I am a student doing my master's thesis and I have a question regarding my study. I am working with country data for 25 countries and I am looking into cultural values, attitudes and sociodemographics as predictors of environmental behavior.

For the dependent variable environmental behavior I would like to combine several aspects (Generation of waste, waste treatment, GHG emissions, energy consumption, public transportation expenditure, car usage, environmental protection expenditures)

Could you please give me some advice on how to combine these into one composite variable? If I transform each variable into quartiles (1 being environment protection and 4 environment degradation) can I then add the "rankings" and form a composite? or is this not a valid way to create a composite?

Also, is it a good idea to combine the variables into one composite dependent variable, or should I focus on making regressions for each aspect of environmental behavior separately?

I use the software Eviews and SPSS. Please if I am not clear enough let me know to give further details.


I would not discretize / categorize your date into quartiles and sum them. This is too coarse-grained as it does not take into account how far a data point is from the boundary between the quartiles, for instance. (For more information on this topic, you might want to read my answer here: How to choose between ANOVA and ANCOVA in a designed experiment, especially the update.) A common approach to a situation like this would be to turn your data into z-scores (i.e., subtract the mean from every value and divide the difference by the standard deviation), and then average the z-scores for the variables you want to combine. This is often done when working with responses to questionnaires, in which case you want to be careful to 'reverse-score' responses to questions that were asked from the opposite perspective (e.g., how much do you dislike..., rather than how much do you like), but I can't tell if there will be an analogue of this in your data.

Before jumping in to combining your response variables, however, I would explore them to see if they really do hang together the way you expect them to. You should look into factor analysis to do this. All of the top returns listed by the Google search for factor analysis are worth exploring, as are the CV threads tagged under .

You may also want to pursue a more comprehensive approach than using a composite variable as your response in a multiple regression model. It may be better to use structural equation modeling instead. Wikipedia's article may be a good place to start, as well as the CV threads under . I don't know much about eviews, but SPSS will not do SEM, you would need to buy their companion product AMOS instead.

  • 1
    $\begingroup$ (+1) In case SEM or factor analysis seem daunting (e.g., given your time frame), you could assess the extent to which your components hang together by inspecting a matrix of their correlations. One step up from this would be to compute Cronbach's alpha using SPSS's Scale...Reliability command. As gung said, you first may need to reverse-score some variables so that in all cases "high=good." $\endgroup$ – rolando2 Nov 11 '12 at 17:31
  • $\begingroup$ thanks for your input. one more quick question: can z-score be applied to country level data as well treating each country as an "individual"? $\endgroup$ – user16724 Nov 11 '12 at 17:39
  • $\begingroup$ If you have a data point $x_{ij}$ for a study unit $i$ on a variable $j$, you can convert that value into a z-score--it's just a linear transformation. It doesn't matter whether your study units are countries or people. $\endgroup$ – gung - Reinstate Monica Nov 11 '12 at 17:56
  • $\begingroup$ ok. thank you very much. I will discuss the input you gave me with my supervisor. $\endgroup$ – user16724 Nov 11 '12 at 18:07
  • $\begingroup$ and one more thing. if a component has a reverse score I can always use -z instead of adding it when computing the composite variable, right? $\endgroup$ – user16724 Nov 11 '12 at 18:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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