This page attempts to answer some frequently asked questions about the "% signal change" script, roi_percent.m, in the GablabToolbox.
For general questions about percent signal change as a measure, check out PercentSignalChangeFaq.
You might also check out RoiDeconvolve for answers to common questions about the "sister" script to roi_percent. The two are used for exactly the same thing, but are best at differing types of experimental designs.
The full text documentation for this program can be found here : percent_signal_change_readme.txt
When running roi_percent, the program pops out 4 files, one of which is a .txt file with four rows (pct, std, conint, and Y). What do all of these columns represent? How is the Std calculated, and how would one convert that to standard error values?
The .txt file you're talking about is the percent_signal_condition.txt file. The output in this file is intended to allow you to compare percent signal responses to different conditions in your experiment. Each condition in your experiment should have its own section in the file, with the first row of each section being the name of your condition. The basic output is the pct row, which is a timecourse of percent signal change values. The timecourse should be 32 seconds worth of TRs long (varies depending on your TR length). That timecourse is the average percent signal change response in that region following the onset of that condition, averaged across all onsets of the condition in the whole experiment. It's also called a time-locked average, or a peristimulus timecourse. The idea is that you can plot this to look at the hemodynamic response timecourse for a particular condition.
The other rows operate on the same timescale, but provide different measures on that time axis. std is the standard deviation of the percent signal change at that peristimulus timepoint across all occurrences of that condition; conint is 1.96 times the standard error for that timepoint, or one-half the 95% error bar value. Y is a peristimulus timecourse of the scaled intensities that the percent signal change is calculated from; depending on how you chose your temporal filtering options, the absolute numbers may vary widely here.
But my trials are only 10 seconds long. What's the rest of that timecourse mean?
Not much. The timecourse that's output covers 32 seconds no matter what, because that's the length of the canonical HRF in SPM. But if your trials are shorter than that, the remainder of that timecourse is likely to be a mishmash of responses to whatever trials follow the condition you're looking at.