While working with a translated (into Lithuanian) version of the IMS (Investment Model Scale) I'm getting a sub-par Cronbach's alpha score of 0,629. The highest increase with an item deleted only goes up to 0,666.
However, when scoring each of the 4 sub scales separately I get good reliability:
,798
,826
,788
,721
I'm sure there's something fundamental I don't understand here, but at this point I'm clueless as to what to do to increase the reliability of the questionnaire. Any help would be appreciated.
Best regards.
Hey.
I meant "sub-par" as in lower then acceptable, >0,7.
It's ok though, I've already found the answer to the question in "Discovering Statistics using SPSS" by Andy Field. Basically he says that since the sub-scales measure different constructs, it's okay if on the whole they don't correlate with each other, as long as each sub-scale has an acceptable alpha score. In my case, for instance, "Quality of Alternatives" is not suppose to be correlated with "Relationship satisfaction" according to the IMS model.
Andy is correct on this a simple collapsing of each scale and cross correlation of the four subclass will show this. if you are presenting the work for publication or defending it in a thesis or viva voce you need to point this out very explicitly as some observers take the overall alpha as most important and the alphas of the scales breakdown as secondary.
This also has implications for the development of a construct as the correlational relationship has a bearing on that too.
I also agree that you should check the correlations among the sub scales, because a low correlation among some sets of them could be pulling down your overall alpha. And if you want to go a step further, you could do a factor analysis, making sure that you use an oblique rotation so that the factors can be correlated (rather than "orthogonal" or uncorrelated).
But it seems like the real issue is your overall analysis plan. Do you intend to use the four sub scales separately, or do you indeed want to sum them into a single scale?
Dear Edgrasas
1. Yes, it is possible.
2. There are several measures. Indeed, reliability measures are: Cronbach’s alpha reliability, split-half reliability and Cronbach’s alpha if item deleted. They are essential and very important measures to assess the survey design.
3. For more details, please refer to:
i. Quantitative and qualitative indicators to assess the performance of higher education institutions. Int. J. of Information and Decision sciences, 2014, Vol.6, No. 4, pp.369-392 (Inderscience).
ii. Statistical characteristics of performance indicators. Int. J. of Quality and Innovation. Vol.2, No.3-4, pp. 385-309 (Inderscience).
Regards,
Zuhair
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Applied Statistics
- Survey Design