Statistics and pain-related fear measures in acute low back pain.

Swinkels-Meewisse et al. (2003) examined some of the psychometric properties of both the Dutch Tampa Scale for Kinesiophobia (TSK-DV) and the Fear-Avoidance Beliefs Questionnaire (FABQ) in acute low back pain patients. We congratulate the authors for their work; early identification of acute (low back) pain patients at risk of becoming chronic may be of prime importance to reduce both economic costs and patients’ disability. In this view, the usage of a reliable and valid measure to assess pain-related fear (of movement) should be encouraged. However, we are concerned about the statistical analysis. Since the items on both the TSK-DV and the FABQ are scored on a Likert scale, they generate ordinal data (even the total scores are ordinal data). A Pearson correlation analysis is therefore inappropriate for both the analysis of the test–retest data and the examination of the concurrent validity of both measures. The Pearson product-moment correlation coefficient is the most commonly reported measure of correlation, but its area of application is restricted to the interval and ratio scales (Portney and Watkins, 2000a). The Spearman rank correlation coefficient should be used with ordinal data (Portney and Watkins, 2000a), and would therefore have been appropriate to analyse the concurrent validity of both measures. Second, the descriptive data of the total scores obtained with the TSK-DV and the FABQ were not reported adequately: the median and the interquartile range are the appropriate descriptive statistics for ordinal data, but Swinkels-Meewisse et al. (2003) reported only the means, medians and standard deviations of the total scores of both the TSK-DV and the FABQ (on p. 32 and 33). The authors were not the first to do so, previous research using the TSK(-DV) consistently reported mean and standard deviations for the total scores on the TSK(-DV) (Vlaeyen et al., 1995; Goubert et al., 2000; Silver et al., 2002). Furthermore, neither a Spearman correlation, nor a Pearson correlation analysis, would have been appropriate for the analysis of the test–retest data. Correlation tells us how the scores vary together, but it cannot tell us the extent of agreement between the two sets of measurement (Portney and Watkins, 2000b). If a test is reliable, then the performance of the test on different occasions should yield (nearly) identical results (Fritz and