Print

Please post once and this message will disappear! Introduce yourself, say hello, jump into a discussion...

[RicePlateReddy]

I cranked through some numbers last night to gain some insight into the consistency of the gang of four. The results contradicted my 'intuition'  .

For each player I computed 5 metrics:

Low Productivity Percentage: The % of innings in which the batsman scored less than 20 runs (not outs included)
Median: The median score of the batsman
Regular Avg: The well known batting average (not outs not counted in division)
Modified Avg: Average score made (not outs counted in division)
Median to Modified Avg Difference: The difference between the median score and the modified average. It makes sense to use the modified average more than the (higher) regular average when calculating this simple, linear variance. In any case, I've shown the difference with respect to the regular average in parenthesis following the figure.


RD:
Low Productivity Pct = 77 / 214 = 35.98%
Median Score = 33
Regular Avg = 54.08
Modif Avg = 47.51
Median to Modif Avg diff % = 30.54% (38.98%)


SG:
Low Productivity Pct = 73 /  178 = 41.01%
Median Score = 27
Regular Avg = 41.92
Modif Avg = 38.40
Median to Modif Avg Diff % = 29.69% (35.59%)


SRT:
Low Productivity Pct = 103 / 242 = 42.56%
Median Score = 31
Regular Avg = 54.64
Modif Avg = 49.00
Median to Modif Avg Diff % = 36.73% (43.27%)


VVS:
Low Productivity Pct = 69 /156 = 44.23%
Median Score = 24
Regular Avg = 43.48
Modif Avg = 37.91
Median to Modif Avg Diff % = 36.69% (44.80%)

Conclusions:

Interestingly, in low productivity percentages, the ranking (best - to - worst) is:
RD, SG, SRT, VVS

In terms of median score variance from the not-out adjusted average, the ranking  (best - to - worst) is:
SG, RD, VVS, SRT

In terms of median score the ranking  (best - to - worst) is:
RD, SRT, SG, VVS

I don't believe that a better treatment of the not outs would yield any appreciable difference, given the large number of innings played. I should also add that VVS probably suffers a small bit from the opening experiment.

[kban1]

Applause for a fresh look at the numbers, especially with relation to consistency issues.





But I have a question -- given that one of the issues we have spent a lot of time debating on the DG is the consistency of these 4 as a group, should we use some other players -- lets say an SMG, Ponting, Lara, Kallis, Richards to see how our maharathis fare compared to the others --would be  a good benchmark comparison, I think

[RicePlateReddy]

But I have a question -- given that one of the issues we have spent a lot of time debating on the DG is the consistency of these 4 as a group, should we use some other players -- lets say an SMG, Ponting, Lara, Kallis, Richards to see how our maharathis fare compared to the others --would be  a good benchmark comparison, I think

Agreed, but requires more time. TBD.

I was more interested in generating the 'Natty curves' later tonight because it will be relatively straightforward with what I already have.
http://www.cricketvoice.com/cricketforum2/index.php/topic,13753.msg170894.html#msg170894

[ramshorns]

I cranked through some numbers last night to gain some insight into the consistency of the gang of four. The results contradicted my 'intuition'  .

For each player I computed 5 metrics:

Low Productivity Percentage: The % of innings in which the batsman scored less than 20 runs (not outs included)
Median: The median score of the batsman
Regular Avg: The well known batting average (not outs not counted in division)
Modified Avg: Average score made (not outs counted in division)
Median to Modified Avg Difference: The difference between the median score and the modified average. It makes sense to use the modified average more than the (higher) regular average when calculating this simple, linear variance. In any case, I've shown the difference with respect to the regular average in parenthesis following the figure.


RD:
Low Productivity Pct = 77 / 214 = 35.98%
Median Score = 33
Regular Avg = 54.08
Modif Avg = 47.51
Median to Modif Avg diff % = 30.54% (38.98%)


SG:
Low Productivity Pct = 73 /  178 = 41.01%
Median Score = 27
Regular Avg = 41.92
Modif Avg = 38.40
Median to Modif Avg Diff % = 29.69% (35.59%)


SRT:
Low Productivity Pct = 103 / 242 = 42.56%
Median Score = 31
Regular Avg = 54.64
Modif Avg = 49.00
Median to Modif Avg Diff % = 36.73% (43.27%)


VVS:
Low Productivity Pct = 69 /156 = 44.23%
Median Score = 24
Regular Avg = 43.48
Modif Avg = 37.91
Median to Modif Avg Diff % = 36.69% (44.80%)

Conclusions:

Interestingly, in low productivity percentages, the ranking (best - to - worst) is:
RD, SG, SRT, VVS

In terms of median score variance from the not-out adjusted average, the ranking  (best - to - worst) is:
SG, RD, VVS, SRT

In terms of median score the ranking  (best - to - worst) is:
RD, SRT, SG, VVS

I don't believe that a better treatment of the not outs would yield any appreciable difference, given the large number of innings played.

