| Message/Author |
|
|
| Syed Ali posted on Monday, September 28, 2020 - 9:38 am
|
|
|
We are trying to compare the effect of two predictors and we are trying to find a proper statistical test to test the difference between two predictors.
In a ML regression model what is the best way to test the difference between two beta coefficients from two different continuous scales where x1 ranges 0-20 and x2 300-500, and the predictor is on a likert scale 1-4. |
|
|
|
|
| Because of the different X variable scales, it seems like you would need to test that the standardized slopes are equal which you can do via Model Constraint. |
|
|
| Syed Ali posted on Monday, October 05, 2020 - 10:47 am
|
|
|
I have seen criticism that using standard betas is too influenced by the variance of the X variable and so we should NOT use the unstandardized coefficients.
I want to answer the question which effect is greater between x1 and x2. Another approach suggests to test the difference through the 95% Confidence Intervals via bias corrected bootstrap. Please see this video for reference : https://www.youtube.com/watch?v=NnqCp5CjTwk
In this short 6 minute video they talk about a more precise way to do this in SPSS Amos towards the end (at time index 6:10) following the procedures by Cumming (2009). Is there something similar in mPlus? |
|
|
|
|
| The critique of standardized slopes is mostly when you compare across groups - where there can be group differences in variances that distort (whereas unstandardized slopes are not so affected). Without looking at this video in whole, I think he is talking about non-symmetric (bootstrapped) confidence intervals for the difference between standardized slopes. You can get these using Model Constraint where you express the standardized slopes and their differences. |
|
|
| Back to top |
