In short, you could first conduct factor analysis and "save" the variables, and then use linear regression!

Use a summative scoring method If you have ordinal rank weighted scales (aka Likert, etc.). Strongly advise calibrating the scale to a minimum of zero, aligned with the sentiment that you are measuring psychometrically (Lower summed score = less intention to purchase, Higher summed score = greater intention to purchase). Four items (survey questions) on each instrument would yield a summative score distribution with a Minimum of 0, and a Maximum of 16 (0 X 4 = 0, 4 X 4 =16), when sentiments are weighted this way, regarding the sentiment "to purchase" as a frequency:

0 - Never

1- Rarely

2 - Sometimes

3 - Often

4 - Always

Note that this is still a five point ordinal weighted scale.

Comparison of the two distributions can be done as strength of association (recommend Spearman's rho correlation rather than GLM Pearson etc., as the distribution is likely to fail the Kolmorgorov-Smironv test for normality of distribution). The strength and significance of the association between the two different distributions of scores (certain brand vs other brand) sort of answers the comparability. Going one step further, use a non-parametric individuals test, Mann-Whitney U to compare these two score distributions (certain brand vs other brand) to resolve the Null hypothesis, namely that there is no significant difference (95% Confidence, p

Hello Tam,

You certainly have a lot of variables!

For the individual scales, I would echo Richard E Gilder 's suggestion that you create summated scores. For many of the scales (e.g., the "Prestige value in social network") it might make sense to factor the scores first to convince yourself that the target scale is unidimensional (and therefore, summing the scores is warranted).

But, to target the two measurements that appear to be of interest (the four opinion items that follow each product pair), I would suggest:

1. sum the scores into a total (or average)

2. for each individual set, run a one-sample t-test, comparing the mean score to 20 (or 5). If different from the target value, then respondents preferred one option over the other. If you're worried that the total (or average) scores won't meet t-test assumptions, then you could treat the comparison as a variant of the median test, wherein you count the instances in which scores are above the target value (20 or 5), below the target value, or at the target value (usually via a chi-square test).

3. For product 1 (shoes) vs. product 2, either a repeated measures anova, or dependent t-test (same thing), or Wilcoxon signed-ranks sum test (if you're worried about interval strength of data). For this comparison use the two total (or averaged ) scores as the data for each case (1 for shoes, 1 for bags).

I'd guess, given the number of variables you have that you're likely to want to explore other research questions. Possibly ordinary least squares regression or ordinal regression could work for you, but without knowing your specific research questions, that's just a guess.

Good luck with your work.