Abstract
Background. Special population studies often report group differences in contrast sensitivity (CS) that are non-uniform across the spatial frequency spectrum. Such interactions have been, at times, consistent; they have also led to mechanistic speculations. Here, we considered whether illness-specific non-uniform reductions in CS could be explained by the heteroscedasticity and rightward skew inherent to CS data. We further considered whether group differences in visual acuity might generate more of a deficit at higher spatial frequencies. Method. We leveraged a publicly available data set of 75 healthy controls and 68 schizophrenia/schizoaffective disorder patients with at least 20/32 acuity (Zemon et al., 2020). Participants attempted to locate briefly appearing (33 ms, 500 ms) laterally-displaced sinusoidal gratings in a staircase-controlled spatial 2AFC design. To maximize the chances of finding spatial frequency interactions, we only analyzed data from the two extreme spatial frequency conditions (0.5 and 21 cpd). ANOVAs were conducted once on the raw (skewed/heteroscedastic) CS data and once again on the log-transformed data. Generalized estimating equations (GEEs, with log-link gamma distribution) were conducted to confirm the results. Results. In the raw data, patient deficits worsened with spatial frequency (p<.001); in the log-transformed data, the interaction reversed (p=.04). The GEEs confirmed the log-transformed results (p=.027). The significant interaction in the GEEs and log-transformed ANOVAs would disappear if we were to match groups on acuity by removing controls with excellent vision (logMAR<0.0). Results were similar for the two stimulus duration conditions. A caveat is that matching groups on acuity is probably only defensible if the acuity differences arose from non-neural factors (e.g., refractive error). Discussion. Our results reconcile seemingly inconsistent findings (e.g., Zemon et al., 2020; Butler et al., 2005) and suggest that past studies may need to be revisited if they do not properly account for heteroscedasticity or group differences in acuity.