LETTER TO THE EDITOR Misinterpreting Schizophrenia Relatives’ Impairments Angus W. MacDonald III* and Seung Suk Kang Departments of Psychology & Psychiatry, University of Minnesota, Minneapolis, Minnesota Received 29 January 2008; Accepted 2 May 2008 TO THE EDITOR: We recently had the pleasure to read Dr. Ma and colleagues’ study of deficits in the relatives of patients with schizophrenia [Ma et al., 2007]. In this study, the researchers evaluated 207 schizophrenia patients, 287 of these patients’ non-psychotic parents and 133 controls on a number of clinical neuropsychological tests. While there was a broad range of impairments in the schizophrenia patients, their parents showed poor performance on a much smaller subset of tests, including the Stroop test, the Verbal Fluency test, the Tower of Hanoi, and the Wisconsin Card Sorting test. The authors interpreted this to mean that the unexpressed genetic liability to schizophrenia was expressed as a ‘‘selective’’ deficit and implicated a prefrontal cortical dysfunction. We agree with the authors that understanding the behavioral impact of the genetic liability to schizophrenia is an important goal [Snitz et al., 2006], and their interpretation was in no way outside the bounds of common practice. In this letter we hope to show that, no matter how common, the practice of inferring that a particular cognitive mechanism or brain system is associated with the unexpressed genetic liability to schizophrenia based on such data is flawed. Importantly, this practice poses an opportunity cost to the field that detracts from the effort to build a cumulative science of the brain systems associated with genetic liability to this debilitating disorder. The flawed inference is a version of the psychometric confound described by Chapman and Chapman [1978]. This confound can occur if the tasks being compared have different degrees of dis- criminating power. Discriminating power is the capacity to distin- guish between two groups and depends on several test properties; it increases generally linearly with reliability, loglinearly with variance and the number of items, and as an inverted u-shaped curve with accuracy. This presents a problem because even tasks that putatively measure specific abilities—say set-shifting or spatial memory—also measure non-specific abilities such as motivation, attention, and strategy formation. One may misinterpret a larger group difference on the set-shifting task as reflecting a deficit in that particular ability, when it may as easily be the result of a group difference in a non-specific ability measured with greater discriminating power. To illustrate the impact of the psychometric confound in neuropsychological testing, we used a computer simulation in which the patient group had a fixed 1.5 SD impairment, and the parents had a fixed 0.75 SD impairment in their latent ability (see Fig. 1). Simply by changing the characteristics of the task used to measure this latent ability, the measured effect sizes changed. Thus, the same impairment in latent ability across several tasks (‘‘A’’ through ‘‘G’’ in the Fig. 1) can result in a pattern of deficits that is very similar to that published by Ma and colleagues (see their Fig. 1) and commonly seen elsewhere to support the inference of a specific deficit. Notice, that although our simulation parameters were not particularly selected to duplicate the tests used by Ma, our Figure incorporates their use of a z-score transformation to compare the extent of the deficit across tasks. This illustrates the principle that despite this transformation, the apparent deficit waxes and wanes across psychometric variants of tasks that measure the exact same latent ability. Although clinical neuropsychological tasks have many superficial differences, data from such batteries do not address the possibility that these contrasts all reflect a general factor measured with different amounts of discriminating power. This point has been raised before in the context of understanding the nature of the cognitive deficits in schizophrenia [Chapman and Chapman, 1978], but it is increasingly important to incorporate into experimental design as we contemplate a future with ever more neuroimaging and molecular genetic studies focused on the unex- pressed genetic liability to schizophrenia. Discerning specific cog- nitive deficits associated with this liability has been shown to be possible, may be extremely useful, and may increase the power of association studies [MacDonald et al., 2007]. However this will take deeper consideration and further refinement of behavioral tasks. Off-the-shelf neuropsychological tests are ill-suited to shine a light into this cave. *Correspondence to: Prof. Angus W. MacDonald, III, Ph.D., N219 Elliot Hall 75 E. River Rd., Minneapolis, MN 55455. E-mail: angus@umn.edu Published online 16 June 2008 in Wiley InterScience (www.interscience.wiley.com) DOI 10.1002/ajmg.b.30800 How to Cite this Article: MacDonald AW, Kang SS. 2009. Misinterpreting Schizophrenia Relatives’ Impairments. Am J Med Genet Part B 150B:443–444. Ó 2008 Wiley-Liss, Inc. 443 Neuropsychiatric Genetics