When do contrast sensitivity deficits (or enhancements) depend on
spatial frequency? Two ways to avoid spurious interactions
Abstract
Studies across a broad range of disciplines–from psychiatry to
cognitive science to behavioral neuroscience–have reported on whether
the magnitude of contrast sensitivity alterations in one group or
condition varies with spatial frequency. Significant interactions have
often gone unexplained or have been used to argue for impairments in
specific processing streams. Here, we show that interactions with
spatial frequency may need to be re-evaluated if the inherent
skew/heteroscedasticity was not taken into account or if refractive
error could plausibly differ across groups or conditions. By
re-analyzing a publicly available data set, we show that–when using raw
contrast sensitivity data–schizophrenia patients exhibit an apparent
contrast sensitivity impairment at low, but not high, spatial
frequencies, but that when using log-transformed data or when using
generalized estimating equations, this interaction reversed. The
reversed interaction, but not the overall contrast sensitivity deficit,
would disappear if groups were matched on visual acuity. However,
matching groups in this way is probably only defensible if acuity
differences arise from optical blur. These analyses reconcile seemingly
discrepant findings in the literature and demonstrate that properly
reporting contrast sensitivity interactions with spatial frequency
requires accounting for refraction error and skew/heteroscedasticity.