4

I'm conducting a 2x3 mixed between-within subjects ANOVA (group x time). I have 20 participants, 10 for each group, who undergo all the 3 measures over time.

Levene's test of equality of variances is not significant for both time1 and time3. Unfortunately, on time 2, group1 (M = 2,869, SD = 1,571) and group2 (M = 6,706, SD = 3,521) have variances significantly not homogeneous (p = 0.016). Is this too bad? Can I trust the ANOVA results?

Thank you!

Federico
  • 153

1 Answers1

4

1) Moderate departure from homogeneity of variance assumption is not a threat to ANOVA unless group sizes are very unequal and/or variance positively correlates with level (i.e. a group with larger variance has also greater mean). You can check the latter in SPSS' Explore procedure ("spread vs level plot") and cure the problem - if observed - by appropriate power transformation, also suggested by the procedure.

2) If you are going to use Linear Mixed procedure rather than ANOVA, please note that it is more flexible and generally does not require homogeneity of variances.

ttnphns
  • 57,480
  • 49
  • 284
  • 501
  • 3
    By reviewing a suggested edit for this post, I noticed the following request: "a citation/reference justifying these would be great if anyone could provide them". – chl Oct 28 '13 at 08:03