7 independent dummy variables - which model is best to use?

[Ali Abdulqader]

Dear All,


I have 7 independent dummy variables in addition to my other 5 independent variables (prior research uses same control variables):

Can you please recommend me a model that will be best able to handle my 7 dummy variables, because I feel that the normal linear regression is not suitable and it is returning to me low R square but significant p value for the model.


Many thanks,
Ali Abdulqader

[Rich Goldstein]

please describe your dependent variable and something about your goal for the analysis

[Ali Abdulqader]

My dependent variable is a percentage (Premium paid), and I'm trying to see if a certain variable (which is a dummy variable) has an effect on premium paid; if a the characteristic is present then it should either have a positive or negative effect or can be zero. Thus I include the other variables as control variables which prior research shown that it has a contribution to premium paid.

When I use a normal regression function on STATA for my equation, it gives me R2 of 3% where as prior research documents 8-10% Adjusted R2 (However they do include other different variables than mine but not much).


So I was thinking that maybe because I have a lot of dummy variables, the normal regression equation can't properly fit through it, so I was thinking a different model might be better.


I will really appreciate your help. and thank you I really appreciate your help

Many Thanks,
Ali Abdulqader

[Ali Abdulqader]

By the way I'm only trying and interested in analysing one variable to see if it is significant, does it matter if the R2 is low?

[Carlo Lazzaro]

Alì:
what do you mean by "normal" regression? A linear one (i.e.: -regress in Stata-ish)?
Moreover, you seem to mix up R2 with Adjusted-R2, which are in part different beasts.
As per FAQ, the best way to increase your chances of getting helpful replies is to post what you typed and what Stata (not STATA, please) gave you back. Thanks.

[Rich Goldstein]

first, please do not send private messages for something that is being publicly discussed

second, since (as per your #3), your dependent variables is a percentage and is presumably limited to the 0-100 range, I would use "fracreg", not linear regression (if the numbers are between 0 and 100, you should divide by 100 to make them proportions first)

[Marcos Almeida]

@ Ali:

Since your outcome is a percentage, you may also consider performing a generalized linear model, selecting logit as the link and binomial as the family.