Dear all,
I am trying to figure this out.
My dependent variable is the total number of days a person has been off work because of sickness in a year together with the total number of days they have been present at work despite being sick in a year. Let's call it totalsick.
totalsick depends on how many days they have been 'exposed'. if you only work 1 day a week and you're sick all of them you might still have a lower totalsick than those that work 5 6 or even 7 days a week. To control for this someone suggested offsetting the amount of days worked.
The problem that I'm faced with is that I want to find out the effects on totalsick for those working less than 5 days a week (excluding the pure exposure related effect..)
--> i.e to find out if this work characteristic causes people to have a lower/higher or not at all different amount of totalsick than those that work 5 or more days a week, whilst still controlling for exposure.
My initial thought was to offset the variable on days worked per week (ranging from 1 to 7) and add an independent dummy variable coded 1 for people that work 1-4 days a week and 0 if 5-7. However I'm fearing this might be problematic in that the dummy variable is created from the variable I used to offset.
Thankful for any input.
Hi Kalle,The days exposed makes sense for an offset, as it provides the maximum potential days in which a person could be sick. The outcome is the actual number of days that a person is sick, which is thus a subset of the total possible days. So if one person has 3 days sick, out of a total 3 days possible, then they are at 100%, whereas another person may have 3 days sick, but that would be out of 30 days possible, which would be only 10% sick.Ariel
Hi again Ariel, Thank you for your answers !
But can I add a dummy for people that work 4 days or less as an independent variable whilst at the same time using the total days worked in a week (exposure) (i.e. 1 - 7 ) as an offset?
Kalle
I wouldn't use a dummy variable for days worked. It unnecessarily categorizes a continuous variable and doesn't add anything of value to your analysis.
The problem that I'm faced with is that I want to find out the effects on totalsick for those working less than 5 days a week (excluding the pure exposure related effect..)
If I don't add the dummy I can't see the effects and if I don't add the exposure then the effects won't tell me very much.
I'm looking to separate the exposure effect from the other effects, but don't really know how.
This sounds like you might want to consider using a regression-discontinuity design, where you draw the cutoff of 5 on the X variable and then detemine whether there is a discontinuity in the Y variable at the cutoff. I am attaching a recent paper I wrote on this. I think you may want to consider this approach.
Article Combining the Regression Discontinuity Design and Propensity...
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression