I used a three-subscale tool to measure the degree of exposure of children to foreign languages. I would like to know if I may reduce the items per each subscale and keep one as a representative of an exposure factor and apply it to MANOVA. I should note that the items showed strong internal consistency and reliability.
Hum, I'd say it could be quite risky in terms of misclassification bias... "internal consistency" can be quite a misleading concept (as Cronbach himself said) and as regards "reliability" it depends on how it has been measured. Moreover, in any case, the exposure given by the combination of your items might not be classified in the same way you get from a single item, regardless of how consistent/reliable/correlated your items are - and, even more important, how would you choose the "representative" one?
Why don't you use a Principal Components Analysis in order to reduce data dimensionality and then put PCA dimensions in the MANOVA? To me, it seems a better and more robust approach.
PS please note that using Likert scales as they are continuous variables might generate weird results, in particular if your Likert scales have only a few levels. Perhaps, you'd better give a try to different statistical methods, or at least use an appropriate correlation index (like Spearman, which is a rank correlation coefficient and thus is suitable also for ordinal variables, or Polychoric correlation) to compute the correlation matrix, and then running a PCA on it.
Reducing your analysis to one indicator per scale with also reduce the reliability of your measures (i.e., they will contain more random error and thus reduce the size of your correlations).
A better approach would be to do a confirmatory factor analysis (CFA) which specifies your factor structure, and then follow that up with a structural equation analysis (SEM) to replace your MANOVA.
Dear Mohamed,
If i undestood you:
Of course. You can reduce the itens number. Use, for exemple, in SPSS:
Analyse --> dimension reduction--> factor...
Select the variable that you will want "to form a package"
In the dialogue box: Extraction you need select
Fixed number of factors
factor to extract --> put the number one!
In the dialogue box: Score
you need select: "save as variable" and mark regression.
The SPSS programm will go save a new variable. it is the package with the union of select variables.
Other possility is the use "The PROCESS macro for SPSS and SAS". it is a macro that use non categorical variables in regression.
Hope this help you
Best regards
Dirceu
- Badges
- Science topic
- Similar topics
- Psychology