I am currently carrying out the data analysis in SPSS of a questionnaire I made for my MSc dissertation and I need some help!
In the study (N=707) I have 7 groups of respondents who were exposed to different stimuli and later evaluated the following dimension:
> "Negative Word of Mouth Intentions" (using a likert scale consisting of 3 items, each was evaluated from 1 to 7; 1 being completely disagree and 7 completely agree)
To compare the results between different groups in order to check if the diferent stimuli presented caused signficant differences in the evaluation of the previous variable mentioned I did a Kruskall Wallis test and indeed I found a statisticall significant diference among the groups. Later I run a K-W pairwise test to find which pairs of groups were significantly diferent.
My question:
1_From the 21 possible combination of pairs, only 4 pairs were different, what can I conclude from this?
2_To find which stimulus are more likely to increase negative word of mouth intentions and wich ones are more likely to decrease that variable. Can I compare all the mean ranks produced by K-W or the median of the variable for each group or I only can do this comparisson for the 4 pairs of groups wich are statistically significantly different?
Kruskal-Wallis is non-para version of ANOVA and should be interpreted similarly in terms of mean-compersions. Limewise, post-hocs should indicate significant pairwise-differences (directional).
Hope this may be of some help,
Matthew
The Kruskal–Wallis test by ranks is non-parametric analogous to one-way parametric ANOVA. So, the results emanated from the test (including mean comparions) is similar to ANOVA. If the test statistic is not significant, then it can be inferred that there is no evidence of stochastic dominance between the samples. However, in your case, as the test was found to be significant, at least one sample stochastically dominates another sample. Therefore, you might use sample contrasts between individual sample pairs, or post hoc tests using Dunn's test, which properly employs the exactly same rankings as the Kruskal–Wallis test and properly employs the pooled variance implied by the null hypothesis of the Kruskal–Wallis test in order to determine which of the sample pairs are significantly different.But it should be kept in mind when performing multiple sample tests, the type I error rate tends to become inflated, sometime raising serious concerns about multiple comparisons.
Pramit Pandit Thanks for such a complete repply I am understanding better the test.
Yes i performed a Dunn's test to compare which pairs are different, and the test also automatically does a Bonferoni correction, to compensate for the inflation of the type error 1, so I think it should be fine.
I have 4 pairs of groups which are statiscally significantly diferent, so can I compare their mean ranks or their median to draw conclusions?
- Badges
- Science topic