Answers are grounded in this paper and cite their sources.
A persistent question in Q methodology is whether the number of participants in the P set may exceed the number of items in the Q sample. Rules of thumb imported from R-technique factor analysis, which require more observations than variables, suggest that it may not. This note argues that the prohibition is unfounded, but that the opposite claim—that the ratio of persons to statements is analytically irrelevant—is equally untenable. Because Q factor analysis correlates persons across items, the person correlation matrix has rank at most Q − 1; whenever P ≥ Q it is necessarily singular, and its non-zero eigenvalues must absorb a total variance of P. Three consequences follow. First, the Kaiser eigenvalue-greater-than-one criterion becomes progressively less discriminating as P/Q grows, because the mean non-zero eigenvalue is at least P/(Q − 1); a simulation of structureless forced sorts shows that with Q = 40 and P = 200 the criterion retains all 39 extractable components from pure noise, and that where structure is present its count tracks the noise level rather than the number of latent viewpoints, whereas a parallel analysis built on permutations of the study's own forced distribution recovers that number. Second, procedures requiring the inverse of the person correlation matrix—maximum-likelihood extraction, Bartlett's test of sphericity, the Kaiser–Meyer–Olkin index and regression-based factor score estimation—become unavailable by construction, whereas principal components and centroid extraction are unaffected. Third, the standard error of an individual factor loading depends on Q alone, so additional participants do not sharpen individual loadings, although they can improve composite factor arrays when they supply clean defining sorts. Beyond these mathematical points, large P sets raise a design question, since additional participants may reproduce configurations already observed rather than extend the range of subjectivities encountered. P-set construction is therefore better governed by purposive coverage of theoretically relevant positions than by numerical representativeness or by a statistical stopping rule.
No takes yet. Share an insight, caveat, or question.
Raffaele Zanoli (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: