Re: Continuous v. dummy coding

Thanks all! I'll have a look at the link and another pass through
Harrell's book.

Marc
Old Mac User wrote:
My initial reaction to this focused on the following statement.

"1. Work tasks (0x/day, 1-5x/day, 6-10x/day, 11-15x/day, More than 15x)

-Coded 0,1,2,3,4"

This immediately raises the following question. Is your response
(measure of performance)
going to be "rates". That is, incidents per unit of exposure.
Exposure in this case might best be in terms of "number of work
assignments"... hence a measure of performance such as
"incidents/number of work assignments."



mcam54@xxxxxxxxxxx wrote:
This is a continuation of a previous thread I started. Thanks to
everyone who helped me!
I have cross sectional data on work injuries that I want to do some
exploration with. I want to look at associations between work
tasks/postures and injury. I am using unconditional LR to look at each

primary risk factor adjusted for background factors (age, gender,etc).
I have two types of questions:

1. Work tasks (0x/day, 1-5x/day, 6-10x/day, 11-15x/day, More than 15x)
-Coded 0,1,2,3,4
2. Postures/activities (Almost never/not at all, About 10% of the day,
About 25% of the day, Half of the day or more)
-Coded 0,1,2,3

We discussed leaving them as linear instead of dummy coding and it
seemed to work well. However, I am probably going to be working with
someone who likes to use n-1 dummy variables in this situation and she
likes to look for linear trends using some sort of SAS function that I
am unaware of.

If I do code them as continuous......1. Is there a reference I can use

to justify my choice? 2. Would I have to change the coding scheme to
reflect the spacing of the categories.


Thanks,
Marc

.

Relevant Pages

  • Re: Continuous v. dummy coding
    ... incidents per unit of exposure. ... We discussed leaving them as linear instead of dummy coding and it ... someone who likes to use n-1 dummy variables in this situation and she ... likes to look for linear trends using some sort of SAS function that I ...
    (sci.stat.edu)
  • Continuous v. dummy coding
    ... I have cross sectional data on work injuries that I want to do some ... We discussed leaving them as linear instead of dummy coding and it ... someone who likes to use n-1 dummy variables in this situation and she ... likes to look for linear trends using some sort of SAS function that I ...
    (sci.stat.edu)
  • Re: Continuous v. dummy coding
    ... incidents per unit of exposure. ... We discussed leaving them as linear instead of dummy coding and it ... someone who likes to use n-1 dummy variables in this situation and she ... likes to look for linear trends using some sort of SAS function that I ...
    (sci.stat.edu)