1. ## Ordinal predictors

I know there was a thread on this. I could not find it.

This is the type of issue that concerns me. These authors had an ordinal predictor that took on from 2-4 values [they may have treated this as a dummy, but there article does not support this].

Can you really have predictors coded this way?

" Restoration was scored by adding the number of services received...(range 0-4). Maintenance was scored in a similar fashion...(range 0-2)."

2. ## Re: Ordinal predictors

Are you referencing the tread on loss of info when categorizing ordinal data?

3. ## Re: Ordinal predictors

Possibly . In that thread we ended up talking about categorical predictors even if that was not the original topic. And that is my point above. How legitimate is it, what are the ramifications, of coding predictors like that rather than as a series of dummy variables.

4. ## Re: Ordinal predictors

I have no idea what your original quote represents. Is it count data or what? Not following the example.

5. ## Re: Ordinal predictors

They did things like count how many services you received. So if you got 2 services they counted that as 2 and if you got 4 they counted that as 4 (although measures commonly had 5 or fewer distinct levels). I am not sure if that is "count data" - its not what I think of as that.

6. ## Re: Ordinal predictors

So how did they treat this variable?

If you think of this in a logistic model, treating it as a dummy variable or continuous would work. So the OR would be 2 vs 1, etc. and as continuous it would be a one task increase = blank greater odds. However for this latter approach, you could not generalized/extrapolate outside the empirical range. I may prefer the former method though, since each group would have a unique slope and this could address a none linear increase.

7. ## The Following User Says Thank You to hlsmith For This Useful Post:

noetsi (02-22-2016)

8. ## Re: Ordinal predictors

They did not use dummy variables as predictors. They used instead a predictor with say 4 likert scale levels. If I understand what you said, this would be valid. I have doubts about using predictors with so few levels unless you are using dummy coding.

I guess I could live with, its ok as long as you don't generalize beyond this range

9. ## Re: Ordinal predictors

Yeah, with 3 or 5 groups, it would be limited and I would also prefer using categories.

10. ## Re: Ordinal predictors

By categories do you mean dummy variables?

 Tweet

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts