I am working on model that involves a dummy dependent variable with probabilty of occurance of event (0,1) and ordinal independent variables (with the value increasing with the number of times another event happened) along with dummy control variables. Moreover, I am analyzing panel data. which model should i use?!
4
-
1$\begingroup$ Dear @user80489, I think it would help if you provided some specific detail about your problem. What kind of question are you trying to answer? $\endgroup$– Rob HallCommented Jun 23, 2015 at 13:41
-
$\begingroup$ Welcome to Cross Validated! We already have several answered questions on ordinal predictors: Logistic regression and ordinal independent variables, Coding for an ordered covariate, Continuous dependent variable with ordinal independent variable, & Logit with ordinal independent variables; so please have a look at them, & if you still need help add some detail to this question as @RobHall suggests. $\endgroup$– Scortchi ♦Commented Jun 23, 2015 at 14:21
-
$\begingroup$ I have edited the question. hopes it clarifies the objective.! $\endgroup$– AsadCommented Jun 24, 2015 at 16:20
-
$\begingroup$ Reading Concepts behind fixed/random effects models, What is the difference between fixed effect, random effect and mixed effect models?, & Difference between panel data & mixed model may help. $\endgroup$– Scortchi ♦Commented Jun 25, 2015 at 8:51
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
4
You could choose among using the Linear Probability, the Logit/Logistic or the Probit models.
The choice between those two models depends on the data you're going to analyze and the assumptions you assume to get the model estimates.
IMHO, you should generally use the Logistic model, as you suggested in the question, since it has more relaxed assumptions and more fat-tails distribution with respect to the normal distribution assumed in the LPM (Linear Probability Model) and probit model. I'd avoid using the LPM because it does not often estimate the predicted probability bounded between a range from 0 and 1, as it should be.
Hope this helps.
-
-
$\begingroup$ Yes, @user80489, but they are a little bit more complicated. For instance, you could use a neural network too, but IMHO it is better to use simpler to do that. If you can tell me the program you're using I can suggest some example of how to do that. $\endgroup$ Commented Jun 24, 2015 at 16:37
-
$\begingroup$ i am using SAS to run these regressions $\endgroup$– AsadCommented Jun 24, 2015 at 20:42
-
$\begingroup$ @user80489, if you own SAS base only, it is convenient to adopt the logistic model and you can find here support.sas.com/documentation/cdl/en/statug/63347/HTML/default/…. $\endgroup$ Commented Jun 24, 2015 at 20:48