12.04.05
Posted in grad school at 2:59 pm by
An open letter to statistics study groups:
Dear Statistics Study Groups,
If you insist upon studying (loudly) at the local coffeeshop for your upcoming statistics test, here are a few things to keep in mind:
1) Just because a group member is the loudest, does not make–much to my personal chagrin–that person the smartest in the group. Thus, allowing this person to lead your little escapade into the world of parametric, inferential statistics may not be the best decision you’ve made.
2) Your study guide was not wrong. You were. In fact, if a strong curvilinear relationship (a quadratic relationship, for example) is estimated using a linear function such as the Pearson’s correlation coefficient, that coefficient will, in fact, drastically underestimate the strength of the relationship, not overestimate it. There’s no sense getting in a huff about how the study guide is wrong.
3) Statistical tests are based upon assumptions. One assumption that you probably shouldn’t make is that no one else in the coffeeshop you’re patronising understands stats as well as you. For, when there are folks who–despite their own ineptitude in statistical procedures–know vastly more than you when it comes to inferential stats, the sheer volume of your misinformation may prove distracting to other coffeeshop patrons. In short, your status as a grad student clearly is a poor indicator of your aptitude at stats.
Still trying to purge the misinformation,
Brandon
Permalink
Trackback URL »
http://www.badchristian.com/2005/12/04/like_fingernails_and_chalkboards/trackback/
Posted in grad school at 2:59 pm by
An open letter to statistics study groups:
Dear Statistics Study Groups,
If you insist upon studying (loudly) at the local coffeeshop for your upcoming statistics test, here are a few things to keep in mind:
1) Just because a group member is the loudest, does not make–much to my personal chagrin–that person the smartest in the group. Thus, allowing this person to lead your little escapade into the world of parametric, inferential statistics may not be the best decision you’ve made.
2) Your study guide was not wrong. You were. In fact, if a strong curvilinear relationship (a quadratic relationship, for example) is estimated using a linear function such as the Pearson’s correlation coefficient, that coefficient will, in fact, drastically underestimate the strength of the relationship, not overestimate it. There’s no sense getting in a huff about how the study guide is wrong.
3) Statistical tests are based upon assumptions. One assumption that you probably shouldn’t make is that no one else in the coffeeshop you’re patronising understands stats as well as you. For, when there are folks who–despite their own ineptitude in statistical procedures–know vastly more than you when it comes to inferential stats, the sheer volume of your misinformation may prove distracting to other coffeeshop patrons. In short, your status as a grad student clearly is a poor indicator of your aptitude at stats.
Still trying to purge the misinformation,
Brandon
Permalink
Trackback URL »
http://www.badchristian.com/2005/12/04/like_fingernails_and_chalkboards/trackback/
walter william de li moicani said,
December 4, 2005 at 8:51 pm
bullshit
David Mackey said,
December 4, 2005 at 9:33 pm
You sounded smart to me, but I don’t know anything about inferential statistics, so I have no way of telling if you are right or just making it up like them as a joke?
dorsey said,
December 4, 2005 at 10:56 pm
No, he’s absolutely right. Pearson’s does indeed underestimate the relationship, but may be less useful if the underlying assumption of normality is violated. Non-parametric correlation methods, such as Spearman’s and Kendall’s may be useful when distributions are not normal; they are a little less powerful than parametric methods if the assumptions underlying the latter are met, but are less likely to give distorted results when the assumptions fail.
jvjannotti said,
December 5, 2005 at 8:38 am
I was a stat major long ago and my grad work was business with quantitative emphsasis. Your post brought some weird numerical curvilinear flashback. Quite strange.
I had to read item number two like three times, it’s been so long since I tried to comprehend that language.
And dorsey’s comment above made my eyes roll back into my head in ecstasy. I actually did like that stuff once upon a time.
Brandon said,
December 5, 2005 at 10:13 am
Dorsey! I never knew you were a stats guy! Just a short aside, I think you may be confusing two assumptions of Pearson’s correlation–the normality assumption refers to normality within the conditional distributions, I think you might mean that when the assumption of linearity is violated that Spearman’s and Kendall’s are more appropriate options.
Sometime, we’ll have to chat stats.
(Holy shit, did I just say that.)
dorsey said,
December 5, 2005 at 10:40 am
Gotcha. I’m sorry.
I was going to see how long I could go with this, but I have too much respect for you to let you be the patsy. So I’ll confess.
I made all that shit up. Ok, I didn’t make it up so much as cut and paste a paragraph from Wikipedia’s entry on correlation.
I did take stats, but I have no idea what I’m talking about.
{impishgrin]heehee![/impishgrin]
rick said,
December 5, 2005 at 10:45 am
I made an “A” in Stats and chicks still dig me.
Brandon said,
December 5, 2005 at 10:53 am
Damn. I hoped, Dorsey, that you’d prove an asset come COM 901 study time.
Audrey said,
December 5, 2005 at 1:51 pm
Wow. Stuff right in my wheelhouse for once …
Pearson’s correlation is appropriate under either of two conditions:
(1) When the variables under consideration follow a bivariate normal distribution; and,
(2) When the conditional mean of one variable is a linear functional of the other variable.
In case (1), Pearson’s correlation is the maximum likelihood estimator. In case (2), Pearson’s correlation is a reasonable estimator. By the way, when case (1) is a sufficient (but not necessary) condition for case (2).
Bottom line: if E{Y|X} = a + bX + cX^2, the value of Pearson’s correlation will depend on both the range in which X is observed and the values of parameters a, b, and c. It is biased towards 0; i.e., its square underestimates the strength of the relationship between Y and X.
Audrey
Zeke said,
December 5, 2005 at 5:51 pm
I remember median, mean, average, and standard deviation. Plus I know the significance of a bell curve and what regression analysis is. But most importantly, I remember Disraeli: “There are lies, damned lies, and statistics.”
dorsey said,
December 5, 2005 at 8:41 pm
Audrey, thanks.
That’s all I was really trying to say. Had I known that 25 years ago, well… I don’t really know what would have happened, but I’ll bet it would have been good.
;-)
Audrey said,
December 6, 2005 at 1:05 pm
Zeke,
Actually, that was Sam Clemens, not Benjamin Disraeli. Mark Twain was well-known for crediting others with some of his more inflammatory bon mots.
Evidence? The first known citation is from Mark Twain’s A Tramp Abroad, and the comment appears nowhere in the writings of Disraeli.
As commentaries on statistics go, however, I prefer this one. Figures may not lie, but liars can figure.
Audrey