For my research, I want to compare the effect that product pictures, product description and consumer reviews have on attitude towards the product and on purchase intention. And I want to compare that effect in the context of t-shirts and facial cream in ecommerce. Simplified, the hypothesis is (for now) that the effect is less positive on facial cream (fc).

I have measured attitude on a 7-point likert scale with 6 items and I want to perform a regression analysis. My supervisor gave me a hint with the following equation: Att = ß0 + ß1*Pic + ß2 * Descr + ß3 * Rev + ß4 * fc + ß5 Pic * fc + ß6 Descr * fc + ß7 * Rev * fc * Errorterm

Now the problem is that I am unsure how to start. Do I take the average (???) of the 6 items measuring attitude for each scenario? Also how do I construct the interaction variables?

I would be extremely grateful if someone could point me into the right direction or recommend a good source of literature (I did not find something that fits my scenario). If you need more specific info about something let me know!

I appreciate your help!


1 Answer 1


Basically you want to reduce 6-item likert scale to a single variable. That is usually done through some kind of factor analysis or principal component analysis (see Wikipedia as a starter or on this site for example here). Averaging is almost always a bad idea because if you have have two statements where one person answers totally agree on the first and totally disagree on the second that person will "on average" be neutral given those two statements.

Moreover, there are questions on the validity of calculating the average on an ordinal scale because it is not immediately clear whether the "distance" between for example "neutral" and "agree" is the same as the distance between "agree" and "completely agree". This problem partly applies to factor analysis as well, although people still use it (see here).

Regarding your interaction variables, that depends on the stats-program you're using. Most programs I know have the option to include interactions either through a specific code (in R) or by selecting them in one of the regression menus (in SPSS I assume).

  • $\begingroup$ Thanks for your help and suggestions! I had already played around with regression in Excel a little bit, but I understand why working with the average is a bad idea. This definitely helps a lot! Much appreciated! $\endgroup$
    – Johannes
    Commented Jul 25, 2017 at 15:52

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.