- Clarrie Grimmett drops 7 places to #23

Would love any feedback!

 

[Spark]

Great stuff, some really interesting results.

 

[viriya]

Great stuff, some really interesting results.

Thanks for the suggestion!

Just based on eye check: Butcher's 173* drops, Younis's 175* drops (what inspired me to work on this), Clarke's 6/9 drops - all good signs. Had to do a lot of work to try to make sure the adjustment wasn't too significant but also non-trivial.

 

[Spark]

Not a related question and I probably could find out on the website itself, but how does a inning's rating scale with size? Is it linear, or is there a decay?

 

[viriya]

Not a related question and I probably could find out on the website itself, but how does a inning's rating scale with size? Is it linear, or is there a decay?

The "Runs scored" factor increases linearly till 200, then decays quickly.

From methodology page (cricrate | Methodology - Test Batting):
"Runs scored: This is the most significant factor. It values runs linearly upto 200, then each additional run becomes less valuable than the last. This is because scores in excess of 200 are pursued mainly as personal milestones and does not always line up with the team objective."