I am developing a time variant fixed effect model with five categorical independent variable. All categorical variables have three categories. I have three questions.
First category is used as reference category. How can I interpret the results? Will it be same as in OLS. Can I interpret that beta of a variable is the change in dependent variable as compared to reference category. I attached the results obtained. can i interpret that for 2 category of DSize dependent variable decrease by 0.0005024 as compared to reference category?
Second, if one category is significant and other is insignificant as compared to reference category how to interpret it.
Third, here how to interpret the value of intercept?
Hi Umar,
it is quite simple - you obtained solution - for each of 5 categorical independent variables and each category except of reference one effect size in Coef. in your outputs.
For variables DSize, DSG both categories 2 and 3 are not significantly different from category 1.
For variables DCR, DTang both categories 2 and 3 are significantly different from category 1.
For variable time (year) the years 2008, 2009 are significantly different from the year 2006, but other years not.
_const is a mean effect with all variables are in reference categories.
BR
Lada
thanx Ladislav Pecen
Can i find the mean effect of individual reference categories. Moreover, since its a fixed effect model. So, is this intercept also include the fixed effect intercept? if yes then how can i find the individual effect of each reference category?
Intercept = mean effect if all 5 independent variables are at reference category.
If you are interested in one independent variable only you have to weight all other 4 variables by relative frequencies of their categories
Thanx Mikhail
it worked for me. However, when i include all three categories of all five categorical variable then only first categorical variable showed coefficients for all three categories while rest of variables showed same result as compared to reference category.
What i did, I use these categorical variables one by one as first independent variable and this is how i get coefficient value for all variables.
However, next thing that was new for me. I found that some of variables showed significant coefficients but when studied with respect to reference category they showed insignificant results.
thanx once again
Dear Mikhail
Can you help once again.
Please find the attached file containing analysis as you said. Again you can see that first categorical variable is showing coefficients for three categories but not for the rest. DCR, DTang, DNDTS and DSG are showing the results for two categories as compared to reference category (i.e. third category).
However, when i used DCR as first variable then it is showing coffecient of its categories but rest of other variables are showing results with respect to their reference categories.
I wonder where the effect of ommitted variables (i.e refernce catefories) since there is no constant option is used.
Dear Mikhail,
You can not include all categories in all questions (DSize, DCR ...) because in such a case the columns of 0-1 values sum to the column of 1 and the matrix X'X (of independent variables) can not be inverted. The first method: to set one of category as the reference one, was better. The intercept is (in this case) the mean value of dependent variable when the reference categorie is "choosen" - the mean value that we could observe in "basic period" (the year that is ommited in report in a part "Time").
Mikhail,
I understand your point but this is more complicated case: we have a few questions with answers (categories) transformed into dummy variables. We can keep all variants only for one of the questions. In other cases we have to choose reference category. That id why the computer program rejects one cataegory for the second and next questions.
Please, see the attached illustration.
Thanx for all of the comments.
However, My objective is to get mean value for all dummy variables.
For instance what will be the value of dependent variable when Dsize = 1, DCR = 2, DTang = 3, DNDTS = 2 and DSG = 1.
reset of categories are equal to zero. In this case what will be the average value of dependent variable?
After about a year I got the solution to get mean scores in such situation. Simply I used "margins" command of STATA to get the mean values of all the factors. This margins command also helped me to find out the marginal effects or derivatives of interaction term between continuous and categorical variables.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Modeling
- Analyses of Variance