Summary: I created a model that tries to explain success factors in crowdfunding initiatives. I collected the data by scraping it directly from the crowdfunding platform. That of course limits my research in terms of data availability.

In order to come up with conceptual model I adopted a theory, that assumes latent variables (such that I do not observe). What I observe is measures or proxies (directly collected form the website) that to some extent depict that latent variables which are embedded in my conceptual framework. Of course I set few hypothesis based on that latent variables.

Example: I have adopt a theory about the online community reputation. For measuring reputation I set 2 measures - N of successfully funded projects, N unsuccessfully funded projects. That example is quite simple, however, but for some other concepts I have multiple measures on different scales.

Questions: 1) How should I test my model reliability or/and validity? Is that necessary?

I think I may not consider Cronbach's alpha in that case, since its not a survey approach, and often the measurement scales are different. Another option I was thinking of is CFA, but SPSS have only EFA. Would setting an EFA with defined number of factors equal to the number of my latent variables is an option?

2) How I should test (accept/reject) my hypothesis? Should I refer to the measures themselves, or should I refer to the latent variables? How I can create the latent variables if necessary?

  • 2
    $\begingroup$ Please explain the abbreviations CFA and EFA $\endgroup$
    – Metrics
    Jul 22, 2013 at 14:35
  • $\begingroup$ CFA (confirmatory factor analysis), EFA (exploratory factor analysis) $\endgroup$
    – user27588
    Jul 22, 2013 at 15:08

1 Answer 1


I recommend testing your measurement model (the model that includes only your latent variable comprised of measured variable) by means of confirmatory factor analysis (CFA). If your measurement model fits well, then you can be moderately confident that your latent variable likely measures what you want it to measure (or at least as confident as you can be in this scenario). This would be the model validity, in a sense.

Then, you can plot out your full model by adding in whatever other variables you are interested in (e.g., latent variable predicting an observed dependent variable). You can use this to test your hypothesis (e.g., latent variable predicting variable Y). You can discuss the variables in latent terms.

SPSS can do CFA, although you need the AMOS program added to it.

  • $\begingroup$ Thanks Behacad. And lets say for my observed variables I used the following transformation ln(y+1)=b0+b1ln(x1+1)+b2ln(x2+1)+⋯+bnln(xn+1). In doing my SEM/CFA should I use the transformed variables ln(x+1) or as they are observed (just x)? $\endgroup$
    – user27588
    Jul 24, 2013 at 11:33
  • $\begingroup$ I'm unsure, sorry, and that might depend on the reasons for your transformation. Probably the transformed values, though. You can always try both and see which fits best and which works best for your purposes. $\endgroup$
    – Behacad
    Jul 24, 2013 at 13:58

Your Answer

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