Stu,

Don't you think I've gotten enough of that from Bill by now? Thanks for the helpful comments.

The reason I am more accepting of Perseverance now is because there is research that supports it. I rejected it before because the reasoning was flawed, and the research I did did not support it.

That "best regards" is meant as a joke, right? You certainly couldn't mean it from the tone of your message.

I might see you in Fredericton in 2005. Or 2006. Or whenever.

Iain.

More response...

Please do not refer to me as an egghead. It's insulting and condescending.

K&R's take is hardly "new". It was published in the original Compendium, which is getting very close to 20 years old now. The take is not much newer than the issue, since save percentage only became an official stat in what, 1981?

I DID NOT close my mind to the idea. I researched it and found no support for it. I therefore provisionally concluded that it was not valid. Upon new research becoming available, I changed my conclusion, now saying it has some validity. That is the exact OPPOSITE of being close-minded. Being open-minded is not the same thing as blindly accepting assertions without verifying the support for said assertions. I base my conclusions on evidence, not gut feeling or sentiment or appearances.

Iain.

Hello Iain,

I never referred to you as an egghead. My use of the word was inflammatory at worst. I don't understand why you chose to take it personally.

Anyway, K&R's take on this issue was indeed new at the time of first Compendium's publication. The issue of goaltender save percentage (nee "Fielding Averages") was first discussed as far back as the mid 1960's. K&R were putting a new spin on a 20-year-old issue that had only recently become an official stat. They were speaking the language of "Joe hockey fan", and I applaud them for deciding that "Joe Fan" might appreciate what they were saying better than trying to figure out what a correlation coefficient was!

On your website you dismiss K&R after running a couple of correlation coefficients and you used the terms "patently ridiculous" and "useless and meaningless" in your "critique" of their work. Nice touch. At no time did you ever "provisionally conclude that it was not valid". Quite frankly Iain, you sh*t all over it.

In May of 2002 you used this forum to reply to Cliff Ash and you seemed to infer through a tounge-in-cheek post that you were taking the lead on this subject...

"I hate to break away from my natural Canadian modesty, but I do consider myself to be a hockey stats revolutionary. I was the one who presented the aforementioned results about shots and save percentage. And I do in fact have something in mind regarding quality of shots and their relationship to save percentage, but to be completely honest I just haven't gotten around to it yet."

Correct me if I am wrong but you did not follow through with any research of substance on the issue.

As for correlation coefficients it is my understanding that plus and minus one indicate perfect positive and negative relationships whereas zero indicates that the X and Y values do not co-vary in any linear fashion. The correlation coefficients can assume _any_ value between -1 and +1, depending upon the _degree of the relationship_.

So, who arbitrarily decides what the _degree of the relationship_ should be in _your_ studies?

Continued best regards,
Stu

I did fully intend to pursue research VERY similar to what Alan Ryder did; he simply beat me to it. And did a very nice job, too.

As for how strong a correlation coefficient is, there is no "right answer". Professional statisticians generally consider a coefficient of plus or minus 0.7 to be very strong, possibly indicating a causational relationship. As always, remember that correlation does not necessarily imply causation. Another way to look at it is the R-squared stat. I believe (and I'd have to look it up) that R-squared is simply the coefficient of correlation, squared. R-squared represents the approximate percentage of variation explained by the variable in question. So a correlation coefficient of .70 would mean and R-squared of .49, meaning that 49% of the variation of one number is explained by the other. The lower you go, the less it is. A coefficient of .20 would mean only 4% of the variation is explained. Again, I don't have my textbook handy, but I think that's right.

As for my earlier work on Perseverance, perhaps I was a bit brasher in those days, and that was indeed a long time ago, in terms of my "career" as a hockey analyst. I was just starting out as a dedicated hockey analyst, whereas now it's just something I do. Would I have written the same words today? Not likely. But the past is the past.

As for "Joe Fan", one of the worst things you can do is "dumb down" what you're doing. Teachers should not reduce their lessons to the lowest denominator, they should bring their students up to an appropriate level of understanding. Not that I'm saying K&R dumbed down their work, they certainly didn't. Most of what is written in the two Compendiums is quite brilliant, including the statistical stuff. Some of it is just incomplete. They assumed that Perseverance was an improvement on save percentage without testing it. They made a slight mathematical boo-boo in their adjusted plus-minus calculations. They also should have compiled more than one season's worth of crucial scoring numbers to test its validity as an individual stat. That's about it.

Iain.

Just had another look myself. The words "patently ridiculous", which you took out of context, were used because it is something K&R themselves wrote. It was a little bit of irony.

And seriously, when was the last time you said "I provisionally conclude that this is not valid"? My website is certainly not an academic journal, and even articles in academic journals don't always use language like that. Obviously, it was provisional, since I have now stated there may be some validity, based on new evidence.

Iain.

Hello you all...

I just re-cracked open my K&R Compendium (the latest one) and was reading about Perserverence. It seems obvious to me that there would be a relationship between number of shots and save percentage, and given the above post it seems that there is one.

What I'd like to know is why K&R came up with .6 ( I Think ) as the number to factor into #shots taken. It seems arbitrary to me. I'd like to see them use a number that comes out of research.