Using a quality of life measure, scores of three different domains were obtained for one population (physical, social and psychological domains). What is the best way to compare the means between the three domains to see if there is any statistical significant difference?
Can the paired sample t-test or the repeated measure ANOVA be used for such situation?
One sample x 3 variables (one time?). It seems to be One-way ANOVA or the Main effects ANOVA
Many thanks for your reply. Yes the 3 variables were assessed one time at the same time. So a One-way ANOVA would be valid for this?
Yep! Seems to be the right approach. Let’s see if someone else will confirm...
Hi Motasum Abu-Awwd, I would agree with Professor Bianco, and suggest you follow his sound advice on the matter - one-way ANOVA, and should the result be significant, follow up with a post-hoc test (I personally like Tukey). All the best, Cain
Your replies are much appreciated! I will follow Prof Bianco's advice. Thank you all very much for being so helpful.
Be careful! It is quite hard to compare the physical, social and psychological domains. Could you please provide us the variables? What did you measure? It is important this, I do not want to be responsible for some scientific shame... ahahah
I agree with Prof @Antonino Bianco, it should depend on which kind of variable you are talking about.
If you choose to use paired t test for three means, you have to run the test 3 times: e.g., 1-2 pair, 1-3 pair and 2-3 pair. This will increase the type I error. And if you choose repeated measure ANOVA, you have to have 3 or more than three follow ups of the same case. So, it will be better to go to one-way or two-way or three-way or four-way ANOVA based on number of independent variables.
Dear Awwad
Of course you cannot campare different type of variable... if you want to campare phisical, social or... in multi-time measurement you should use multivariate and ofcourse the data are not independente. Surly you cannot compare different type of variable together.
For three measures from the same respondent, please use repeated measures Analysis of Variance, which I assume is what Professor Bianco meant when he suggested one-way ANOVA. If the assumptions regarding normality and sphericity are not met, you can use Friedman's nonparametric ANOVA. For traditional ANOVA, you can use single df tests for post hoc analyses or use repeated-measures (paired) t test. If you use Friedman's ANOVA, you may use Wilcoxen t for post hoc analyses.
You can use Friedman two-way analysis of variance if your groups are small (n< 30) but you use related samples (the same persons).
I am more of a methods person than psychometric, but my understanding is what is wanted is a T affiliated solution to a multiple comparison problem.
Probably this ship has sailed, but wouldn't the easiest T affiliated fix be a Bonferroni adjustment for the number of planned comparisons? Assuming assumptions are reasonably met, you could make the adjustment on your dependent T. You might not get the power of other procedures (e.g.Tukey HSD) but if you are just doing paired comparisons it would let you use the T and, if the effect size was sufficient, should catch significance. (e.g. https://stats.stackexchange.com/questions/56980/correcting-for-type-1-error-in-multiple-paired-t-tests)
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA