Controlling for type-1 error

We have conducted an experiment that asks respondents to rate whether they thought an article was biased on an 11-point scale, with 0 being biased to one side, 6, not biased at all and 11, biased against the other side.

We used two organizations, each that supports different sides of the issue covered in the news article.

We altered in separate manipulations using a fully crossed design, the source of the article (in group, out group or neutral source) and the slant of the article's content (favoring or opposing each group and neutral). We analyze each organization separately using ANOVA, and when we make comparison's across conditions , we adjust for multiple comparisons using the Sidak correction.

However, within each group we use t-tests to see whether for each condition the rating of bias is different from the midpoint, which indicates bias. So, for each group, we do 9 such t-tests in groups of three. Within out-group source we compare the against-slant, no slant and for-slant against the midpoint--none of these are compared against each other. We do the same within neutral source and for in-group source.

The question is: do we have to make corrections to control type 1 error for the t-tests, and if we do so, do we treat each source group separately, or do we correct for all 9 together?

Thank you in advance for the help!


Less is more. Stay pure. Stay poor.
Lost me here, " we do 9 such t-tests in groups of three". Please provide more details.

You don't necessarily need to conduct all of this tests, why don't you just report the mean values with 95% confidence intervals. That way you don't have to keep changing you comparison values in the one-sample t-tests. A person would just look to see if interval included 6 or other values.