Monday, July 10, 2006
Post the Seventeenth
Wherein your Host Praises the Gamma Distribution
Your host is a stupid man. But he is a stupid man who recently discovered one of his own mistakes and so is feeling pretty good right now!
I have a project that I’ve been working on for a year or more now and it just isn't going anywhere. I think the underlying ideas are interesting and important but the analysis wasn't holding up to statistical scrutiny.
One of the problems I have had with quantitative methods is learning to think statistically. This is hard to do when there is just so much out there that I simply haven’t learned yet. How could I think of framing a research question on differences in variation across some category when I didn’t know that heteroscedastic regression existed? How could I come up with a topic that examines different effects across various levels an observation is nested in when I didn’t know about hierarchical models?
It is also hard for me to think statistically because, as I have related previously I am math phobic. So today in class we were covering Generalized Linear Models. This is something a very able professor at UVA had attempted to teach me previously but we only covered the binomial and Poisson distributions. She had informed us that other probability distributions were available for analysis but, in my math-stupidity, I didn’t fully grasp what this meant.
Now I have discovered the glorious gamma distribution which far better fits my data than the normal, binomial or Poisson distributions. I get it now!
This:
Looks more like this (the red line):
Than it does like any of these:
Your host is a stupid man. But he is a stupid man who recently discovered one of his own mistakes and so is feeling pretty good right now!
I have a project that I’ve been working on for a year or more now and it just isn't going anywhere. I think the underlying ideas are interesting and important but the analysis wasn't holding up to statistical scrutiny.
One of the problems I have had with quantitative methods is learning to think statistically. This is hard to do when there is just so much out there that I simply haven’t learned yet. How could I think of framing a research question on differences in variation across some category when I didn’t know that heteroscedastic regression existed? How could I come up with a topic that examines different effects across various levels an observation is nested in when I didn’t know about hierarchical models?
It is also hard for me to think statistically because, as I have related previously I am math phobic. So today in class we were covering Generalized Linear Models. This is something a very able professor at UVA had attempted to teach me previously but we only covered the binomial and Poisson distributions. She had informed us that other probability distributions were available for analysis but, in my math-stupidity, I didn’t fully grasp what this meant.
Now I have discovered the glorious gamma distribution which far better fits my data than the normal, binomial or Poisson distributions. I get it now!
This:
Looks more like this (the red line):
Than it does like any of these:
After very quickly running some new analyses, it appears that, indeed, my hypothesized relationships do hold up to scrutiny (given a gamma distribution) and I might have a good conference paper (cross my fingers and pray it is publishable) on my hands now.
Credo ut intelligam.
Labels: ICPSR
