Здравата потвърдителна факторна анализа

Standard CFA assumes that the observed indicators follow a multivariate normal distribution. In practice, Likert scales, clinical ratings, and behavioural counts are rarely normal — they tend to be skewed or heavy-tailed. When normality fails, the maximum likelihood chi-square statistic inflates, standard errors shrink, and model fit indices become misleading. Robust CFA fixes this by computing a scaled test statistic and sandwich-type standard errors that remain valid even when the distributional assumption is violated. The factor structure being tested is still specified in advance — the confirmatory logic is unchanged — but the inferential machinery is made resistant to non-normality.