2. Final solution with item 17: GHQ-28 characterised by greater homogeneity

When the analysis was repeated by using item 17 in place of item 50, the first four factors accounted for 50.2% of the variance (see Table 8). All the items were found to show a high load on pre-assigned factors. Following analysis, item 17 ("Difficulty in getting off to sleep"), which had been chosen as the eighth item in factor 1 (Anxiety and Insomnia), registered an even higher load on the same scale as items 47 and 49, moving up to sixth place.

Table 8. Factor structure of the 28-item GHQ version (varimax rotation of the 4-factor solution, accounting for 50.2% of variance). Most homogeneous solution with a scale of 8 items, including item 17

As in the first solution, the different scales are not pure measures of the four factors. Table 9 sets out the mean factor loads for each scale on each factor in this second possible final solution.

Table 9. Mean factor loads of the 4 scales on the four factors. Most homogeneous solution with a scale of 8 items, including item 17 (50.2% of variance on these factors: varimax rotation of the 28-item questionnaire)

Comparison of the factor structures of the different versions of the GHQ

After performing an unrotated principal components analysis, the British authors obtained 11 factors with a eigenvalue greater or equal to one, accounting for 63.4 % of the variance. We obtained 14 factors in a general urban population, accounting for 59.9% of the variance (versus 13 in Vázquez-Barquero's general Spanish rural population study in 1988, and 14 in Medina-Mora's Mexican study in 1983).

Table 10 compares the results yielded by varimax rotation of the first six factors in our sample against those yielded by the English version. As will be clearly observed, the greatest stability in both studies lies in the scales that form part of the final four-factor solution, namely: "Severe depression", "Sleep disturbances" and "Social dysfunction". Furthermore, the "Anxiety and dysphoria" scale in the English study shares a single item (item 55) with the "Anxiety" scale in our study, and another (item 50) with "Loss of self-esteem".

Table 10. Comparison between the factor structures of the 30-item GHQ version obtained in our sample and the English version (varimax rotation of the 6-factor solution)

In the study by Vázquez-Barquero et al. -which was undertaken with a different objective to that of construction of a GHQ-28- all items having a load of over 0.5 were retained, whereas in our, Medina-Mora's and Goldberg and Hillier's studies the five with the highest load were retained. Although this difference in criterion limits comparison, the 6 factors obtained by Vázquez-Barquero et al from all 60 items proved to be different to those reported in the other studies. It is noteworthy that the factor, "Somatic symptoms", was identified in neither of the two Spanish samples, while Goldberg and Hillier were able to differentiate 2 factors of general illness and somatic symptoms (only one factor was observed in the Mexican data).

The results obtained in our sample compared to those of the original English version (Goldberg and Hillier, 1979) and the Mexican study (Medina-Mora et al., 1983)

Both possible final solutions proposed in the Results section of this study differ from the Goldberg and Hillier version in 8 items (Table 11). The Mexican study by Medina-Mora et al. differs by one more. Despite the differences, there are greater similarities between the English version and the solution yielded by our study, than between our results and the Mexican solution, as can be seen from Table 12: the first two coincide in 20 items (versus 19) and combine the items of Anxiety and Sleep disturbances in a single dimension, while in the Mexican version, "Sleep disturbances" constitutes a pure scale.

Table 11. Comparison between the two possible factor structures of the 28-item GHQ version obtained in our sample and that of the English version (varimax rotation of the 4-factor solution)

Table 12. Comparison of items comprising the GHQ-28 in three studies. The English and Mexican versions conducted on primary-care populations and our study conducted on general Spanish population (varimax rotation of the 4-factor solution)

DISCUSSION

On applying the same methodology to the general Spanish population as that used by Goldberg and Hillier (1979) in developing the original scaled version , we found that shorter versions can be derived from the GHQ-60. In this paper, 2 short versions have been included: one with 6 scales of 5 items each (N=30), and another with 4 scales (N=28). Our study confirms the hypothesis that the questions obtained would not be the same as those featured in the English version.

