Jesus Sanz, Department of Clinical Psychology, Universidad Complutense de Madrid
14 May 2005
Cronbach’s alphas for the Greek version of the State-Trat Anxiety Inventory (STAI), Form Y, were surprisingly low (0.19 for the State subscale and 0.39 for the Trait subscale) and not consistent with alpha coefficients reported in literature. Previous studies have found high alpha coefficients around 0.80s- 0.90s for both subscales and for a variety of samples: psychiatric patients, university students, adults from the general population, and so on (see, for example, Abdel-Khalek, A. M., 1989, Personality and Individual Differences, 10:277-285; Gauthier, J., 1993, Canadian Journal of Behavioural Science, 25:559-578; Iwata, N., 2000, Journal of Clinical Psychology, 56:793-806; Stanley, M. A., et al., 2001, Behaviour Research and Therapy, 39: 221-235). Furthermore, test-retest correlations for the Greek version of the STAI were so high as 0.96 and 0.98 (for State and Trait, respectively), whereas alphas were so low as 0.19 and 0.39. Although test-retest reliability and internal consistency reliability are not just the same, both indices are reliability estimates and, therefore, such high discrepancies do not make sense. Given that both subscales of the STAI are comprised of direct (or anxiety-present) items (e.g., I feel nervous) and inverse (or anxiety-absent) items (e.g., I am happy), a failure to reverse the inverse items prior to calculate the alpha coefficients may explain the very low alphas reported in this study. I would suggest authors to review the calculation of Cronbach’s alphas and check if the inverse items of the State and Trait subscales of the STAI were properly reversed prior to calculate alpha.
Annals of General Psychiatry
A possible error in calculating alpha coefficient
14 May 2005
Cronbach’s alphas for the Greek version of the State-Trat Anxiety Inventory (STAI), Form Y, were surprisingly low (0.19 for the State subscale and 0.39 for the Trait subscale) and not consistent with alpha coefficients reported in literature. Previous studies have found high alpha coefficients around 0.80s- 0.90s for both subscales and for a variety of samples: psychiatric patients, university students, adults from the general population, and so on (see, for example, Abdel-Khalek, A. M., 1989, Personality and Individual Differences, 10:277-285; Gauthier, J., 1993, Canadian Journal of Behavioural Science, 25:559-578; Iwata, N., 2000, Journal of Clinical Psychology, 56:793-806; Stanley, M. A., et al., 2001, Behaviour Research and Therapy, 39: 221-235). Furthermore, test-retest correlations for the Greek version of the STAI were so high as 0.96 and 0.98 (for State and Trait, respectively), whereas alphas were so low as 0.19 and 0.39. Although test-retest reliability and internal consistency reliability are not just the same, both indices are reliability estimates and, therefore, such high discrepancies do not make sense. Given that both subscales of the STAI are comprised of direct (or anxiety-present) items (e.g., I feel nervous) and inverse (or anxiety-absent) items (e.g., I am happy), a failure to reverse the inverse items prior to calculate the alpha coefficients may explain the very low alphas reported in this study. I would suggest authors to review the calculation of Cronbach’s alphas and check if the inverse items of the State and Trait subscales of the STAI were properly reversed prior to calculate alpha.
Competing interests
No competing interests