Hi Mads Lauridsen .

You want here to test the correlation, and on of the variables is dichotomous variable (Yes/No, 0/1, PRO freedom of speech,/ANTI freedom of speech). In the case you should perform point–biserial correlations.

It is applicable also with your sample size.

Dear MADS: There are a lot of approaches that might be applied here, but let me suggest a very basic approach that would have potential for publication. To start, You are working with a 5-point Likert Scale and you may numerically score it as +2 (very agree), +1, 0, -1 , -2 (very disagree). Google “How do I score a Likert Scale” and you will find ResearchGate replies to this query made earlier [Such coring has long history in the Social-Science Literature]. After scoring --to explore the dimensionality of your 9 questions -- I would initially suggest a basic Principal-Components Analysis (PCA) with eigen-value 1 cutoff, Varimax rotation and least Squares estimation of each participants rotated PCA scores [input data matrix with 41 rows (one for each participant (S) and 9 columns (reflecting each of the questions)]. Often offered as a unified option in many stat-packages , this analysis is a traditional first cut at understanding if the 9 questions are unidimensional or multiple [which very frequently happens in forming attitude questionnaires].

If the PCA reflects a single component dimension [hooray], then divide the component scores for the 41 Ss into your two groups and analyze with a t-test or equivalent F ...with whichever you are most comfortable [note that due to the fact of near equivalency of the moderately-large two group sample-sizes AND PCA score dependencies on 9 questions, a non-parametric is not required {See classically Sec 10.2, that builds on classical Box results, Scheffe, 1959, Analysis of Variance, Wiley]. The finding of more than one Principal Component (EV>1) somewhat complicates as you would then need to identify the nature of the identified question components [a somewhat artful task, but see Bittner et al., 2005 and Sandquist et al. 2012 for examples] and then separately analyze the scores for each of the identified components again with t-test or 2-group F.

I would be remiss if I didn’t point out you could simply add the (+2 to -2) scores of the 9 questions and do a T-test or F-test of the group differences. But the PCA, even if only one component, would almost certainly be substantially more powerful (as PCA component scores take into account the reliabilities of the individual questions; as well as, satisfy future reviewers regarding your instrument. Potentially more importantly, the PCA results would be very helpful, if you ever wished to employ the 9-item questionnaire in the future [illustrated by Bittner et al, 2007). At the same time, I would be remiss, if I didn’t point out that there are a considerable number of more complex approaches that might be applied with factor-analysis vs. PCA (e.g., Sanquist et al. (2012 vs. Bittner et al. (2005; 2007).

I believe, however, that the above Scoring-PCA approach very likely will meet your specific needs with a minimum of complexity.

Regards, ALVAH

Bittner, A.C., Croffut, R.M. & Stranahan, M.C. (2005) Prescript-Assist TM probiotic-prebiotic treatment for irritable bowel syndrome: Randomized, placebo-controlled, double-blind clinical study. Clinical Therapeutics, 27(6):755-761.

Bittner, AC, Croffut, RM, Stranahan, MC & Yokelson, TN (2007). Prescript-AssistTM probiotic-prebiotic treatment for irritable bowel syndrome (IBS): Open, partially-controlled, one-year extension of previously published controlled clinical trial. Clinical Therapeutics, 29(6):1153-1160.

Sandquist TF, Orr H, Shui B & Bittner AC (2012). Lifestyle Factors in U.S. Residential Electricity Consumption. J. of Energy Policy, 4:354-364