Overall interpretation of multinomial ordinal logistic regression model, and how to improve with more variables bit by bit

10 ビュー (過去 30 日間)
Dhruv Ghulati
Dhruv Ghulati 2015 年 12 月 24 日
I have 3 categories to predict which are ordinal - Dead, Inactive, and Highly Active.
I tried MATLAB's logistic regression with 2 explanatory variables (my aim is to keep building up variables bit by bit to get the strongest model, with the correct variables).
I assumed my response variable was ordinal (3 categories to predict - totally dead, inactive, and highly active) - http://uk.mathworks.com/help/stats/mnrfit.html#btpyj65
y = ordinal(convertednbs.activity,{'Dead','Inactive','Highly Active'});
[B,dev,stats] = mnrfit([convertednbs.artists_paying_for convertednbs.bookmarks]...
pvalues = stats.p;
interpretation = [B(1:2)'; repmat(B(3:end),1,2)];
I get the following for my stats.p (p values for the 2 predictors bookmarks and artists paying for to predict the ordinal "activity" variable):
Does this mean that artists paying for and bookmarks is significant, given that the p values (bottom 2) are tiny? The coefficients that come out are (only bottom two rows, the others are coefficients, but don't understand the matlab explanation to be honest):
-3.84122521578429 2.44911025915193
-0.178107681649582 -0.178107681649582
-0.321731337655824 -0.321731337655824
Am I using the MATLAB logistic function the right way? How can I continue my method to one by one build up a model? There doesn't seem to be a way that shows "I am doing better than a 50/50 coin flip for each variable I include in the model, for predicting inactive, dead and active".

回答 (0 件)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by