Hello,

I am currently studying the relationship between firm size and access to finance by using the World Bank Enterprise Survey. In my empirical analysis, first I have tried the ordered logit estimation which violates the proportional odds assumption. When ı check the Brandt test, I have seen that the problematic variable is the regions of Turkey which is a categorical variable between 1 and 12. For that reason, ı decided to move on with generalized ordered logit. However, ı have a question in mind. How should ı specify the categorical region variable as nonparallel (npl)? When ı included it with i. notation ı encounter with this warning (Warning: variance matrix is nonsymmetric or highly singular) and also standard errors are not reported.

My regression equation is as follows,

gologit2 access_to_finance firm_age i.firm_size private_foreign manag_experience export_orientation external_funding manager_female i.sector i.region, npl(i.region)

access to finance : categorical variable between 0 (no obstacle) and 4 (very severe obstacle)
firm size: categorical between 1(micro) and 4 (large)
firm_age, private_foreign (the percentage ownership by foreigners) , managers_experience, export_oritation (the percentage of sales provided from export), and external_funding (the percentage of funding obtained from external resources such as banks for day to day operations)
manager female (binary equal to 1 if female)


Thanks in advance for your advice and time,

Cemre