Why psychometrics is important - a response to: Aubeeluck, Buchanan & Stupple (2013) Journal of Huntington's Disease 2(4) 453-454.

Dear Editor, We thank Aubeeluck, Buchanan and Stupple (ABS) for their comments [23] regarding our recent paper on the psychometric properties of the Huntington’s Disease Quality of Life Battery for Carers (HDQOL-C) and the Alzheimer’s Carers Quality of Life Inventory (ACQLI) among carers of people with Huntington’s disease (HD) [1], and the Editor for providing us with this opportunity to clarify the issues raised in the letter. It is true that our sample size was small. This was acknowledged [1], together with the fact that studies [2] have shown that samples of this size (or smaller; n≥ 20) can be useful and provide stable interpretations relative to larger samples for basic traditional psychometric analyses. For example, the FACIT-F, a widely used fatigue scale, was initially tested with a sample of n= 49 [3]; subsequent larger-scale studies have not contradicted initial results. Arguing that single-centre samples are problematic means that single-centre studies are inappropriate for the HDQoL-C. This is because traditional psychometric properties are not fixed scale characteristics but sample dependent, and should be supported within the sample in which a scale is used [4–6]. For example, the developers of the SF-36, one of the most widely used rating scales in the health sciences, argue that basic scaling assumptions should always be met before interpreting SF-36 data [7, 8]. We find the comment that our data are homogenous curious. First, and relevant also to the previous point, our sample characteristics were very similar to those previously reported in psychometric HDQoL-C studies [9, 10]. Secondly, all HDQoL-C scores ranged across their full or close to full possible ranges [1]. This aspect is important, since failure to do so limits the interpretability of psychometric data due to lack of sample representation across the latent continuum [11]. Had our HDQoL-C data been restricted to a narrow range, there would have been a problem related to “homogenous data”. This was not the case. Although our sample was recruited from a single centre, the recruitment area covered four English counties (a fact that unfortunately was unacknowledged in our paper). We agree that factor analysis (FA) is common in psychometric studies. However, there are good arguments against factor analysing item-level data [12]. The prime reason for conducting item-level FA is to assess dimensionality [13]. Due to our limited sample size and inherent problems with such analyses, we took an alternative approach, multitrait/multi-item scaling analysis [11, 13]. Although addressing the same fundamental issue as FA, multitrait/multi-item scaling