A tour of statistical techniques

A tour of statistical techniques

(To the tune of: There's a hole in my bucket)
Me: I've made a correlation matrix (grid), using Pearson's correlation coefficient (r), for a selection of variables, using my trusty statistical software SPSS.
Statistician: Yes, well done, but having done that, some of the variables are not continuous, so you should use Spearman's rho correlation coefficient (ρ).
Me: No problem, how's this?
Statistician: OK, fine, but having said that, SPSS just uses Pearson's coefficient anyway, if the sample size is as large as this one, whether you want it or not.
Me: Is it still OK?
Statistician: Hmm, yes, but there are also some variables with few categories, so you ought to use Cramér's V instead, based on chi-square (χ2). You just need to do each variable pair as a crosstabulation and request that statistic.
Me: Hmm, OK, I can find out how to do that...
Statistician: Well, we've made progress, but having done that, there are even some variables with only two values, and you should use the phi coefficient for them, it's simpler.  And, to be consistent, you should turn all the variables into binary forms.
Me: That won't be too fiddly to code...
Statistician: Yes, very good, but there's no need for that, because the phi coefficient (φ) for binary variables in 2x2 tables is just the same as the Pearson coefficient, which is easier to calculate.  Oh, and SPSS will present Pearson results as a nice grid, so why don't you use that instead...?
Me: ...

Topics: 
research
Author: 
DrDave
Post date: 
24 October 2013 08:15