Length of trange resulting in "inaccurate" results

[shann...@gmail.com]

Hi everyone,

I've been playing around with different baseline and task lengths to understand how or if it changes my outputs, sometimes I get this error for some subjects and get sharp artifacts within my outputs (either during the baseline portion or the task):

Design matrix is poorly scaled...(RCond=8.698193e-23)
Warning: Matrix is close to singular or badly scaled. Results may be inaccurate. RCOND =  8.698193e-23.

What does it mean when it says "results may be inaccurate"? How do I check its accuracy?

For context:

- I have at least 30 seconds standing baseline and 30 seconds of walking.

- I'm using hmrDeconvHRF_DriftSS

- I've tried tranges of: -20 40, -20 35, -15 40, -15 35, -10 40, -10 35, -5 40, -5 35

- The error typically comes up for at least one subject using any of these ranges or the outputs are empty for one subject. I'm trying to find a trange that could work for all subjects but am concerned about the "results may be inaccurate" warning.

[shann...@gmail.com]

I'm trying to convert my analysis to Homer3 (was previously running everything in Homer2) and this time I get an error which doesn't finish running the GLM:

Design matrix is VERY poorly scaled...(RCond=5.172345e-26), cannot perform computation

Does anyone know what this means and what I need to do to run my data? I've tried multiple variations of the trange and none of them work.

Thanks!

[Boas, David]

This can happen for several reasons. It means that in the GLM that estimation of at least one of the regressors is ill-conditioned and your estimation of the coefficient for that regression will be highly susceptible to noise.

 

Looking at the screen of a run and the parameters (thanks for including that!) I have two thoughts:

1. Do you have short separation measurements? Your GLM is trying to do a short separation regression. It looks like you have short separation measurements, so all should be fine here.
2. Maybe more likely is that your stim seem to be presented in a periodic fashion, i.e. the same time delay between stim onset, at least for groupings of different stim conditions. If you have a fixed interval between stim onsets and that interval is less than tRange for the HRF, then you will definitely get this message about a “poorly scaled design matrix”.

--
You received this message because you are subscribed to the Google Groups "Homer-fNIRS and AtlasViewer Community" group.
To unsubscribe from this group and stop receiving emails from it, send an email to homer-fnirs...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/homer-fnirs/e4ae9598-fefc-4980-969b-176b42d1b2een%40googlegroups.com.

[shann...@gmail.com]

Hi David,

Thanks for your initial thoughts.

1. Yes, I have short separation channels (8 of them)

2. I have a random interstimulus interval of 40-120 sec between trials of similar tasks and even longer between conditions. So the tranges should not be overlapping. Also, even with a trange of -2 to 20, I get this error. (30sec task length)

What else could be happening?

Shannon