Violation of normality assumption for One Way Anova - Non parametric test?

BOAZ

New Member
Hoping somebody can help with this?

Recently we conducted a small study with ten participants. Each participant completed the same two tests, 'Test A' and 'Test B', during one session. Five particpants (Group AB) performed test A first, while the remainder performed test B first (Group BA). The goals were:

1) To see if the two tests were equivalent
2) To establish whether the order of the tests had any effect on the relative scores on each test.

Goal 1
The combined data failed the Levenes test for homogeneity of variance (P = .028), and Test B data failed the Shapiro Wilk test for normaility (P = .008). A Wilcoxin Signed Ranks test revealed no significant difference between tests A & B (Z = -.634, P = .536)

Goal 2
Achieving goal two is a problem. We have used a One Way ANOVA to establish that group (i.e. AB or BA) was not a significant factor for either test (P =.496, and P =.436). However, while the data for both tests A and B passed Levene's tests, only group AB scores on test A passed a Shapiro Wilk test. We are unsure of which non parametric test to use as the Kruskal-Wallis ANOVA by ranks is only suitable for 3+ groups. It may be the case that we do not really know how to set this up properley in SPSS!

Any help would be much appreciated!

d21e7x11

New Member
Hi BOAZ,

Few notes/questions:
- I believe in 1) you meant the goal is to test if the tests A and B are different. Correct?
- each participant completed both tests - were this accounted for in the analysis you've described?
- since each participant had both tests, you may want to consider doind analysis of diferences (outcome) and the type of difference i.e. (A-B) or (B-A) as a factor. In other words, you can do one-way ANOVA on the differences testing if i) the difference is 0; ii) there is no difference between AB and BA.

BOAZ

New Member
Thanks d,

I response to your questions
- I believe in 1) you meant the goal is to test if the tests A and B are different. Correct? - each participant completed both tests - were this accounted for in the analysis you've described?
Correct and yes. I treated the data as paired, rather than independent.

- since each participant had both tests, you may want to consider doind analysis of diferences (outcome) and the type of difference i.e. (A-B) or (B-A) as a factor. In other words, you can do one-way ANOVA on the differences testing if i) the difference is 0; ii) there is no difference between AB and BA.[/QUOTE]
OKay I will calculate the differences scores this evening and see i fthe resultign data is normal. Although, if I only have two sets fo difference scores (A-B) and (B-A) wouldn't this jsut be a t-test or U test? I am not sure how to set up the data you have described for an ANOVA.

Thanks again, BOAZ