Hello everyone,
I am a student and conducting a research on EWOM type (namely EW1, EW2, EW3) impact on accordingly 3 main problems (Ease lack of trust - ET, Brand Trust -BT, Brand Loyalty - BL). To be more detailed:
- Independent variables: EW1, EW2, EW3
- Dependent variables: ET, BT, BL
=> My hypotheses is about the impact of each EWOM type on ET, BT, BL is different.
My survey questions are designed using Likert-scale and I am considering Kruskal-Wallis test to analyze the data on PSPP platform. So, my plan is to have 3 different independent observation in which I will exert each EWOM type information (Group 1: EW1, Group 2: EW2, Group 3: EW3) and ask them the same questions set regarding ET, BT, BL accordingly. Later on, for the data analyze stage using Kruskal-Wallis test , I will allocate the data again as follow to find out the answer for my hypotheses:
- ET- (EW1, EW2, EW3)
- BT- (EW1, EW2, EW3)
- BL- (EW1, EW2, EW3)
However, I am confused since for each dependent variable (For example: ET), I have around 5 items (questions) (using Likert-scale) while most of the case using Kruskal-Wallis test is just about one item (question) and run the test. I want to ask in my case, Is it still possible to run Kruskal-Wallis test. If yes, can you please explain me how or Do you have any materials for instruction? And If no, Do you have any suggestion for my case?
Thank you so much!
Hello Chi,
If I understand your query, you have 1 independent variable (EWOM type, having three levels, EW1, EW2, EW3, which you'd like to treat as groups) and three dependent variables: an ET score, a BT score, and a BL score. You'd like to compare the three groups for equality on the three scores.
First, to create the scores, you could: (a) sum or average responses across the individual items purported to contribute to each score; or (b) run a factor analysis of the full item set (most likely with polychoric correlations given the Likert-type scale), then estimate factor scores for each respondent on each factor (assuming that a 3-factor solution that represents ET, BT, and BL is supported by your data).
In either case, a one-way anova type comparison could follow, if you wish to evaluate the scores individually. Otherwise, a one-way manova type comparison would be preferred.
If you're worried about whether the scores are of interval strength (e.g., you've chosen path "a" from above), you could apply either a hybrid ordinal/interval method, or apply a nonparametric (m)anova. See references below.
For path "b", ordinary (m)anova would likely work. Of course, you should check the data set for conformance to the usual assumptions.
Good luck with your work.
Ordinal/Interval hybrid methods:
Puri, M. L., & Sen, P. K. (1969). A class of rank-order tests for a general linear hypothesis. The Annals of Mathematical Statistics, 40(4), 1325-1343. Available: https://projecteuclid.org/download/pdf_1/euclid.aoms/1177697505)
Conover, W. J., & Iman, R. L. (1981). Rank transformations as a bridge between parametric and nonparametric statistics. The American Statistician, 35(3), 124-129.
Nonparametric (m)anova:
Hettmansperger, T.P. (1984). Statistical inference
based on ranks. New York: Wiley. (See chapter 6, pp. 279-296.)
S code (which usually runs in R as well) for these methods:
Sheu, C-F. & O'Curry, S. (1996).
Implementation of nonparametric multivariate statistics with S.
Behavior Research Methods, Instruments & Computers, 28(2), 315-318.
You may read the assumptions at: https://www.statisticshowto.datasciencecentral.com/kruskal-wallis/
For each variable (ET, BT,....) you are using 5-point multi-item scales; calculation reliability using Cronbach alpha is the first step. then you may run EFA or sum up the cores (calculate mean/median) and use ANOVA to test equality of means.
- Badges
- Science topic