I am analyzing survey results from three different countries. Survey utilized Likert scales and consisted of 5 constructs - independent variables such as "trust" and "social influence" (each construct consisted of 2-4 items on a 7 point scale; i.e. 16 items in total for independent constructs). I have then included questions related to a dependent variable (3 items on a 7 point scale).

I am interested in analyzing two aspects. First, I would like to know whether each construct had an effect on the dependent variable. Second, I would like to know whether people in particular countries place more emphasis on a particular construct (for instance, "trust" has a higher effect on the dependent variable in countries with low uncertainty avoidance).

I have received around 250 responses (at least 60 per country, which according to research should be enough, as I am only looking into 5 constructs).

However I am not sure how to analyze this data. I am considering running Exploratory Factor Analysis (EFA) first, to assess the validity and reliability of data (first performing KMO to assess adequacy and then computing Cronbach's Alpha matrix to possibly remove some items from any of the constructs).

  1. Does it make sense to do EFA in this particular case?
  2. Can I then create what is by some called "super variable" by combining items from each construct? Should I use the sum of the factors or the mean or the mode? I will be using SPSS for the analysis and would like to take advantage of the transform function.
  3. Can I then run a regression tests for each construct (independent variable) and the dependent variable? Does it make sense to use multivariate regression technique or non-parametric ANOVA? Is there a more appropriate technique to see construct's influence on the dependent? Is there a technique allowing me to test each construct impact on the dependent at the same time?
  4. Should I analyze the data for each country separately or used combine results from all countries (to to test first question of my paper related to whether each construct has a positive effect on the dependent variable)? Perhaps do both, i.e. separately and combined?
  5. After I have finally determined which factors (constructs) have a positive influence on the dependent variable, how can I address the cross cultural aspect of the paper? Is it correct to simply say, that for instance "trust" has a higher effect on the dependent variable in Country X, because it has a higher correlation coefficient?

I have a limited statistical knowledge and intend to use only Excel and SPSS for the analysis. I have noticed that other papers are utilizing CFA and SEM, however I am not familiar with these methods and I am not sure whether they would be appropriate in this particular case. I haven't found any comparable papers which would try assessing both, impact of various factors on the dependent and at the same time compare differences in multiple countries, therefore any help would be highly appreciated.

  • $\begingroup$ Welcome to Cross Validated! Please take a moment to view our tour. $\endgroup$
    – Tavrock
    Commented Mar 10, 2017 at 17:37
  • 1
    $\begingroup$ I count 11 questions in your (well-conceived) post. This is in the 99th percentile for posts on this site :-) You are seeking not an answer but help formulating an entire analysis plan. You may obtain a few fragmented answers here, but I bet you'd be better off with some extensive collaboration with a mentor or consultant. Good luck ~ $\endgroup$
    – rolando2
    Commented Mar 10, 2017 at 18:03
  • $\begingroup$ If your question has been answered to your satisfaction, you can accept an answer by clicking the check mark under the voting arrows. $\endgroup$ Commented Jul 18, 2017 at 16:32

1 Answer 1


1. I don't think so. Exploratory factor analysis is a method to decide on how to group a disparate collection of items. But you've already decided how to group them, into the 5 constructs you mentioned.

2. You sure can. The sum and the mean should lead to identical results, because the mean is just the sum divided by the number of items. The mode is probably a poor choice because, with just a few items on a 7-point scale, modes will often fail to be unique. You could also try to estimate individual weights for each item, using CFA, for example. The median would also be a sane choice.

3. You probably want a multiple regression model in which each construct is a separate predictor.

4, 5. If you're using multiple regression, the obvious thing to do would be to include countries as predictors, and add in country-by-construct interaction terms.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.