A comparison of alpha coefficients in independent samples: a complementary study to the Spanish-language version of EUROQUEST

Dear Editor:

It is important to report confidence coefficients since they allow us to determine the precision and stability of a given measurement.^1,2 The magnitude of the value also provides information on the degree of influence on subsequent statistical analyses.³ In their validation of the Spanish-language version of EUROQUEST, Marhuenda et al.⁴ report the α coefficient, which is one of the most widely used estimators of the reliability of a score and is interpreted as reflecting what percentage of the variance observed corresponds to true variance.^5,6 They then compare their α values with those obtained in a French study,⁶ concluding that their own results are more reliable than those found by the French researchers.

While α coefficients may vary between studies due to differences between the samples analysed, the multicultural approach of EUROQUEST means we may expect the estimated true variance to be statistically similar between the groups analysed.⁸ This would enable us to at least partially determine the absence of bias, among other factors, although these conclusions must be supported by empirical findings.⁹

According to Marhuenda et al.,⁴ the α coefficient values from their sample are more reliable than those found in the French study; however, they reach this conclusion by means of a descriptive comparison. Both this approach (heuristic judgement) and the conclusions reached may have an impact on subsequent studies on EUROQUEST. This is relevant because there is a possibility of researchers erroneously concluding that differences exist. It is therefore important to be aware of the existence of procedures specifically designed for comparing α coefficients obtained from independent samples.^10,11 These comparison methods account for the magnitude of the α coefficient, the sample size, and the number of items, and enable differences between coefficients to be evaluated statistically.

The complementary analysis is based on the total samples of the Spanish⁴ and the French⁷ studies, and the number of items per domain reported in Table 1 of the Marhuenda et al. study. Sensory hyperaesthesia, psychopathological disorders, and self-rated quality of life were excluded from the calculation because the number of items in these domains was not reported.

ALPHATEST¹² was used for the comparison of data, revealing statistically significant differences between the α coefficients found in the French and the Spanish studies; this was the case in all domains except psychosomatic disorders (Table 1). Based on this supporting data, we may therefore concur with the Spanish researchers’ conclusion that their results are more reliable.⁴ Nonetheless, more complex approaches require testing of measurement invariance in order to study potential biases in greater depth.¹³ Researchers are encouraged to use analytic procedures going beyond mere observation in order to provide a solid foundation for their arguments.

Table 1.

Comparison of α coefficients from the French and Spanish EUROQUEST studies.

	No. of items	αSpanish study (n = 759)	αFrench study (n = 768)	χ^2(1)
Organic neurological symptoms	11	0.78	0.71	12.070 (P < .001)
Psychosomatic disorders	15	0.77	0.79	1.380 (P = .240)
Cognitive symptoms	10	0.89	0.79	63.874 (P < .001)
Depressive symptoms	7	0.86	0.78	28.894 (P < .001)
Structural and/or functional symptoms	11	0.91	0.85	40.904 (P < .001)
Sleep or affective disorders	4	0.71	0.58	15.565 (P < .001)
Intoxication	11	0.73	0.64	13.086 (P < .001)

Funding

The author has received no funding of any kind.

Conflict of interest

The author has no conflict of interest to declare.