The 6 factors obtained from all 60 items proved to be different from those reported by Goldberg and Hillier (1979). While they were able to differentiate 2 factors of general illness and somatic symptoms, we observed only one in the Spanish data. Solely three of the five questions comprising this scale appear in the English version, two (3 and 4) in their "General illness" factor and another (9) in "Somatic symptoms". Moreover, the "Anxiety and dysphoria"scale in the English study shares item 55 with the "Anxiety" scale, and item 50 with the "Loss of self-esteem" scale which is not represented in the English version. "Sleep disturbance" coincides in 4 of five items, while "Severe depression" is exactly the same. The remaining factor, "Social dysfunction", coincides in three of five items (28,30,36). In the shortened 28-item GHQ, only 20 items coincide with the English version. Unlike the Mexican version which features a scale with items exclusively relating to Sleep disturbances, in the English version and our study these are combined in a single scale along with items related to anxiety disorders. Lastly, in line with the different philosophy used for the construction of the GHQ-28 versus other versions of the General Health Questionnaire, it would seem wise for priority to be given to the criterion of homogeneity in the scales of any 28-item version of the GHQ proposed designed for use in the general Spanish population (in an attempt to ensure that such scales represent common dimensions of symptomatology). The fact that not all the scales would have the same length becomes of secondary importance, when it is recalled that the point at issue is the factor structure of the results of the GHQ-60 obtained in the study population. Our version therefore highlights an aspect that differentiates it from other studies, namely, to obtain four homogeneous scales by selecting 28 items, one of the scales must contain an additional item, with the ultimate consequence that two scales of 7, another of 6 and a fourth of 8 items are thus obtained.

As regards variance, Goldberg was able to account for 53.5% with 6 factors obtained from 60 items, and 59% with 4 factors from 28 items, while we could account for only 43.6% and 50.2% respectively. This is probably due to the fact that in his data the first general component accounted for 35%, while in our data it accounted for only 22.9%.

There are three reasons that might account for the differences between the versions studied. Firstly, there is the possibility that the statistical processing of the samples might be different, inasmuch as the use of a factor analysis technique other than principal components analysis with varimax rotation would necessarily yield different results (Vázquez-Barquero, 1988). This is not the case here, since the method used both for our study and for the Mexican version of the GHQ-28 was the same as that used for the original version. Secondly, there is the possibility that the number of items analysed might be different, since the fewer analysed, the fewer the resulting factors (Worsley et al. 1978). However, this was not the case either, since all the studies cited took the GHQ-60 as their starting point.

Thirdly, there are the characteristics of the samples used. Our sample totalled 654 subjects, a figure large enough to comply with recommendations for the ideal sample size for a GHQ factor analysis, i.e., at least five times greater than the number of variables to be examined (Goldberg and Williams, 1988). The study design used by Goldberg and Hillier and by Medina-Mora et al. is a systematic sampling of a general medical practice. Furthermore, in both cases the proportional number of women in the sample was higher than that of men, given the greater demand for primary care among the former (Molina, 1998). As mentioned above however, the Tres Cantos study featured a different setting (general population), as well as a different sampling design and sample characteristics, with a balance between the numbers of men and women that is not reflected in the other studies.

In this regard, one must not overlook the extent to which the different nature of the sample can cause the relationship between the factors comprising the structure to vary. Our study is unique, inasmuch as the GHQ-28 was constructed on the basis of the general urban population while earlier studies were conducted in a primary-care setting. In other words, the relative severity of the symptoms or moment in the evolution of the illness ("before seeking advice about a symptom" versus "a symptom sufficiently intense to induce one to seek advice") may explain differences in the factor structure. Similarly, sight must not be lost of the proportional number of women making up the study samples (higher in primary-care studies), since studies conducted on samples composed solely of women are known to register a lower number of principal components compared to other studies (Benjamin et al. 1982; Parkes, 1982; Hobbs et al.1983).

The practical value of this study resides in its having produced a real adaptation of the GHQ-28 based on the general Spanish population. This adaptation places an instrument at our disposal, not only capable of detecting possible psychiatric cases in the community but also affording guidance via its subscales. Moreover, this study calls for further field work, aimed at replicating the development of the GHQ-28 in a primary-care setting and, subsequently, comparing the male-female factor structures both in the general population and in primary care. It would be interesting to corroborate whether a GHQ-28 having two scales of 7, one of 6 and another of 8 items would also be obtained in a Spanish primary care sample. Furthermore, the question of the number of items that should ideally comprise the scaled version of the GHQ remains to be ascertained. It falls to subsequent predictive assessment studies to furnish the parameters of sensitivity, specificity and percentage of misclassified subjects for this version's use in a Spanish population.

