Add support for analysis of deviance from car #580
Conversation
- added roxygen tags `@keywords internal - fixed some markdown syntax
…g the old behavior as a default)
| if(all(colnames(x) != "F values")) y$df.residual <- NULL | ||
| if(any(colnames(x) == "Cp")) y$df <- NULL | ||
|
|
||
| if(is.null(y$statistic)) { |
There was a problem hiding this comment.
Hmm, this I don't understand. Shouldn't canonize do this?
There was a problem hiding this comment.
I would consider these fixes specific to analysis of deviance that do not fit into a generic function such as canonize() -- moving this to canonize() would imply that we would need a list of test statistics that should be reported with degrees of freedom and a list of test statistics that should not be reported with degrees of freedom.
There was a problem hiding this comment.
I'm not sure I understand. This section does not deal with degrees of freedom, as far as I can tell. I'm referring here to the section where statistic is null and is replaced by deviance.
There was a problem hiding this comment.
I accidentally deleted my last comment.
Maybe the most concise answer is the following: In one very specific case (chi-squared tests for generalized linear models), a column named Deviance contains the difference in deviance which is distributed as chi-squared, and is used as the test statistic.
In canonize(), however, we rename Deviance to deviance, not statistic, because columns of this name typically do not contain the test statistic, but simply the deviance of the model (in addition to the test statistic contained in another column). Because column Deviance containing the test statistic is the exception, not the rule, I would opt for sticking with the current behavior of canonize() and deal with the exception within the specific method.
There was a problem hiding this comment.
What a mess. That makes sense, thanks for clarifying!
Hi @crsh,
This is my first shot at supporting analysis of deviance output from the car package. I need this for one of my manuscripts. There are probably a lot of things we could do to improve the functionality, but this will do in my use case, and I think something is better than nothing.
Best,
Marius