Dynamic Multinomial Logit Model

[ESTHER OSADOLOR]

Hello everyone,

I am working on informality and poverty dynamics using panel data from Nigeria and I am trying to run an analysis using the "dynamic multinomial logit model" to estimate the informal employment dynamics. In this case, I am trying to ensure that the state dependence of the employment status i.e. the effect of past employment status on present employment status, is accounted for. However, using the "gllamm" command in stata (got this from the Stata forum), I am able to estimate the model but nothing about the state dependence shows in the result. I also tried the "gsem" command but no result on the state dependence. Hence, it appears I am not implementing it well.
Below is a snippet of my data:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(pid w_year w_inf_type)
  1 2010 2
  1 2012 4
  1 2015 .
  1 2018 .
  8 2010 3
  8 2012 3
  8 2015 .
  8 2018 .
 15 2010 3
 15 2012 2
 15 2015 2
 15 2018 .
 21 2010 .
 21 2012 .
 21 2015 .
 21 2018 .
 25 2010 .
 25 2015 3
 25 2018 .
 25    . .
 34 2010 2
 34 2012 2
 34 2015 .
 34 2018 .
 41 2010 2
 41 2012 1
 41 2015 .
 41 2018 .
 45 2010 3
 45 2012 2
 45 2015 .
 45 2018 .
137 2010 5
137 2012 6
137 2015 .
137 2018 .
142 2010 6
142 2012 1
142 2015 .
142 2018 .
148 2010 .
148 2012 .
148 2015 .
148 2018 6
149 2010 5
149 2012 .
149 2015 .
149 2018 .
161 2010 4
161 2012 .
161 2015 4
161 2018 .
167 2010 5
167 2012 .
167 2015 6
167 2018 .
173 2010 4
173 2012 .
173 2015 .
173 2018 .
178 2010 .
178 2012 5
178 2015 6
178 2018 .
191 2010 3
191 2012 2
191 2015 .
191 2018 .
242 2010 .
242 2012 .
242 2015 .
242 2018 .
249 2010 .
249 2012 .
249 2015 .
249 2018 .
252 2010 .
252 2012 5
252 2015 .
252 2018 .
257 2010 6
257 2012 6
257 2015 6
257 2018 .
261 2010 .
261 2012 .
261 2015 .
261 2018 .
264 2010 4
264 2012 .
264 2015 .
264 2018 .
266 2010 .
266 2012 .
266 2015 .
266 2018 .
272 2010 4
272 2012 .
272 2015 .
272 2018 2
end
label values w_inf_type inff1
label def inff1 1 "Formal self employment", modify
label def inff1 2 "Formal wage employment", modify
label def inff1 3 "upper-tier informal wage", modify
label def inff1 4 "Lower-tier Informal Wage", modify
label def inff1 5 "Upper-tier Informal self", modify
label def inff1 6 "Lower-tier Informal self", modify

Thank you

[Andrew Musau]

I am trying to run an analysis using the "dynamic multinomial logit model" to estimate the informal employment dynamics. In this case, I am trying to ensure that the state dependence of the employment status i.e. the effect of past employment status on present employment status, is accounted for. However, using the "gllamm" command in stata (got this from the Stata forum), I am able to estimate the model but nothing about the state dependence shows in the result.

You do not show any code. Do you include a lagged dependent variable on the RHS to make the model dynamic? If so, what does "nothing about state dependence shows in the result" mean? Stata 17 and later versions have the command xtmlogit which estimates a panel multinomial logit model, so previous replications of this command using gllamm and gsem are not needed. Just use the command directly. In any case, I have to see an empirical application of such a model to give any useful advice, so if you have some literature references, include them here.

help xtmlogit

[ESTHER OSADOLOR]

Dear Andrew,

Thank you very much for your feedback. I did not include a lagged dependent variable on the RHS because I wasn't sure if I needed to do that as a similar analysis for poverty dynamics using xtpdyn didn't require that, but I will do that now.

Regarding the xtmlogit command, I have tried implementing it but it wasn't converging. I really cannot trace the cause of that, I will really appreciate your thought in that regard.

Thank you.