Linear Regression in jamovi - factors & intercepts

Discuss the jamovi platform, possible improvements, etc.

by bekhanaro54 » Tue Oct 27, 2020 10:07 am

I'm teaching a course on HR analytics, and I'm having my students use jamovi to replicate SPSS analysis examples from a textbook. In the data sets that accompany the textbook examples, many of the categorical variables have been recoded to numerical (e.g. 1=male, 2=female).

If you are conducting a linear regression in SPSS, you only have the option to add all of your predictors into the Independent Variables box. However, jamovi gives you the option to divide your predictors into either covariates (numerical) or factors (categorical). One of my students found in jamovi that if you drag all variables, categorical or otherwise, into the covariates box, the output is identical to the SPSS examples, but if you add the recoded cat variables to the factors box, the model intercept shrinks.

I'm pretty sure this has something to do with the fact that factors in jamovi are automatically dummy coded with a reference category, but I don't quite understand the mechanism that causes this. Any insight is appreciated, thanks!
Last edited by bekhanaro54 on Wed Oct 28, 2020 9:56 am, edited 1 time in total.
Posts: 1
Joined: Tue Oct 27, 2020 10:00 am

by jonathon » Tue Oct 27, 2020 10:28 am

yup, so depending on the coding, the levels of the factors are coded in different ways. here's a description of different coding:

so 1=male, and 2=female is a different way of coding it ... (and it doesn't feel like a correct way of coding it either, but perhaps someone else will chime in).

User avatar
Posts: 1474
Joined: Fri Jan 27, 2017 10:04 am

Return to General