stepwise regression

Discuss statistics related things

by marshall » Sat Jul 13, 2019 10:47 am

Hi,
Is there a way to run a stepwise regression analysis in Jamovi?
Or perhaps, is there a way to generate the semi-partial correlations in Jamovi?
Thank you.
marshall
 
Posts: 1
Joined: Sat Jul 13, 2019 10:43 am

by jonathon » Mon Jul 15, 2019 1:37 am

hi,

i think we're mostly opposed to stepwise regression - we don't really find it to be a principled approach, and think model selection using things like BIC, AIC, etc. is a better approach. so it's not something we've implemented (although, its certainly something which could be provided by a module).

wrt semi-partial correlations, i'm not crystal clear on that, so i'll leave that for ravi to chime in.

cheers

jonathon
User avatar
jonathon
 
Posts: 899
Joined: Fri Jan 27, 2017 10:04 am

by coledavis » Mon Aug 26, 2019 9:54 am

I cover some of this in my books. In short, I would say that in these days of easy computing, I would agree that the use of stepwise regression for the usual purposes is not particularly ethical. Even if you wanted to use it for whittling down data, wouldn't factor analysis be more useful and valid.

However, there is the application for sensitivity analysis: sources of uncertainty may be analysed by use of different parameters.Another possibility is some combination with hierarchical regression, whittling down theoretically unclear variables before subjecting surviving variables to blocks.

Do note that I have used the term 'stepwise' in the general way it has been introduced. In fact, 'stepwise' (or 'bidirectional elimination') is one of three general techniques within sequential regression, the other main categories being backward elimination and forward selection
coledavis
 
Posts: 19
Joined: Sun Jul 02, 2017 1:35 pm


Return to Statistics