GEN Is Not HGN

ceiling effect would not be present. Indeed, the ceiling effect strongly suggests that the A&SQ may perform better in samples of amblyopes with acuity much poorer than that of the sample we tested. We agree with Drs. Kelderman, Felius, and Passchier that our modest sample size of 102 amblyopes limits the extent of any interpretation of the Rasch analysis that can be made from our results, and we highlighted that point in our paper. We also agree that the Rasch model is not a panacea, as no model is. It does, however, provide scientific measurement properties of data that are otherwise ordinal. The main problem with the A&SQ does not appear to be whether it is unidimensional, but whether it provides a valid measurement of the quality of life of patients with strabismus and/or amblyopia. In the original Likert-scaled A&SQ, responses to all items (questions) were weighted equally. Unless all items are equally difficult or important, then simply adding them all up to get a final score does not provide a valid measurement. This observation is particularly true of the A&SQ when subjects are arbitrarily given a score of 4 if they are unable to answer a particular question. Our Rasch analysis of the A&SQ highlighted how different some items were from others (see Fig. 2 in our paper), so that responses to item 7, “miss the other person’s hand when shaking hands,” are weighted very differently from those to item 21, “squint or shut one eye in bright sunlight,” for example, and the difference seems logical. We also believe that the results from the A&SQ are probably not unidimensional. In their letter, Kelderman et al. criticize the validity of the Rasch analysis for assessing the dimensionality of the A&SQ and favor the use of traditional psychometric analysis (i.e., factor analysis). Comparisons between factor analysis results and Rasch analysis should be performed with caution, as the fundamental intent of each method differs. The principal component analysis identifies factors within a correlation matrix (i.e., factor structure underlying the items of the A&SQ), whereas the Rasch analysis determines whether there are other dimensions left once the initial latent trait (i.e., vision-related quality of life) has been extracted. Once subscales have been demonstrated to exist by factor analysis, they should be assessed for unidimensionality. For example, a questionnaire with several subscales (demonstrated using factor analysis) may be considered unidimensional (according to the Rasch analysis), since all the items in the questionnaire measure a single underlying trait. If both are not used, which one should be chosen? Factor analysis assumes that the data being analyzed are linear measures and not the ordinally labeled stochastic observations that are provided by the A&SQ. Studies that have compared factor analysis and the Rasch analysis have concluded that the Rasch analysis is much better at determining the identification of the core construct, particularly when the data are ordinal and factors correlate highly. In our report, not only do the results of the Rasch analysis strongly suggest that there is an important second dimension, but this dimension also includes a relatively large number of items (5 of a total of 23 items) and the inherent qualities of those items (mainly psychosocial) are all similar and seem logically different from most of the remainder, which assess difficulties with functional activities. Thus, we suggest that it may be more appropriate to provide two scores, since the breach of unidimensionality does not allow for appropriate summation of the items within the A&SQE. In turn, our proposed strategy may help to draw meaningful clinical conclusions about the consequences of living with strabismus or amblyopia or both. Marta Vianya-Estopa David B. Elliott Brendan T. Barrett