Great effort SSL. 

To me as great as this is to give an idea of the overall consistency there need to be some considerations made for the degree of difficulty. 

1) For example Sachin's score against 240 odd not out against BD is counted as an inning but in no way shape or form reflects the uselessness of such a knock in his overall modified average.

2) Another thing is you totally discounting notouts which is not totally right either IMO.  For example VVS numerous times playing at No.6 remained not out on 60 odd runs atleast 4 or 5 times from top of my head.  So who is to say he could have not scored a 100 if he had a chance to bat up.  So some consideration needs to be given in such scenarios to make it as close to accurate as possible.

For now just skimming through it that is what I came up with. 

But if we work through this we can come up with a model as they did in Tennis by an expert panel in which after taking all things into consideration (like number of  Grandslam wins, Finals appearances, Year end Masters champion, other non-grandslam wins) Ivan Lendl came up better than Sampras, McEnroe, Borg, Connors, Agassi etc.  At the time when I last looked at it Federer was not in the picture.

I should also add that VVS probably suffers a small bit from the opening experiment.

Is there a way you can re-do VVS's numbers just for my curiosity from the end of 2000 when he became a regular as a Middle order batsman as opposed to a non-regular before that when his postion as a bastman was all over the map.

[ramshorns]

SSL:Actually VVS only played 155 innings not 156 if you refer this

http://www.howstat.com/cricket/Statistics/Batting/BattingCareerRuns.asp?Stat=5000.

I agree with it and I will tell you why. 

There was this one match where India lost a wicket but immediately there was bad light and we were close to drawing a Test.  So next player padded up was VVS.  RD just sent him to appeal for the light and comeback which he promptly did.  He did not even take gaurd or face a ball.  So it did not count as an inning but cricinfo continues to show that as an inning and it is 156 as opposed to 155.

[gouravk]

but whats the point of this analysis ?

[RicePlateReddy]

 
1) For example Sachin's score against 240 odd not out against BD is counted as an inning but in no way shape or form reflects the uselessness of such a knock in his overall modified average.

Of course. I did indicate this in the post immediately prior to natty's in the link I provided http://www.cricketvoice.com/cricketforum2/index.php/topic,13753.msg170841.html#msg170841 That is one reason why I resisted making any definitive conclusions. This is just an indicator and I always wondered what the metrics would look like and hence cranked it out. Having said that, it surprised me to see VVS lower than the others and SG higher than SRT. I would have bet against it.

2) Another thing is you totally discounting notouts which is not totally right either IMO.  For example VVS numerous times playing at No.6 remained not out on 60 odd runs atleast 4 or 5 times from top of my head.  So who is to say he could have not scored a 100 if he had a chance to bat up.  So some consideration needs to be given in such scenarios to make it as close to accurate as possible.

I get the feeling that not outs affect everyone here perhaps equally. They are difficult to handle in general. Doing the Bhogle and Rao "10 closest innings average added on with cap" may be reasonable. Again if you refer to the link -- http://www.cricketvoice.com/cricketforum2/index.php/topic,13753.msg170894.html when they did that, the not out factor ended up not contributing much. Hence I avoided paying too much attention to it. When I get some time, I may clarify with not outs considered differently to see if there is any shift. For a start, I should exclude not outs for scores less than 20 in the lower productivity metric.

[LosingNow]

In terms of median score variance from the not-out adjusted average, the ranking  (best - to - worst) is:
SG, RD, VVS, SRT

Appears to me (all things equal) the person with lowest average would have this number as the lowest because the difference between median and average has to become small as average falls.. (RD is an interesting and exceptional case..wow!).. so I am not sure if this measure of consistency is useful. Also, consistency around a low average is not necessarily good. My hypothesis is that this measure, when looked by itself, probably penalizes the century makers ... those with lot of 100s and/or with bigger centuries...because that will spread the gap between median and average.

I guess I would look at average first.. if it is high that is good.. and then look at consistency.. and if that is high too .. that is even better.

 

[kban1]

In terms of median score variance from the not-out adjusted average, the ranking  (best - to - worst) is:
SG, RD, VVS, SRT

Appears to me (all things equal) the person with lowest average would have this number as the lowest because the difference between median and average has to become small as average falls.. (RD is an interesting and exceptional case..wow!).. so I am not sure if this measure of consistency is useful. Also, consistency around a low average is not necessarily good. My hypothesis is that this measure, when looked by itself, probably penalizes the century makers ... those with lot of 100s and/or with bigger centuries...because that will spread the gap between median and average.

