Welcome to RunSubmit.com

Question

2

Hello,

Which method do you use to assess the relative strength of each driver in a multivariate logistic regression ? This is very useful to show this on a pie chart to a non modeller audience.

Kind regards,

Toloc

edited Feb 15 at 2:48

jay.l.stevens♦
783●8

asked Feb 4 at 10:07

Toloc
475●6

Mark Lamias · Answer 1 · 2010-03-08 21:43:08Z UTC

2

Why not use the standardized regression coefficients? These will assess the average change in the dependent variable for a one unit change in the given beta. This is an indicator of the impact of the dependent variable, holding all other variables constant.

answered Mar 8 at 21:43

Mark Lamias
21●1

Hello Mark, By the way, this is the method that I'm using. Some Data miners colleagues don't like this method as the result may be influenced by multicollinearity (quite scarce to my experience) and prefer the one proposed by Jay. – Toloc Mar 10 at 11:06

jay.l.stevens · Answer 2 · 2010-02-09 13:15:47Z UTC

1

@Toloc, One of the easy things to do after-the-fact might be to generate and visualize a straight correlation coefficient between the input variables and the output of the regression. Although its not perfect, it might be a good place to start and would give you at least a sense of which variables might be contributing to the final output of the model.

answered Feb 9 at 13:15

jay.l.stevens♦
783●8

	What exactly do you mean by "output of the regression"? The dependent variable? The fitted values? – Matt Parker Apr 2 at 17:08
	I mean taking the output of the logistic regression (i.e. the probability value between 0 and 1) and then using running correlations between that value and the inputs. – jay.l.stevens♦ Apr 9 at 4:27

BMiner · Answer 3 · 2010-02-10 22:45:43Z UTC

How about an odds ratio? or plot (table) the probability of Y=1 holding all other variables at their means - except the one you are showing, which you show the probability of Y=1 at meaningful values (maybe just the quartiles or 0/1 depending on the data type).

Assessing the relative strength of driver variables in a multivariate logistic regression

Remember to vote up questions/answers you find interesting or helpful (requires 15 reputation points)

3 Answers