Hi all,
I am looking at three measures of social problem solving.
Two of the measures are scored on the same scale with the same variables (one is an analogue task and the other is a diary card task). The third is just a self-report measure. I found that these variables generally were not correlated with each other, thinking that this means they aren't indexing the same processes. When I conducted paired sample t-tests on the outcome variables for the analogue task and diary card task, there were no differences. Wouldn't this mean that performance on those tasks then are similar? If so, why would they not be correlated? Hope this is enough information!
Hi
I think if your target is which of social problem solving among 3 groups works better. you can used to Kruskal-Wallis test, if scales are ordinal or nominal. also, partial correlation you must used for continuous-quantity parameters.
Dear Dana,
There are so many points you should think of, even in such a quite simple design. First of all, you didnt say anything about your sample size, therefore you should consider this and think about your effect sizes additionally or instead of significance testing. For example, your correlations/differences could be substantial, but due to a small sample size, nothing will show up as significant. In your case, where you didnt find any effect, you should ask yourself if the cause is maybe low power or a really small effect size.
Independent of that, correlation and mean differences are not the same (e.g. you could create a new variabel, add one unit to your original variable and would find a difference, but a correlation of 1, since correlations are not altered by linear transformations, or you create two random variables with different mean values, they have different means and are not correlated), there can be all combinations of situations: ns cor/ ns dif; sig cor/ns dif; ns cor/sig dif: sig/cor/sig dif. For example in a classic pre-post designs, you would expect sig dif/sig cor, but if the intervention did not work it would be ns dif/sig cor. In your case you could ask, why there should be a difference AND correlation? Or the other way around, if the answers of the two scales are independent (ns cor), why is there any need that they differ? And yet another question, if these scales are uncorrelated and they just happen to have the same scale, why is the difference of them of interest at all?
The fact that two variables are both non-significant merely says that each of their effects is not different from a random relationship with the dependent variable. It does not say anything about their relationship with each other.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA