jamovi

I have a question about the standardized estimates (betas) generated for dummy variables in a linear regression analysis.

I conducted the same analysis in SPSS (versions 27 and 28) and jamovi 2.2.2

Although I obtained the same results for the R2, F-statistic, unstandardized regression coefficients, and t-values, the values computed for the standardized estimates are markedly different (i.e., almost double).

NOTE: To have jamovi create the dummy variables, the nominal-level variable was placed in the FACTORS textbox when setting up the analysis.

Any advice on what may have caused the discrepancy and how to resolve it would be much appreciated.

i *think* spss standardizes the estimates for dummy variables ... so it takes the values of 0 and 1 and converts these to -0.707 and +0.707 when performing it's estimate, where as jamovi leaves this as is ... i *think*.

cheers

jonathon

Thanks for the quick response.

However, I am still concerned about the discrepancy in the values generated for the standardized estimates. Reporting and interpreting a beta as .71 is very different from reporting it as .35.

Other than creating dummy variables using the Transform function is there any way to resolve this discrepancy?

hang a sec. i don't think spss allows to you provide factors to linear regression? ... or have they added that in a more recent version? so in spss you're manually dummy coding? ... and, yes, if you dummy code the same way in jamovi, you'll get the same results.

of course, you may decide that scaling dummy variables for standardized estimates doesn't really makes sense.

cheers

jonathon

Okay thanks.

FYI (For Your Information), SPSS does have a function to Create Dummy Variables as part of its TRANSFORM options.

In health research, it is not uncommon to want to include a nominal-level variable when conducting a multiple linear regression.

oh yup. good to know. thanks.

yeah, that's why we allow the inclusion of factors.

cheers

jonathon

Hi,

I stumbled across the same thing and I'm still trying to understand the issue. For metric predictors, the standardized coefficient tells me that if the predictor variable increases by one standard deviation, the criterion variable increases by beta standard deviations.
But how do we interpret the standardized coefficients for nominal/categorical variables in the standardized coefficients column? They differ from the unstandardized version (I think I understand why: because a new model is estimated with the standardized variables - but I don't understand the interpretation of this "new" variable for nominal variables).

Thanks a lot for your help (and by the way thanks for providing the best statistics software for teaching)!

Hi
to interpret the beta's of categorical variables (in jamovi) you first need to be sure what the coding system means. This is because different coding systems define different comparisons. "Simple" (the default coding system in jamovi) compares each group with a reference group. Now, the beta associated with each comparison is the difference in the standardized Y between the two groups being compared.

Thanks a lot for your reply ! I will think that through for my example.