Multi-nominal regression interactions with margins and margin plots

[em lowthian]

Hi everyone,

Does anyone know how to do the margin command for multinominal regression for a certain outcome?

I'm trying to get margins to plot school year's use of energy drinks over time using repeated cross-sectional data.

The mlogit command is (I want sugar2=0 to be the reference category):

mlogit sugar2 i.schlyear##c.year gender ethnic2 ses, base(0)

margins, at(year=(1998 2000 2002 2004 2006 2009 2013 2014 2015 2017) schlyear=(1 2 3 4 5 6 7))

 marginsplot

But this gives me the margins for outcomes 0, 1 and 2. I just would like outcome 1 and 2 to be separate graphs as it is too much in one graph.

I have tried various ways of plotting margins but stata has returned errors.

e.g.

1)

margins, at(year=(1998 2000 2002 2004 2006 2009 2013 2014 2015 2017) schlyear=(1 2 3 4 5 6 7)), predict(outcome(#1))
invalid 'predict' 

2)

. margins schlyear##year, predict(outcome(#1))
factor 'year' not found in list of covariates

3)

margins i.schlyear##c.year, predict(outcome(#1))
only factor variables and their interactions are allowed

4)

 . margins schlyear year, predict(outcome(#1))
factor 'year' not found in list of covariates

P.S.
Details of variables:
sugar2 = sugary drinks (never, weekly, daily) 0=base/reference category
schlyear = school year (1-7)
year = wave of data (1998 - 2017)
gender = boy (1) girl (2)
ethnic2 = white (1) BAME (2)
ses = high (0) low (1)

Any help would be greatly appreciated! Thanks in advance.

[Clyde Schechter]

margins schlyear, at(year = (1998 2000 2002 2004 2006 2009 2013 2014 2015 2017)) predict(outcome(2))  
marginsplot

will give you the graph for outcome sugar2 = 1 (because with base category 0, 1 is the second outcome). Then do the same with -predict(outcome(3))-

[Richard Williams]

This might help.

https://www3.nd.edu/~rwilliam/stats3/Margins05.pdf

Also check out the user-written combomarginsplot. Even if you wind up not using it, the help shows how you can do things like combine satisfied and very satisfied into a single line.

[em lowthian]

Hi Clyde, Richard,

Clyde, thank you very much for providing me with this. I will try that code this morning.

Richard - that you for providing that, I use a lot of your materials and I must have not come across this one. Your materials are very helpful so thank you for providing these further.

Best wishes,

Em