Hi all,
I'd appreciate a lot some advice on the following. I'm running an ordinal logit regression, where my dependent variable is marks obtained in a test (fail, pass, merit, and distinction) and my independent variables are all categorical with more than two categories (public school, private school, mixed school; the level of education of parents; socio-economic status). I'm very interested in learning about the effect that each level of my categorical independent variables have over my variable of interest. As such, I have added the "i." command to obtain a disaggregated analysis in my regression. I' concerned on:
1) How to make sure that my interpretation of the ologit model takes into account the ommited category for every categorical variable that is analysed with "i.". For example, do I have to specify that every interpretation for high socio economic status is taking into account that my base category is low ses (assuming that my base category is SES)? or is it not necessary? I have the same concern when interpreting the predicted marginal effects for those variables. Do I have to consider each variable's probability accounting for its base category?
2) I am not quite sure that I am interpreting my marginal effects correctly. Can they be analysed as percentages? For example, if I have a -0.033 coefficient for the "x" dummy variable in my outcome 1, which is Failing the test. Would it be right to say that an individual with "x" has 3% less likelihood of failing the test?
3) Also in regards of point 3, should I just analyse the marginal effects of variables that are statistically significant on my ologit model?
4) Last, but not least, how fair would it be to make a "robust check" of my model by running a logit regression on the probabilities of failing or passing the test? Meaning that, I would not look at the variation of grades, but only to a fail or pass the test scenario. This was not my first model, because I believe it is rich to see the variation of grades, but not sure if it could be used as a backup model to double check my first model's results.
Thanks a lot for your help in advance.
Best wishes
S
I'd appreciate a lot some advice on the following. I'm running an ordinal logit regression, where my dependent variable is marks obtained in a test (fail, pass, merit, and distinction) and my independent variables are all categorical with more than two categories (public school, private school, mixed school; the level of education of parents; socio-economic status). I'm very interested in learning about the effect that each level of my categorical independent variables have over my variable of interest. As such, I have added the "i." command to obtain a disaggregated analysis in my regression. I' concerned on:
1) How to make sure that my interpretation of the ologit model takes into account the ommited category for every categorical variable that is analysed with "i.". For example, do I have to specify that every interpretation for high socio economic status is taking into account that my base category is low ses (assuming that my base category is SES)? or is it not necessary? I have the same concern when interpreting the predicted marginal effects for those variables. Do I have to consider each variable's probability accounting for its base category?
2) I am not quite sure that I am interpreting my marginal effects correctly. Can they be analysed as percentages? For example, if I have a -0.033 coefficient for the "x" dummy variable in my outcome 1, which is Failing the test. Would it be right to say that an individual with "x" has 3% less likelihood of failing the test?
3) Also in regards of point 3, should I just analyse the marginal effects of variables that are statistically significant on my ologit model?
4) Last, but not least, how fair would it be to make a "robust check" of my model by running a logit regression on the probabilities of failing or passing the test? Meaning that, I would not look at the variation of grades, but only to a fail or pass the test scenario. This was not my first model, because I believe it is rich to see the variation of grades, but not sure if it could be used as a backup model to double check my first model's results.
Thanks a lot for your help in advance.
Best wishes
S
