Skip to content

An in-progress repository to create a method of visualization for multinomial logistic regression results. Mlogit results are notoriously difficult to interpret, so this program I am creating will take the predicted probabilities for variables and use hierarchical clustering and heatmaps to categorize variables that have similar effects on the o…

kelseygonzalez/mlogitviz

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

20 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

mlogitviz

An in-progress repository to create a method of visualization for multinomial logistic regression results.

Multinomial logistic regression models are notoriously difficult to substantively interpret due to the multiple layers of reference categories. This paper addresses this difficulty often found in social science research by creating a visualization technique for multinomial logit models inspired by blockmodeling using the R programming language. Using the base multinomial logistic regression model results, this method mines the predicted probabilities of each variable level. I employ a method of quasi-blockmodeling through agglomerative hierarchical clustering to sort predictors by similarity of effects on the dependent variable. The sorted predicted probabilities are visualized using heatmaps which display the deviation from the mean for each normalized column. The resulting methodological technique aims to help social science researchers extrapolate general trends and uncover relationships within the context of multinomial logistic regression instead of becoming too enmeshed in the multi-layered reference categories and p-value significance.

Original multinomial logistic regression output: Table of mlogit model

mlogitviz output: mlogitviz

About

An in-progress repository to create a method of visualization for multinomial logistic regression results. Mlogit results are notoriously difficult to interpret, so this program I am creating will take the predicted probabilities for variables and use hierarchical clustering and heatmaps to categorize variables that have similar effects on the o…

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages