Re: Concerning Ordinal Predictors, Please step through this code

[Paul Johnson <>]

Cheng-shan Frank Liu wrote:
> Thank you, PJ. It is clear for me now that we have to check every time 
> when we attempt to include ordinal variables in a model.
>
> 1. Does this rule of thumb apply to an index or scale-like variables, 
> such as a 24-point media attention index?
Matter of opinion, I think.

> 2. If the anova test rejects the null hypothesis that there is no 
> difference between the two models (mod3 and mod5 in your example), does 
> it suggest that we need to transform the ordinal variable? How to do so?
NO, that's backwards. If there isno difference, leave it as a scale 
variable.

If there is a difference, that means you have trouble, need todo 
something else.

> 3. My two more technical question are: How to use multiple variables in 
> avas() in the "acepack" library? (The manual only shows how to transform 
> dv and iv into a linear relationship.) And, can we transform only 
> specific variables and put it back to our lm or glm regression?
Don't know. Don't go in that direction, given shortage of time. Instead, 
enter them as "ordered" and if it makes a difference, then leave it that 
way.  Remember, interpret predictions, not b's.

pj


> Frank
>
>
> Paul Johnson wrote:
>
> > If you are one of the people with a predictor like "Education" or 
> > "PartyID", please test this code and read the commentary to see if you 
> > can pull your head out. Or my head out.


--
Paul E. Johnson                       email: pauljohn_AT_ku.edu
Dept. of Political Science            http://lark.cc.ku.edu/~pauljohn
1541 Lilac Lane, Rm 504
University of Kansas                  Office: (785) 864-9086
Lawrence, Kansas 66044-3177           FAX: (785) 864-5700