I am modeling insurance status in a logistic regression as separate dummy variables for private, Medicare, Medicaid, uninsured, etc. For people that are dual eligible, should I have a separate "dual eligible" (i.e. having both Medicare and Medicaid insurance types) variable or will this status be caught in having both the dummy variable hitting for Medicare and Medicaid equal to 1 catch this? Should this be an interaction term?
$\begingroup$
$\endgroup$
1
-
$\begingroup$ It's not completely clear what your situation is. What's the response? $\endgroup$– Glen_bCommented Dec 19, 2014 at 1:14
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
4
It is highly likely that people that fall under more than one category don't have a typical response close to the (transformed) sum of the two effects.
Interaction terms should be able to pick it up and might be a suitable approach if you expect the response (whatever it is) for people under Medicare and Medicaid to be different, but on the other hand -- depending on circumstances -- you may be able to have a new variable that's just "has at least one of these" that would adequately model what you need. If you're just trying to get a 'covered by some kind of insurance' variable, it would be a better way to go.
More details might allow for a better answer.
-
$\begingroup$ I believe I'm going forward with just making a new variable for these folks. Thanks for the input. $\endgroup$– j_bro_rxCommented Dec 24, 2014 at 6:56
-
$\begingroup$ My main variable for insurance goes something like: private, Medicare, Medicaid, dual eligible, other; with the reference category for each being uninsured. $\endgroup$– j_bro_rxCommented Dec 24, 2014 at 6:58
-
$\begingroup$ I'm also interested in a "ever uninsured" variable which is if someone had any period of time over two years that they were uninsured. Many of those that are categorized as having a certain insurance type in the above mentioned variable do hit on this "ever uninsured" variable but, of course, the majority are those that are uninsured during the whole time period. I am now wondering if there is going to be an issue including both insurance categories and this "ever uninsured" variable. $\endgroup$– j_bro_rxCommented Dec 24, 2014 at 7:00
-
$\begingroup$ I don't see that there's any inherent problem there. You may in practice run into multicollinearity issues, but that's another problem. $\endgroup$– Glen_bCommented Dec 24, 2014 at 7:11