2
$\begingroup$

I have 1 continuous DV and 1 categorical IV with 4 levels. I set up orthogonal dummy coding for 3 planned contrasts. I then used the linear model to observe if the planned contrasts are significant.

My question is, do I have to use a Bonferroni correction in this instance? The sites I've read says that it is necessary when conducting multiple comparisons using t-test but as I'm using the linear model should I still correct the alpha?

Many thanks!

Improve this question
$\endgroup$
1
  • $\begingroup$ You are using a linear model and t-tests. The fact that you did an anova first wouldn't get you out of the problem that the following t-tests will, in most cases as there might be exceptions, require corrections for multiple comparisons. $\endgroup$
    – David Ernst
    Commented Nov 4, 2017 at 14:24

2 Answers 2

Reset to default
2
$\begingroup$

If you planned to do them in advance for scientific reasons then there is no reason to adjust for multiplicity. The form of test you were using is not relevant. If, on the other hand, you indulged in data-driven exploration of your data then you might want to consider adjusting for multiplicity.

Improve this answer
$\endgroup$
3
  • 1
    $\begingroup$ If you do thousands of comparisons on a high dimensional dataset, I would suggest correcting the p-value, whether you planned the comparisons in advance or not. $\endgroup$
    – JAD
    Commented Nov 9, 2016 at 21:19
  • 1
    $\begingroup$ @JarkoDubbeldam I agree entirely, I thought that was the implication of my answer but perhaps I was not clear enough. $\endgroup$
    – mdewey
    Commented Nov 10, 2016 at 6:54
  • $\begingroup$ "Planning" your tests doesn't get you out of the problem of multiple comparisons. The error rate is inflated by 3 planned tests that are uncorrected for multiple comparisons just as much as the error rate is inflated by 3 UNplanned tests that are uncorrected for multiple comparisons. $\endgroup$
    – Bonferroni
    Commented Mar 5, 2017 at 10:26
0
$\begingroup$

Yes, it is considered best practice to use some correction to control family-wise alpha whenever multiple tests are performed. This isn't a consideration specific to t-tests.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.