The question regarding the consistency of parameter magnitudes in the GLM model.

When configuring a GLM model in AFNI, I've noticed a significant difference in the scaling between regression term parameters and head motion parameters. As a result, when examining the 'x1d' file using '1dplot,' I observe minimal fluctuations in the head motion curve. Could this discrepancy potentially impact the statistical outcomes?

Thank you so much!

Note that the plots will scale individually using "1dplot -sepscl".

But to your question, exactly which regressor is scaling from -220 to 225? Is that regressor an amplitude modulator?

In general, it is indeed preferable to keep regressors somewhat close to unit-scaled.

I would not expect this scaling to have a big impact, but especially if it went towards the millions, it probably would make the matrix less stable, even though the results would be theoretically identical (with betas scaled reciprocally to regressors). The cost of large scaling will probably be some decimals of accuracy in the betas. It might take a somewhat bad matrix to affect the betas by a large fraction, depending on the number of regressors and such.

  • rick

I apologize for the delay in remembering this question. As you mentioned, the final outcome was indeed not significantly affected. Thank you for your response! :pray::star2: