Cron Bach value was 0.04

Hi Parami,

it is vital that you understand what a (common) factor model is and its difference to principal components analysis:

Fabrigar, L. R., Wegener, D. T., MacCallum, R. C., Strahan, E. J., MacCllum, R., & Strahan, E. J. (1999). Evaluating the use of exploratory factor analysis in psychological research. Psychological Methods, 4(3), 272-299. doi:10.1037/1082-989X.4.3.272

Preacher, K. J., & MacCallum, R. C. (2003). Repairing tom swift's electric factor analysis machine. Understanding Statistics, 2, 13-43.

In this regard, when your set of measures intends to contain congeneric measures of the same underlying latent variable, THEN you are in the realm of factor analysis. If the set is rather a mixture of things and intens to reflect some kind of domain or breadth, then the result will be an index and application of factor analysis would be the wrong tool.

Next, understand that that the application any measure of internal consistency (like alpha) requires the validity of the factor model. Otherwise, you may have a bunch of items measuring different things that--although they are all reliable, still would result in a bad alpha estimate (as they don't have to correlate strongly with each other).

In addition, if your factor model is valid, then using McDonalds omega would be a better choice than alpha, as it has weaker assumptions:

McNeish, D. (2018). Thanks coefficient alpha, we’ll take it from here. Psychological Methods, 23(3), 412.

Hope that helps,

Holger

Thank you so much, Prof Holger. Now it is clear to me that i need to do the factor analysis first and assess the validity of the factor model. I will then test for internal consistency. I will get back to you with new test results. Thanks again

If, indeed, the Alpha coefficient is 0.04 (or even 0.4), I'd stop everything and try to find out why it is so low. Is it an error in calculating the coefficient? Is the questionnaire too heterogeneous, trying to measure too many unrelated concepts? With a coefficient that low I do not see any reason to continue the study without major revision. Good luck.

I am going to see the exact reason. May be the unrelated concepts as I have had 5 domains related to primary care quality. Thank you so much for the suggestion

In my opinion Cronbach Alpha should be used for validation. It is direct and can be explained easily.

The previous points are quite apt. I would add that alpha increases with the total number of items and with their average intercorrelation. 26 is a lot of items for any scale. With that many items, I would expect an alpha above .4 (and certainly well above .04), given even a very modest average intercorrelation. The low alpha value suggests to me that some subsets of your 26 items might be phrased in the opposite direction from others, and/or, some of the subdimensions indicated by your factor analysis correspond to theoretical constructs that are expected to correlate negatively with other subdimensions.

If the empirical 26-item correlation matrix includes both positive and negative correlations, then the average intercorrelation could be close to zero, thus producing a low alpha. If so, then you need to reverse-score some of the items, so that a high score on every item corresponds to a high value of the construct that you think it measures, and also exclude from your alpha analysis any items that you think measure a different construct.

Thank you so much. My tool is for assessment of patient perceived primary care quality. It was a challenge indeed to restrict the items to 6 domains and 26 questions. I will follow your suggestion. This helps me a lot.

Most importantly, I have some suggestions for you; 1) Doing second pilot test is not necessary. 2) Calculating Cronbach Alpha, Item-Total Correlation, Commonality coefficients, etc., should come first and Factor Analysis (FA) should come after them. 3) As mentioned above, factor analysis (FA) is in principle different from principal components analysis (PCA). Regarding psychometric instrumentations, the items that are created in terms of the Likert type scale should be considered as ordinal variables when calculating or analyzing dimensionality or factorial structure. 4) Multivariate normality test is so important to be done before running Exploratory Factor Analysis (EFA), and in most cases, Likert type-based items by their ordinality cause the dataset distribution's deviation from multivariate normality and thus, PCA is not appropriate extraction method at all to do EFA. 5) One important point is that you should estimate the number of (dataset or instrument) factors and dimensions by means of Parallel Analysis (PA), Dimensionality Assessment Test, etc., before conducting factor analysis test (EFA), and do not be sure about PCA results about factor number.....

There are many references that need to be mentioned here, but to make this long text short, you can find the citations or references as well as full description of all the points described above in my first published article (link below). when conducting the project of Master thesis, I faced the problems you are now dealing with..

Preprint Dimensional Structure and Psychometric Properties of the Per...

https://www.researchsquare.com/article/rs-3126417/latest

Thank you so much for your insight. It is a relief that I don't need to do another pilot test. Actually this tool is for my M Phil in Family Medicine.