Authors' reply

Friedman et al raise doubts as to the two-factor structure of the HAD scale reported by us. The size of our sample (n=51 930) allowed us to test our finding in several sub-samples. Using principal-components analysis, the same two-factor solution was also found in all sub-samples reporting somatic and psychiatric problems, as well as in all age— and gender-groups from 20 to 89 years. This indicates that the two-factor structure of the HAD scale is robust and stable. Therefore, eventual minor biases due to response rates cannot account for the discrepancy between Friedman et al's and our findings. Our third factor, which emerged only in sub-samples with low depression scores, always showed a low eigenvalue. Our results are in accordance with the conclusions of a recent literature review on the HAD scale (Bjelland et al, 2002) which concludes that a two-factor solution is most commonly found.

Friedman et al (2001) have a sample (n=2669) characterised by major depression (DSM-IV), which corresponds to high depression and probably variable anxiety scores on the HAD scale. When performing factor analysis, composition of the sample is essential for the results. If an inclusion criterion restricts the variance and covariance of the variables entered in the factor analysis, this will influence the factor solution found. The results by Friedman et al can be interpreted as a consequence of their restriction of their sample to major depression only, as this restricts the covariance between items on the HAD scale. In our sub-sample with various mental problems (n=2098) the two-factor solution is robust with high explained variance (82.1%).

Friedman et al's findings are of interest, however, since they answer the question: What is the factor structure of the HAD scale when anxiety appears in major depression? Comparing the fit coefficients between two— and three-factor solutions using confirmatory factor analysis must show the advantage of a three-factor solution. Friedman et al seem to presume that the factor structure of anxiety found in major depression is identical to that found for anxiety in the general population.

The advantage of population samples is that selection bias is minimised. In several of our studies based on the unselected HUNT-II population (from the Nord-Trøndelag Health Study) we have found results at variance with those of clinical samples (Engum et al, 2002; Wenzel et al, 2002). This could also explain the discrepancy between Friedman et al's and our results.