Adjusted risk ratios (ARR) after a multinomial logistic model

[Luis Ortiz]

Dear members of the list,

I am trying to generate the adjusted risk ratios (ARR) corresponding to one of my independent variables in a multinomial logit model explaining undergraduates' expectations upon bachelor degree graduation

mlogit expect3 ib2.edu_parents female progresion i.codi_a2 rendimiento, baseoutcome(4) 

After the output (which I omit because the table is quite long), I ask Stata to generate the adjusted risk ration of my independent variable of interest, a categorical variable that includes different levels of parental education

adjrr  ib2.edu_parents 

Quite unfortunately, this results in the following error message that I have not been able to discern, neither in the help of the 'adjrr' command nor in other messages posted in Statalist

. adjrr  ib2.edu_parents
factor-variable and time-series    operators    not    allowed 

I have tried with other ways of naming the independent variable of interest, but I have not find what I am doing wrong. For instance....

. adjrr edu_parents
at level for factor edu_parents not present during estimation 

Any guidance would be much appreciated

Kind regards and thanks for your attention

Luis Ortiz

[Rich Goldstein]

I am not familiar with this command, but in general there are at least 2 ways to do this: (1) no new programming involved here - use -fvrevar-; see

h fvrevar

(2) make a copy of the program, save it under a new name and change the syntax statement as shown in the following blog: https://blog.stata.com/2015/11/25/pr...tor-variables/

[Luis Ortiz]

Many thanks for your answer, Rich.

The second option looks a bit difficult to me. I have tried with the first one. It has indeed allow me to use the 'edu_parent' with adjrr, which is something I was not able to do before, but it seemingly results in treating that variable as as a continuous one, because the output is not provided category by category.

When you were so kind as to respond my initial post, I was trying to develop it in order to clarify why I am interested in 'adjrr'.

I am using a multinomial logit model to explain the effect that undergraduates' social background on their expectations of postgraduate enrollment upon graduation. In the dependent variable, there is also the possibility of stating that they expect 'Looking for a job', which is the category that I use as a reference category in the dependent variable, in the multinomial model that I present below.

mlogit expect3 ib2.edu_parents female progresion i.codi_a2 rendimiento, baseoutcome(4)

I am mostly interested in the effect of parental education (ib2.edu_parents) on expectation of postgraduate enrollment. I have already got very valuable advice here, in this list, about the use of the command 'margins' for estimating the average marginal effect of parental education on the expectation of the postgraduate enrollment. I learnt that this average marginal effect of parental education on the expectation of postgraduate enrollment (the category I am interested in my multinomial dependent variable) should not be interpreted with reference to the reference category in the dependent variable.

Yet, I am also interested in the effect parental education on the probability of expecting master enrollment relative to the effect it makes on the probability of 'Looking for a job'. I know that I may use relative risk ratios, adding RRR as an option at the end of the command, but RRR does not seem to give the size of the difference of the effect of parental education in one outcome ('Doing a MA') and the other ('Looking for a job') At least, they do not do directly.

This is the reason I explored and I came across with 'adjrr'. According to the documentation of this command, it looks as a nice way of getting this difference.Yet, I do not seem to be able to proceed wit it.

I wonder there is any other possibility of arriving to what I want.

At any rate, many thanks again for your help, Rich

Kind regards

Luis Ortiz

[Rich Goldstein]

I don't fully follow, but why not look at -margins-; you might also want to look at Richard Williams' write-up which you can find (margins05.pdf) here: https://www3.nd.edu/~rwilliam/stats3/index.html

[Luis Ortiz]

Many thanks, Rich...¡

That's a great source. I did know Richard William's material, but not so well concentrated

Many thanks

Luis Ortiz