Weightings for average, strike-rate, wickets per innings et al

[Days of Grace]

I am currently working on the next edition of my top 100 test batsmen and bowlers in addition to my soon-to-be released top 50 ODI batsmen and bowlers.

The final sticking point I have is how much weightage to assign to the career metrics for batsmen and bowlers. Here are the current weightings:

Test batsmen:
Average: x4
Runs per innings: x2
Strike-rate: x1

Test bowlers:
Average: x4
Wickets per innings: x2
Strike-rate: x1

ODI batsmen:
Average: x2
Runs per innings: x1
Strike-rate: x1

ODI bowlers:
Average: x2
Wickets per innings: x1
E/R: x1

Obviously strike-rate and economy-rate are more important for ODI batsmen and bowlers when compared the strike-rates of test batsmen and bowlers. Am I giving too much weight to strike-rate for test players? Should runs and wickets per innings count for as much as 50% of the average?

If anyone has any opinions about this, let me know. You will have made a contribution to the final rankings.

 

[TheJediBrah]

wicket per innings should be 0, if you're trying to measure quality

runs per innings too for that matter. Average already tells you everything you need to know in that regard

 

[Days of Grace]

wicket per innings should be 0, if you're trying to measure quality

runs per innings too for that matter. Average already tells you everything you need to know in that regard

I think you need to include wickets per innings since it's a measure of productivity and you are penalizing spin bowlers if you don't include it.

 

[Line and Length]

Wickets per innings favours bowlers who 'carry' an attack (eg Hadlee or Murali) and is harsh on bowlers who are part of a strong, balanced attack such as the Windies Pace attack at their peak.

 

[TheJediBrah]

I think you need to include wickets per innings since it's a measure of productivity and you are penalizing spin bowlers if you don't include it.

You're not penalising spin bowlers if you don't include it, you're just not rewarding them excessively.

re. productivity that's definitely a valid point but it circles back to analysing what the purpose of the exercise is. Are you trying to judge quality of the players, or productivity, because they're not exactly the same.

 

[Days of Grace]

You're not penalising spin bowlers if you don't include it, you're just not rewarding them excessively.

re. productivity that's definitely a valid point but it circles back to analysing what the purpose of the exercise is. Are you trying to judge quality of the players, or productivity, because they're not exactly the same.

I'm analyzing both. Quality, productivity and career length.

 

[Days of Grace]

Wickets per innings favours bowlers who 'carry' an attack (eg Hadlee or Murali) and is harsh on bowlers who are part of a strong, balanced attack such as the Windies Pace attack at their peak.

This is a very valid point. I have considered adjusting wickets per innings but it's too complicated. I will consider giving less weight to wickets per innings and runs per innings, though.

 

[TheJediBrah]

I'm analyzing both. Quality, productivity and career length.

Taking into account RPI and WPM gives you a less accurate measurement of quality, which is the only reason I would disagree with it personally

Average by itself is a better measurement than any combination of average + WPM or average + RPI

So the way I see it you're putting in more steps and more work to get a less accurate result

 

[srbhkshk]

Create a list of around 30 players who you are confident ranking subjectively and run a regression to get the weights.

 

[Days of Grace]

Create a list of around 30 players who you are confident ranking subjectively and run a regression to get the weights.

Could you explain this more or give an example?

 

[srbhkshk]

Could you explain this more or give an example?

To be very honest it was a semi-joke, but if you actually go down this path it's critical that you be extremely confident in how you rank those 30 players subjectively. You can probably reduce it to 25ish if you don't want to do 30 but anything below that is probably too low.

After that you create this data -

Name - Value - Avg - SR - RPI

(example) -
Bradman - 100 - Avg - SR - RPI
Sachin - 50 - Avg - SR - RPI
Some batsman definitely worse than Sachin - 47 - Avg - SR - RPI

etc. etc. The numbers are subjective, i.e. you decide the numerical value for Sachin (and the other 28 guys) if Bradman is fixed to 100. This value then becomes your dependent variable and Avg , SR and RPI are your independent variables. Then you run a simple linear regression (or a ridge penalty regression) with the equation -

Value ~ Avg + SR + RPI and whatever coefficients the model throws at you become your weights. If you don't want to assign values then just use rank, so Bradman - 30 , Sachin - 29 , Other guy - 28 and so on... , just need to ensure that the number is higher for who you decide is the better Batsman.


In case you don't know how to run a linear regression, here's a link that tells you how to do it in excel.

https://www.rwu.edu/sites/default/f...displaying_regression_statistics_in_excel.pdf

 

[RossTaylorsBox]

You've also got to make sure your 30 guys are the right mix of ATGs, good players and scrubs or you'll get some weird results. Plus Bradman as a massive outlier.

 

[Coronis]

I am currently working on the next edition of my top 100 test batsmen and bowlers in addition to my soon-to-be released top 50 ODI batsmen and bowlers.

The final sticking point I have is how much weightage to assign to the career metrics for batsmen and bowlers. Here are the current weightings:

Test batsmen:
Average: x4
Runs per innings: x2
Strike-rate: x1

Test bowlers:
Average: x4
Wickets per innings: x2
Strike-rate: x1

ODI batsmen:
Average: x2
Runs per innings: x1
Strike-rate: x1

ODI bowlers:
Average: x2
Wickets per innings: x1
E/R: x1

Obviously strike-rate and economy-rate are more important for ODI batsmen and bowlers when compared the strike-rates of test batsmen and bowlers. Am I giving too much weight to strike-rate for test players? Should runs and wickets per innings count for as much as 50% of the average?

If anyone has any opinions about this, let me know. You will have made a contribution to the final rankings.

Objectively SR is very hard to define as a rating metric in both tests and ODI’s. Obviously it is more important in ODI’s.

