I'm usually using Bonferroni for ANOVA post hoc testing, but I was also told this test is too "demanding" and I might lose some interesting data with it.
I usually need this analysis to compare sociodemographic characeristics in neuropsychological tests performance.
Thanks!
Without claiming to be a great statistician (I'm not), I've always suspected that the Bonferroni correction is mainly a way of discouraging researchers who might otherwise be inclined to "shotgun" data sets with lots of variables. But it accomplishes this by being excessively stringent. Take the case of a correlational study with 80 cases and 10 variables (this could be something as simple as the subtests of an ability measure or the scales on an omnibus personality test). The correlation matrix will have 50 entries. Using the conventional p
Bonferroni is a correction of your post-hoc test families. You take 3 to 5 - 6 p-values and correct them to get rid of too weak significances. If you have e.g., several t-tests and some of them are close to p < 0.05 they may be falsely significant and you need to be more rigid by e.g. taking only those p-values at 0.01 into account. Tukey HSD (honestly significant difference) is good. What post-hoc tests you use may depend on your n. With small n and non-parametric test you can use Mann-Whitnay. Usually they are provided directly by the computer program you use.
Thank you Beatrice!
I'll try also Turkey. I honestly only use Scheffe or Bonferroni, and didn't know the difference between them. I'l keep in mind Mann Whitney. Thanks!!
Without claiming to be a great statistician (I'm not), I've always suspected that the Bonferroni correction is mainly a way of discouraging researchers who might otherwise be inclined to "shotgun" data sets with lots of variables. But it accomplishes this by being excessively stringent. Take the case of a correlational study with 80 cases and 10 variables (this could be something as simple as the subtests of an ability measure or the scales on an omnibus personality test). The correlation matrix will have 50 entries. Using the conventional p
Dear Stephen,
Thanks a lot for your extended answer. It was very helpful! I'll keep that in mind. Thank you for your time!
let me support the good suggestions by Stephen Joy, yet endeavouring to act as a clinical gerontologist, not being [unfortunately] a statistician
best regards, Mauro Colombo
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA