## How to display multinominal logit results graphically?

Do you have any suggestions for the best way to represent multinominal logit results graphically? I am using stata.

My reply: I don’t know from Stata, but here are my suggestions:

1. If the categories are unordered, break them up into a series of binary choices in a tree structure (for example, non-voter or voter, then voting for left or right, then voting for left party A or B, then voting for right party C or D). Each of these is a binary split and so can be displayed using the usual techniques for logit (as in chapters 3 and 4 of ARM).

2. If the categories are ordered, see Figure 6.4 of ARM for an example (from our analysis of storable votes).

1. Mutter says:

Check out:

Potârcă, Gina, Melinda Mills, and Laurent Lesnard. 2013. “Family Formation Trajectories in Romania, the Russian Federation, and France. Towards the Second Demographic Transition?” European Journal of Population 29(1):69-101. doi: 10.1007/s10680-012-9279-9

The multinomial logit graph in that paper was also made in Stata, likely using:

Long, J. Scott, and Jeremy Freese. 2006. Regression Models for Categorical Dependent Variables Using Stata, 2nd ed. Stata Press.

• Anonymous says:

An ungated link to the graph in the paper (possibly without the rest of the paper) would be appreciated.

2. Paul G. says:

The Margins and commands in Stata (introduced in version 12, perhaps earlier) are quite powerful and allow you to predict the probability of various outcomes, plot the probabilities by levels of factor/indictor variables and continuous variables. Try here for example: http://www.ssc.wisc.edu/sscc/pubs/stata_margins.htm

3. Robert Grant says:

I agree with Paul G the first thing to try is Stata’s own -marginsplot- command (you need v12+) which is really good and will save you lots of time. You get predicted probabilities marginalized over whatever combination of independent variables – and their interactions – you want! But if you want to pile up these curves for each of the categories on top of one another (like you sometimes see item-response people doing), Stata doesn’t do transparent graphics, so you’ll have to -serset- the lines, export them, and do it in R.

But without having really spared enough time to think it through… I wonder if you could also do some kind of biplot, feeding a predictor vs outcome contingency table into the -mca- correspondence analysis command? However, my experience of biplots is that your ‘clients’ will really struggle to understand them.