In both formats the game has changed markedly and overall batting SR has risen whilst bowling SR has fallen - due to the changing nature of the game its not as important as other factors to me.

For example, Bill O’Reilly (considered by many to be the 3rd greatest spinner) has a SR of 69.3 compared to Ashwin (generally considered inferior) with a SR of 51.3. Is it fair to penalise him for this?

 

[shortpitched713]

ODI batsmen:
Average: x2
Runs per innings: x1
Strike-rate: x1

ODI bowlers:
Average: x2
Wickets per innings: x1
E/R: x1

Obviously strike-rate and economy-rate are more important for ODI batsmen and bowlers when compared the strike-rates of test batsmen and bowlers. Am I giving too much weight to strike-rate for test players? Should runs and wickets per innings count for as much as 50% of the average?

If anyone has any opinions about this, let me know. You will have made a contribution to the final rankings.

Average and strike rate, as well as bowling average and ER should be equal for ODIs, IMO. It breaks from the conventional wisdom, but I believe it is much more accurate for players who are having impact in ODIs (which I generally find are played slightly too conservatively to this day).

 

[shortpitched713]

For Tests:

Batting strike rate should have a very low positive counting, due to context specificity.

Bowling strike rate is even worse in how context specific it is. Lower is always better for your "ace" strike bowler, for mine, but then we get into the how to rate the 2nd or 3rd bowler, or spinner, as some amount of their role will involve being a "stock" bowler, where actually a lower ER and thus higher SR are preferred. I personally, would omit it altogether for bowlers.

 

[Days of Grace]

To be very honest it was a semi-joke, but if you actually go down this path it's critical that you be extremely confident in how you rank those 30 players subjectively. You can probably reduce it to 25ish if you don't want to do 30 but anything below that is probably too low.

After that you create this data -

Name - Value - Avg - SR - RPI

(example) -
Bradman - 100 - Avg - SR - RPI
Sachin - 50 - Avg - SR - RPI
Some batsman definitely worse than Sachin - 47 - Avg - SR - RPI

etc. etc. The numbers are subjective, i.e. you decide the numerical value for Sachin (and the other 28 guys) if Bradman is fixed to 100. This value then becomes your dependent variable and Avg , SR and RPI are your independent variables. Then you run a simple linear regression (or a ridge penalty regression) with the equation -

Value ~ Avg + SR + RPI and whatever coefficients the model throws at you become your weights. If you don't want to assign values then just use rank, so Bradman - 30 , Sachin - 29 , Other guy - 28 and so on... , just need to ensure that the number is higher for who you decide is the better Batsman.


In case you don't know how to run a linear regression, here's a link that tells you how to do it in excel.

https://www.rwu.edu/sites/default/f...displaying_regression_statistics_in_excel.pdf

Thank you for explaining. That is a little too complicated for me at this stage. What I am currently doing is giving each batsmen or bowler a score out of 800 for their overall quality. To do this, for test batsmen, at first, I get the top 10 ranked averages, runs per innings and strike-rates for all batsmen who have scored 1000 career runs (my personal cut-off). I average out the top 10.

For example, the average of the top 10 batting averages is 60, the average top 10 runs per innings is 55 and the average top 10 strike-rate is 75. From that point, any player is judged by how close they come to achieving those career figures. In essence, how close are they to "perfection".

A theoretical player who averaged 60, scored 55 runs per innings and scored their runs at 75 runs per ball would get a career rating of exactly 800.

A player who averaged 30, scored 37.5 runs per innings and had a strike-rate of 37.5 would get a career rating of exactly 400.

Now, Bradman is an outlier in this respect, because he has adjusted career figures of 90.10, 78.84 and 63.42. He scores 1112 points out of 800. This is calculated by the following:

Average: 90.10/60 = 1.50
Runs per innings: 78.84/55 = 1.43
Strike-rate: 63.42 = 0.85

((1.50*4)+(1.43*2)+(0.85*1))/7 = 1.39

1.39 * 800 = 1112


I could max him out at 800 but this wouldn't show the chasm between him and the next best player.

 

[Days of Grace]

Objectively SR is very hard to define as a rating metric in both tests and ODI’s. Obviously it is more important in ODI’s.

In both formats the game has changed markedly and overall batting SR has risen whilst bowling SR has fallen - due to the changing nature of the game its not as important as other factors to me.

For example, Bill O’Reilly (considered by many to be the 3rd greatest spinner) has a SR of 69.3 compared to Ashwin (generally considered inferior) with a SR of 51.3. Is it fair to penalise him for this?

I adjust strike-rates by era and opposition.

 

[Coronis]

I adjust strike-rates by era and opposition.

Mm I still don’t find it an inherently useful measure of skill personally i.e I don’t subscribe to higher SR = better. There’s just too many variables in an innings and too many situations where longer slower innings are as/more useful for me to consider it a big factor.

 

[TheJediBrah]

Mm I still don’t find it an inherently useful measure of skill personally i.e I don’t subscribe to higher SR = better. There’s just too many variables in an innings and too many situations where longer slower innings are as/more useful for me to consider it a big factor.

There's diminishing returns on Strike-rate in Tests. Over the course of a career though, higher is almost always better. You win more games overall, both directly and indirectly because of faster scoring and more aggressive batting in general. Even though defensive batting and focus on survival can be valuable in many situations.

 

[Days of Grace]

I realize that one drawback in my system is that an ODI bowler with an E/R of 6 is equivalent to averaging 30 runs per wicket. Both are 1.5 times the figures which give a bowler maximum quality points (4.00 E/R and 20 runs per wicket). In no way should this be the case. So instead of using multiplication and division, perhaps a linear scale would be better, i.e. maximum points for an E/R of 4.00 and zero points for an E/R of 6.00. How can I get excel to calculate this for me?