Bonferroni Adjustment
By
jen ripley
Created 3 years ago
Duration 0:09:24
32
Statistics Psychology. How to do a Bonferroni Adjustment for Planned Comparisons with Grouped Data

Embed

Social
Select the file type you wish to download
Slide Content
Tags:
Psychology

Slide 1  Bonferroni Adjustment
 A technique to manage Type I error in planned comparisons

Slide 2  What is it?
 The often overlooked statistic to control for Type I error for PLANNED comparisons
 If conduct several comparisons with same dataset then chances of Type I error increase (similar to running lots of T tests)
 Change the acceptable alpha (p) value for a result to be significant
 Follow a formula
 There are actually a few formulas all called “Bonferroni Correction” if you look on the internet. Some of that is the type of application of stats (math, sociology, psychology, etc.) has different standards based on the types of analyses conducted in those fields.
 For our class I will focus on a single formula…wait for it….

Slide 3  Not for post hoc tests
 Not for Scheffe, tukey, Bonferroni posthocs (what we’ve been doing in labs)
 Post hoc ASSUMES a lower bar of stringency in the data analysis

Slide 4  Steps
 Determine if your study needs to modify the p (alpha) value to control for Type I error for PLANNED Comparisons
 If running more comparisons than levels of the IV/s
 If oneway ANOVA it’s fairly easy. How many levels of the IV?
 If 3 levels (groups). So can do (31) = 2 comparisons without adjustment
 If 4 levels. (41) = 3 comparisons without adjustment
 So if a 2 X 2 ANOVA, that’s 4 mean scores, can only do (41) 3 comparisons without adjustment
 If a 2 X 3 ANOVA, can only do (61)= 5 comparisons without adjustment
 3 X 3 ANOVA, (?1)
 Determine how many comparisons you will be doing

Slide 5  Example of Steps to conduct
 If Dr. Williams had 6 different conditions in experiment
 Can do 5 comparisons without needing to correct alpha (stick with p<.05)
 Otherwise alpha = (# conditions 1) (.05) / number of comparisons you want to do
 If Dr. Williams wants to do 10 comparisons
 (61) (.05)/10 = .025. Should use .025 as alpha test for significance (from Keppel Text)
 Sorry, have to do this one by hand.

Slide 6  Another example
 Dr Taylor examining a 2 X 3 ANCOVA
 Wants to do 7 planned comparisons
 What is the alpha level?
 Answer: (61)(.05)/7 = .04 (rounded)