0todd0000
commented
2 years ago For normal CIs you can use:
- spm1d.stats.ci_onesample
- spm1d.stats.ci_pairedsample
- spm1d.stats.ci_twosample
There are also nonparametric versions:
- spm1d.stats.nonparam.ci_onesample
- spm1d.stats.nonparam.ci_pairedsample
- spm1d.stats.nonparam.ci_twosample
See more details here
You should also be able to construct CIs for t contrasts from a GLM. You'd just need to transform the t-value and the SPM critical t-value back to the data units. This is not directly supported in spm1d so you'd need to calculate the transformation that is appropriate for your particular GLM.
Hi Todd and spm1d support team,
We have been using spm1d to determine differences between groups using both a t-test and GLM with covariates. I often get comments from co-authors about including the other two parts of the ‘holy trinity’ as per current convention for discrete statistics: the effect size and measures of precision. I understand from previous postings that the SPM graphs are best for effect size (issue 48). For measures of precision, I read Mark’s posting (issue 143) advocating use of spm1d's CI functionality when using an ANOVA, which would work with a t-test, or instead of a t-test, but is there anyway to calculate confidence intervals for the difference between groups for a GLM using spm1d? Or can the CI just be calculated from the mean difference between the groups over the suprathreshold range as an estimate even though a suprathreshold range doesn't necessarily determine a local effect?
Thanks for your help
Trevor