search

friendly / VCDR

Visualizing categorical data with R
0 stars 0 forks source link

3+ variables passed to chisq.test()? #34

Closed philchalmers closed 9 years ago

philchalmers commented 9 years ago

Does this issue bite anywhere in the book, or in the vcd/vcdExtra packages? https://bugs.r-project.org/bugzilla/show_bug.cgi?id=16300

I imagine you gentlemen have rolled your own chi-squared tests for these arrays, but you never know if this comes up in auxiliary functions.

philchalmers commented 9 years ago

Apparently this was patched a month ago. Still could appear in the book though if it's not compiled using R-dev, but if it occurs in your packages you'll know next time you check for a CRAN release (hopefully).

davidjohannesmeyer commented 9 years ago

Actually, Kurt fixed this because I pointed him to the issue a few weeks ago, so I know about this. Btw, summary() for tables accepts higher-dimensional arrays...

Am 03.04.2015 um 18:16 schrieb Phil Chalmers notifications@github.com:

Does this issue bite anywhere in the book, or in the vcd/vcdExtra packages? https://bugs.r-project.org/bugzilla/show_bug.cgi?id=16300

I imagine you gentlemen have rolled your own chi-squared tests for these arrays, but you never know if this comes up in auxiliary functions.

— Reply to this email directly or view it on GitHub.

philchalmers commented 9 years ago

Great, glad we both stumbled across this.

Thanks for the tip on summary(), I wasn't aware of that. However, it doesn't seem to give as much returned information as I would like (no expected values, residuals, etc), which was really what I wanted to see in the first place. Too bad the chisq.test() patch was just to throw an error, because clearly summary() is computing the X2 value for higher dimensional tables, and therefore native support for higher-dimensional X2/expected values has already been written somewhere in the stats package (perhaps the summary method for tables should return information similar to chisq.test()?).

davidjohannesmeyer commented 9 years ago

The problem with higher-dimensional tables is actually that there is more than just one null hypothesis - why should we choose total independence? So in general, you are bound to specify a log-linear model anyway. What summary() does is actually problematic.

Best David

Am 04.04.2015 um 16:39 schrieb Phil Chalmers notifications@github.com:

Great, glad we both stumbled across this.

Thanks for the tip on summary(), I wasn't aware of that. However, it doesn't seem to give as much returned information as I would like (no expected values, residuals, etc), which was really what I wanted to see in the first place. Too bad the chisq.test() patch was just to throw an error, because clearly summary() is computing the X2 value for higher dimensional tables, and therefore native support for higher-dimensional X2/expected values has already been written somewhere in the stats package (perhaps the summary method for tables should return information similar to chisq.test()?).

— Reply to this email directly or view it on GitHub.