The most dangerous form of motion is when it appears to be correlated with the effects of interest in the design matrix. In principle, you can use MANCOVA (another form of GLM) to assess whether there is evidence that estimated head motion is related to the effects of interest. spm_cva could be used for assessing this, whereby X contains the effects of interest in the design matrix, X0 contains effects of no interest and Y would contain the matrix of estimated movement parameters.
I strongly suspect that you will invariably find a statistically significant relationship between them. Even if there is no head motion, the estimated parameters are slightly influenced by the signal that you are looking for. This has been pointed out a few times in various papers of Freire, Mangin and others, with a proposed solution that involves incorporating GLM fitting into the motion correction.
Best regards,
-John