Class is Permanent, Form is... an Illusion?

[Jungle Jumbo]

It might just be my eyesight and an attempt to overread the Atapattu graph, but would there be a (very slight) correlation, somewhere around the line of y=1.5x-75?

 

[Goughy]

This is something that has been disussed for a while and neatly illustrate by Neil here.

There will be times when confidence is high or low or technique is tuned or sloppy but most 'form' is coincidence and random.

Highlights the problems with selecting average players who are 'in-form' and why quality players with a proven track record should always be prefered (unless taking a flyer on an inexperienced talented youth).

 

[Top_Cat]

Definitely have a problem with looking at form as a function of scores alone. Too many uncontrolled-for factors at play to conclusively say (or even strongly suggest, really) that form does/n't exist. The lack of a statistical association in the form of a correlation doesn't mean there isn't one, either.

It's a complicated issue. I mean, sure, you could argue that a big score in one knock won't mean you're going to get one in the next dig. But that could be explained by everything from fatigue to bad luck i.e. there's a bunch of confounds with physical activity. This sort of analysis pre-supposes that if you then don't follow-up one big score with another, you're 'out-of-form' too. What about the effect on the next 10 innings? Is there even one? I dunno. Also is predicated on the assumption that scoring runs is entirely dependent on what the batsman does.

An example; Andy Strauss arrived in Aus in 2006 looking like he was hitting the ball beautifully yet after a run of outs, only really looked genuinely out-of-form in Sydney. Before that, looked great but just kept getting out. Yet, by that run of scores, was out-of-form for virtually the whole series. Rick Ponting in 2001, right before he went to India, looked in rare form and we all know what happened to him there. From all reports he was smashing them in the nets too.

Point is, psychology, luck, bowling quality (not just in terms of averages but on the day too), fielders, captaincy by the opposition, etc. play massive parts in a sequence of scores by batsmen so using purely runs scored, in terms of the quality of measurement, is a fairly insensitive way of looking at it.

And using Excel for stats, Neil? For shame!

 

[gio]

I think the more important graph is the second one, as it's liekly to show a more accurate representation. Any innings can be ended early due to a great ball/bad decision etc. therefore if form did exist, it would be more likely to be shown in the second graph.

Any chance of adding a line of best fit Neil, just for ****s and giggles?

 

[GIMH]

And using Excel for stats, Neil? For shame!

what should he use?

 

[James_W]

Microsoft Works Spreadsheet Machine Maker 3000 tbh.

 

[Top_Cat]

what should he use?

Anything but.



For something like this, it's probably alright but there are usually fairly significant errors in Excel's inferential functions. I use SPSS at work but SAS, Stata, etc. are all good.

 

[pasag]

Anything but.



For something like this, it's probably alright but there are usually fairly significant errors in Excel's inferential functions. I use SPSS at work but SAS, Stata, etc. are all good.

Evil program

 

[GIMH]

Anything but.



For something like this, it's probably alright but there are usually fairly significant errors in Excel's inferential functions. I use SPSS at work but SAS, Stata, etc. are all good.

I've been getting to grips with Excel in a big way of late, moritified to hear that it isn't the best thing sliced bread itbt

Whoever mentioned MS Works - I couldn't believe it when I found out I actually have that on my laptop. Might have to use it given that my Office trial has just ran out

 

[Top_Cat]

I've been getting to grips with Excel in a big way of late, moritified to hear that it isn't the best thing sliced bread itbt

Depends on what you're using it for. As a spreadsheet, all good. That's what it's designed for. As a stats package, for which is has an add-on, not so great. It's like a lot of software which does its primary job well but scope creep means it has other functionality added to it but doesn't do it nearly as well. Nero is a decent burning program but it's media functions aren't great in my experience.

 

[chasingthedon]

Very interesting discussion, Neil.

Thing about baseball is, they play so many games that the sample sizes are usually much larger than, say, Tests - a baseball career may cover 1,500 games or more, and inferences on the existence of form are thus more easy to draw. Plus, they typically look at hitting - e.g. when discussing form, they might measure how often they get one hit followed by another, very small events in time as compared to looking at complete innings scores in cricket.

And as Top Cat states, there can be lots of variables between successive innings in cricket. These variables can really only be minimised by looking at a whole career.

I think it's fair to say though that, looking at Test careers, some batsmen are more "mercurial" than others. Some time ago I was thinking about averages and how they can be misrepresentative of some players, for example Ian Bell (avg around 42) and Marvan Atapattu (avg around 39). I ran an investigation into z-scores, which measures how likely a player is to make a particular score based on his career spread (mean and standard deviation). Bell and Atapattu are about the same in probability in making a 50 (42.9%-41.3% in Bell's favour), but Atapattu is much more likely to make a higher score (three times more likely to score 150, a whopping eight times as likely to score a double), and hence more likely to turn a game.

