Hi Everyone,

I have trouble conducting multinomial regression. I have derived four latent classes using LPA and then to identify class predictors using different independent variables, I run multinomial regression using latent classes as DV. I have got very large OR and standard error for one of the variables with no CI output. The sample size of this particular variable in the total sample is small which is 43 (3%), and in each outcome variables group that derived from LPA are: in class-1 = 20 (3%); in class-2 = 18 (3%); in class-3 = 2 (1%); and 3 (3%). Is there any minimum sample size we should consider as rule of thumb to exclude the variable? If yes, can you recommend me a reference article or book I should look at?

The syntax i used is below:

mlogit predclass_Rearranged Q1_Age i. Q2_Gender Education_recoded i.SES i.Location i.BMI_2Cat i.have_kids_8 Q29 Q9_1 Q7_recoded i.Q10 i.Q8_3 i.Q8_4 i.Q8_5 i.Q8_6, baseoutcome(3) rrr


Many thanks,

Liyu