I guess I would look at average first.. if it is high that is good.. and then look at consistency.. and if that is high too .. that is even better.

 

 a valid proposition at first glance.

But consistency around a low average would not be valuable if the average was significantly low --remember the metric is scores 20 or below or 20 or above.

In that scenario, a person who scores 20 above the cutoff more often than not is not a bad bat to have. I think a CI stat analysis also showed how SG has an amazing frequency of scoring close to his career average of 40+ (a separate one showed how he was one of the few never to have dropped below 40).

RD's performance is of course phenomenal if only because he averages 12 more than SG, yet is second on that list in terms of the differential. 

[inoc]

In terms of median score variance from the not-out adjusted average, the ranking  (best - to - worst) is:
SG, RD, VVS, SRT

Appears to me (all things equal) the person with lowest average would have this number as the lowest because the difference between median and average has to become small as average falls.. (RD is an interesting and exceptional case..wow!).. so I am not sure if this measure of consistency is useful. Also, consistency around a low average is not necessarily good. My hypothesis is that this measure, when looked by itself, probably penalizes the century makers ... those with lot of 100s and/or with bigger centuries...because that will spread the gap between median and average.

I guess I would look at average first.. if it is high that is good.. and then look at consistency.. and if that is high too .. that is even better.

 

you are right, consistency has little meaning without the average. you can consistently score zero.

every time another way of analysis has been formulated it has pointed to simple averages as being the easiest and gold standard at evaluating performance.

[Libran]

 
1) For example Sachin's score against 240 odd not out against BD is counted as an inning but in no way shape or form reflects the uselessness of such a knock in his overall modified average.

Of course. I did indicate this in the post immediately prior to natty's in the link I provided http://www.cricketvoice.com/cricketforum2/index.php/topic,13753.msg170841.html#msg170841 That is one reason why I resisted making any definitive conclusions. This is just an indicator and I always wondered what the metrics would look like and hence cranked it out. Having said that, it surprised me to see VVS lower than the others and SG higher than SRT. I would have bet against it.

2) Another thing is you totally discounting notouts which is not totally right either IMO.  For example VVS numerous times playing at No.6 remained not out on 60 odd runs atleast 4 or 5 times from top of my head.  So who is to say he could have not scored a 100 if he had a chance to bat up.  So some consideration needs to be given in such scenarios to make it as close to accurate as possible.

I get the feeling that not outs affect everyone here perhaps equally. They are difficult to handle in general. Doing the Bhogle and Rao "10 closest innings average added on with cap" may be reasonable. Again if you refer to the link -- http://www.cricketvoice.com/cricketforum2/index.php/topic,13753.msg170894.html when they did that, the not out factor ended up not contributing much. Hence I avoided paying too much attention to it. When I get some time, I may clarify with not outs considered differently to see if there is any shift. For a start, I should exclude not outs for scores less than 20 in the lower productivity metric.

I can't understand the issue of uselessness of an innings...it is test cricket goddammit!! So nothing can be useless...Going by that logic, VVS innings in Sydney in the tour before 2004 should be one of the most useless ones..because we anyways lost the match and the series

[doubt]

good analysis

i read somewhere about the median of other players (prolly 1 yr old)

   1. Bradman 56.5
   2. Hobbs 40
   3. Dravid 35
   4. Kallis 34.5
   5. Lara 33.5
   6. Yousuf 33.5
   7. Hayden 33
   8. Viv Richards 32.5
   9. Ponting 32
  10. Hutton 32
  11. Tendulkar 31.5
  12. Chappell 31
  13. Border 31
  14. Miandad 30
  15. Inzamam 30
  16. Gavaskar 29
  17. S Waugh 25.5

[natty]

But I have a question -- given that one of the issues we have spent a lot of time debating on the DG is the consistency of these 4 as a group, should we use some other players -- lets say an SMG, Ponting, Lara, Kallis, Richards to see how our maharathis fare compared to the others --would be  a good benchmark comparison, I think

Agreed, but requires more time. TBD.

I was more interested in generating the 'Natty curves' later tonight because it will be relatively straightforward with what I already have.
http://www.cricketvoice.com/cricketforum2/index.php/topic,13753.msg170894.html#msg170894

Note that the projection at 50% on the Y axis to the X axis (via the natty curve) is the median.. This curve will contain a lot of the reasonable info one needs to know about a batting career; the median is just one point on the curve. 

aside: I dont see the need for dubious metrics like the difference between modified avg. and median.  One problem with measures of variance is that the variance is confounded with the mean (better batsmen will have higher variances) as far as batting distributions go.. what this means is that the difference between 10 and 20  is not the same as the difference between 110 and 120 (the scaling changes with position on the scale).  So defining consistency is trickier than it seems.