You can also use z-scores to predict the statistical likelihood that a player will make a score he hasn't actually achieved yet, based on his innings scores to date. For example, looking at Pietersen and Hussey, Hussey is statisitically more likely than Pietersen to score a 50, 100, 150 and 200, but Pietersen is more likely to score 250.

So the above can be used to look at consitency of form and which players are more streaky. Small sample size and other variables notwithstanding, of course!

 

[zaremba]

Definitely have a problem with looking at form as a function of scores alone. Too many uncontrolled-for factors at play to conclusively say (or even strongly suggest, really) that form does/n't exist. The lack of a statistical association in the form of a correlation doesn't mean there isn't one, either.

It's a complicated issue. I mean, sure, you could argue that a big score in one knock won't mean you're going to get one in the next dig. But that could be explained by everything from fatigue to bad luck i.e. there's a bunch of confounds with physical activity. This sort of analysis pre-supposes that if you then don't follow-up one big score with another, you're 'out-of-form' too. What about the effect on the next 10 innings? Is there even one? I dunno. Also is predicated on the assumption that scoring runs is entirely dependent on what the batsman does.

An example; Andy Strauss arrived in Aus in 2006 looking like he was hitting the ball beautifully yet after a run of outs, only really looked genuinely out-of-form in Sydney. Before that, looked great but just kept getting out. Yet, by that run of scores, was out-of-form for virtually the whole series. Rick Ponting in 2001, right before he went to India, looked in rare form and we all know what happened to him there. From all reports he was smashing them in the nets too.

Point is, psychology, luck, bowling quality (not just in terms of averages but on the day too), fielders, captaincy by the opposition, etc. play massive parts in a sequence of scores by batsmen so using purely runs scored, in terms of the quality of measurement, is a fairly insensitive way of looking at it.!

But if there is no discernible statistical correlation between one's recent scores and one's next scores, even using these sorts of "raw" data, then even if form exists (as it seems intuitive to suppose it does) it is a much, much less significant force than we had ever previously supposed.

 

[Neil Pickup]

Anything but.



For something like this, it's probably alright but there are usually fairly significant errors in Excel's inferential functions. I use SPSS at work but SAS, Stata, etc. are all good.

I don't think it's possible to get the Pearson coefficient wrong. Surely.

I would use Minitab but I have formatted the computer and my housemate from Uni has the CD.

 

[Burgey]

First Chance Average.

That is all.

 

[Top_Cat]

But if there is no discernible statistical correlation between one's recent scores and one's next scores, even using these sorts of "raw" data, then even if form exists (as it seems intuitive to suppose it does) it is a much, much less significant force than we had ever previously supposed.

It's too much of a blunt force measure to really draw that or any conclusion for certain, I reckon and just raw number of runs scored is too macro for my tastes. It's the end result, sure, but the factors which determine how many runs scored are enormously complicated. One obvious example; the pitch. No pitch is totally even because it's packed soil. So even if everything is in order, you might get a ball which holds up and boom, you're out. Or, worse, you might get a few in a row and after a couple of knocks like that, suddenly form has deserted you?

'Form' and maintenance of said form would surely be related to any number of emotional/cognitive/physical factors (confidence, emotional stability/lability, neuromuscular/physical conditioning, etc.) yet discussing the statistical association completely ignores all of them. Which is completely understandable because psych is still testing the waters with them too.

And they're all likely related and feed off each other. If you're feeling good from a conjugal visit from the missus on tour and have had your morning coffee (or diuretic pill ) but you've got a slight injury from the 150+ you scored last week, maybe it'll play on your mind and you'll be worried about injuring it further so you'll 'protect' it and might play a poor shot and get out early, even though you were 'in form'. What if the injury takes a bit longer to heal and you start to worry about it? What if it really, really hurt and even though it healed quickly, you're worried about hurting it again so you shelve the cut shot, put yourself under pressure to score with less preferred shots and get out trying to up the rate? Etc., etc.

It takes a fairly extraordinary confluence of psychological factors to do well in any sport, let alone one like cricket. It's why very few people can do it and it's no coincidence that the big difference between a top shelf grade cricketer and a Test cricketer is, a lot of the time, a mental one. Even those who apparently find it easy have many off days and I can't imagine the mental effort required for a bloke like Mike Slater, with bipolar depression, to do well in a sport.

Point is, if just being able to play the sport is so dependent on state of mind and various interactions between/within psychological feedback looks, I don't think it's beyond the realms of possibility that the right confluence of these factors can still be in play and someone in form whilst still not scoring all that well. It comes down to what you define as form, really. If it's entirely down to the end result (runs) then sure, the effect is in question. But I'd argue it's a narrow definition for what form is and how long it could last if it really does exist.

