@Elena For establishing the validity of your variables, you can use Cronbach's Alpha on variable wise items (check reliability of items together meant for measuring one variable) if you're using validated scales for your questionnaire. Then compute composite variables from respective items (IVs and DV). Now you can run liner regression analysis provided your data, after reliability testing, fulfills the assumptions of regression analysis. Refer to any book on multivariate analysis for assumptions of linear regression.

Hello Elena,

The first order of business is addressing the dimensionality of each of your target variables. Having multiple questions relating to the same attribute/charcteristic/stimulus doesn't guarantee that all will, in fact, be measuring the same common factor. Examining internal consistency reliability is certainly one step towards that inspection, but not the only one. Perhaps a confirmatory factor analysis would be in order. If the adherence of items to your target factors was good, then you could use the estimated factor scores as the variables for your analysis.

With respect to estimating Cronbach's alpha, you can't estimate it for a single item/question from your instrument. You have to look at a set of items (at least two!).

For your regression analyses, you'd have better variables by using composite scores, regardless of whether you got them by: (a) summing the relevant item scores together; (b) averaging the relevant item scores; (c) obtaining factor scores from the factor analysis; or (d) using item response theory to obtain scores for each target variable.

Good luck with your work, and do post a follow-up when you've finished your disseration!

Dear David Morse ,

Thank you very much for your reply! It was very clear and a useful trail to follow.

Yes, I am considering the Cronbach alpha first and then the CFA on the factors that proved reliable in the alpha.

It is really useful to know I can use multiple ways to merge the questions for each variable in the regression, thank you!

I have a couple more questions (perhaps silly).

1. Do you think I need to conduct correlation before the regression as well? Or after Alpha and CFA can I go directly into the regressions?

2. I have negative worded questions (to avoid social desirability bias), I am pretty sure for those I need to reverse the scale scoring too, correct?

Thank you again for your help!

Elena

Hello Elena,

The specific statistical analysis steps depend on the tool (regression, AMOS, SmartPLS, etc.) you are using to test the hypothesis and the nature of your data. However, generally the steps are:

1. Prepare the dataset by removing invalid cases, replacing missing values, and removing outliers.

2. Ensure the measures are valid (factor analysis).

3. Ensuring scales reliability (CA/CR, convergent validity, and discriminant validity)

4. Computing descriptives and a table of correlations.

5. If using a parametric tool (regression, covariance-based SEM) test for regression assumption satisfaction.

6. Test your bivariate hypotheses.

7. Test your multivariate hypotheses.

8. Conduct any post hoc analysis desired.

9. Evaluate your findings.

Dear Kenneth W. Cromer Thank you for your reply!

I have done steps 1 to 3, preparing the data by eliminating invalid responses, checking Cronbach Alpha values to eliminate items lower than .07 and run EFA to check if all the items measure the correct variable.

Does it matter whether I do Cronbach alpha before EFA?

I have seen different journal articles from my literature review do Alpha first or EFA first. Which is the best way?

Other than that, I was now planning to run some descriptive statistics (mean and SD) and then check for regression assumptions before running the various regressions.

Thanks again,

Elena

Usually you do EFA first, that is because it confirms unidimensionality and that all factors load sufficiently on the construct. Then you do CA. The reason for this is that EFA is used to validate the items, while CA is used to validate the scale as a whole. If you delete items using CA first, you end up with a reliable scale, but you never know if the construct was multidimensional (because you may have deleted the other dimension doing CA). In the long run, it is best to avoid that risk and work the items first (EFA) then the scale as a whole (CA).

Hi Kenneth W. Cromer great thank you very much for confirming this! Now back to the drawing board and reversing the steps, good thing I saved the initial data set!