ACKNOWLEDGEMENTS

We should like to thank Dr. Pedro Enrique Muñoz for his continued confidence and help over the 7 years devoted to the project, "Aportaciones al desarrollo de la versión española del General Health Questionnaire (GHQ) de Goldberg de 28 ítems" (Contributions to the development of the Spanish version of Goldberg's 28-item General Health Questionnaire), as well as all the various professionals who collaborated in the field work entailed in this research.

REFERENCES

BENJAMIN, S., DECALMER, P. and D. HARAN (1982). Community Screening for mental illness: a validity study of the General Health Questionnaire, British Journal of Psychiatry, 140: 174-180.

DOHRENWEND B.P. and B.S. DOHRENWEND (1982). Perspectives on the past and future of Psychiatric Epidemiology, American Journal of Public Health 72, 11: 1271-1277.

GOLDBERG, D. (1972). The Detection of Psychiatric Illness by Questionnaire. Maudsley Monograph No. 21. Oxford : Oxford University Press.

GOLDBERG, D.P. and V.F., HILLIER (1979). A scaled version of the General Health Questionnaire, Psychological Medicine, 9: 139-145.

GOLDBERG, D.P. and P. WILLIAMS (1988) The user's guide to the General Health Questionnaire. Slough: N.F.E.R./Nelson.

GOLDBERG, D.P., COOPER, B., EASTWOOD, M., KEDWARD, H.B., and M. SHEPHERD (1970). A standardized psychiatric interview for use in community surveys, British Journal of Preventive and Social Medicine, 24:18-23.

HOBBS, P., BALLINGER, C.B. and D.M.W. SMITH (1983). Factor analysis and validation of the General Health Questionnaire in women: a general practice survey, British Journal of Psychiatry, 142: 257-264.

LOBO, A., PÉREZ- ECHEVARRIA, M J., and J. ARTAL (1986). Validity of the scaled version of the General Health Questionnaire (GHQ-28) in a Spanish population, Psychological Medicine, 16: 135-140.

MEDINA-MORA, M.E., PADILLA, G.P., CAMPILLO SERRANO, C., MAS, C.C., EZBAN, M., CARAVEO, J. and J. CORONA (1983). The factor structure of the GHQ: a scaled version for a hospital's general practice service in Mexico, Psychological Medicine, 13: 355-362.

MOLINA, J.D. (1998). Aportaciones al desarrollo de la versión española del General Health Questionnaire (GHQ) de Goldberg de 28 ítems. Doctoral Thesis. Madrid: Universidad Autónoma.

MUÑOZ, P.E, VÁZQUEZ-BARQUERO, J.L., PASTRANA, E., RODRÍGUEZ, F. and C ONECA (1978). Study of the validity of Goldberg's 60 items in its Spanish version, Social Psychiatry, 13: 99-104.

MUÑOZ, P.E, VÁZQUEZ-BARQUERO, J.L., RODRÍGUEZ, F., PASTRANA, E. and J. VARO (1979). Adaptación española del General Health Questionnaire (GHQ) de D.P. Goldberg, Archivos de Neurobiología, 42: 139-158.

PARKES, K.R. (1982). Field dependence and the factor structure of the general Health Questionnaire in normal subjects, British Journal of Psychiatry, 40: 392-400.

SEVA, A., MAGALLÓN, R., SARASOLA, A. and J.A. MERINO (1992). GHQ-28 validation in a Spanish general urban population, The European Journal of Psychiatry, 6: 147-153.

VÁZQUEZ-BARQUERO, J.L., WILLIAMS, P., DIEZ-MANRIQUE, J.F., LEQUERICA, J. and A. ARENAL (1988). The factor structure of the GHQ in a community sample, Psychological Medicine, 15: 597-608.

WORSLEY, A., WALTERS, W.A.W. and E.C. WOOD (1978). Responses of Australian patients with gynaecological disorders to the General Health Questionnaire: a factor analytic study, Psychological Medicine, 8: 131-138.