 

[Top_Cat]

I don't think it's possible to get the Pearson coefficient wrong. Surely.

The stats add-on for the last version of Excel I used regularly (2003) Excel didn't do a significance test of the correlation.Just looking at a plot isn't enough to say much because there can be a strong correlation without it being significant. Anyone know if the later versions include that at least?

</Excel sucks as a stats package>

And mate, you'd be surprised at how silly some of the numbers these packages spit out can be. Even SPSS versions > 12 can be twitchy, especially if you use the GUI to come up with your numbers.

 

[Burgey]

For a giggle, can someone do Bradman's output?

 

[zaremba]

It's too much of a blunt force measure to really draw that or any conclusion for certain, I reckon and just raw number of runs scored is too macro for my tastes. It's the end result, sure, but the factors which determine how many runs scored are enormously complicated. One obvious example; the pitch. No pitch is totally even because it's packed soil. So even if everything is in order, you might get a ball which holds up and boom, you're out. Or, worse, you might get a few in a row and after a couple of knocks like that, suddenly form has deserted you?

'Form' and maintenance of said form would surely be related to any number of emotional/cognitive/physical factors (confidence, emotional stability/lability, neuromuscular/physical conditioning, etc.) yet discussing the statistical association completely ignores all of them. Which is completely understandable because psych is still testing the waters with them too.

And they're all likely related and feed off each other. If you're feeling good from a conjugal visit from the missus on tour and have had your morning coffee (or diuretic pill ) but you've got a slight injury from the 150+ you scored last week, maybe it'll play on your mind and you'll be worried about injuring it further so you'll 'protect' it and might play a poor shot and get out early, even though you were 'in form'. What if the injury takes a bit longer to heal and you start to worry about it? What if it really, really hurt and even though it healed quickly, you're worried about hurting it again so you shelve the cut shot, put yourself under pressure to score with less preferred shots and get out trying to up the rate? Etc., etc.

It takes a fairly extraordinary confluence of psychological factors to do well in any sport, let alone one like cricket. It's why very few people can do it and it's no coincidence that the big difference between a top shelf grade cricketer and a Test cricketer is, a lot of the time, a mental one. Even those who apparently find it easy have many off days and I can't imagine the mental effort required for a bloke like Mike Slater, with bipolar depression, to do well in a sport.

Point is, if just being able to play the sport is so dependent on state of mind and various interactions between/within psychological feedback looks, I don't think it's beyond the realms of possibility that the right confluence of these factors can still be in play and someone in form whilst still not scoring all that well. It comes down to what you define as form, really. If it's entirely down to the end result (runs) then sure, the effect is in question. But I'd argue it's a narrow definition for what form is and how long it could last if it really does exist.

I don't think this analysis is as narrow as you suggest. It doesn't proceed on the basis of any assumed reason for "form", it simply describes whether runs of form exist at all.

In particular, the analysis is not testing a hypothesis that "you scored runs last time and for that reason you're likelier to score runs next time" (ie A causes B); rather, that the fact that you scored runs last time may tend to indicate that you are in form, and a further consequence of that form is that you should be likelier to score runs next time (ie A and B are both caused by C).

Or to put it more simply if "form" existed then you'd expect runs of good scores to occur. But this analysis seems to show that such runs do not occur.

I should stress that I'm not a statistician and I'm not sure I really know what I'm talking about.

 

[Top_Cat]

I don't think this analysis is as narrow as you suggest. It doesn't proceed on the basis of any assumed reason for "form", it simply describes whether runs of form exist at all.

In particular, the analysis is not testing a hypothesis that "you scored runs last time and for that reason you're likelier to score runs next time" (ie A causes B); rather, that the fact that you scored runs last time may tend to indicate that you are in form, and a further consequence of that form is that you should be likelier to score runs next time (ie A and B are both caused by C).

Or to put it more simply if "form" existed then you'd expect runs of good scores to occur. But this analysis seems to show that such runs do not occur.

I should stress that I'm not a statistician and I'm not sure I really know what I'm talking about.

The results really say there's no positive association between runs scored in a knock and previous innings/average of previous 10. Extrapolating that to say 'form is a myth' is where I think it falls down.

 

[Redbacks]

So: without a reasonable definition of form you can't define an acceptable model..? It might be better suited to a baseballer if he enters each bat with the intent get a hit or home run. Rarely would a batsman in test cricket bat the same in the 1st and 2nd innings where a 25 off 100+ balls to save a match, declarations etc make it near impossible to make sense of the data without a test by test analysis at least. E.g. this method would highlight the great run of 100's Kallis went on in successive test matches, Bradman, Hayden et al.

Would a batter be considered in form who scored 113, 13, 12, 68, 160, 30? Not alot of correlation but conforming to the rule of a score a test and going @ 66.