David Peterzell

Discovering Sensory Processes Using Individual Differences: A Review and Factor Analytic Manifesto.

In the last century, many vision scientists have considered individual variability in data to be “error,” thus overlooking a trove of systematic variability that reveals sensory, cognitive, neural and genetic processes. This “manifesto” coincides with old and recent prescriptions of a covariance-based methodology for vision. But the emphasis here is on using small samples to both discover and confirm characteristics of visual processes, and on reanalyzing archival data. This presentation reviews, briefly, 215 years of sporadic and often neglected research on normal individual variability in vision (including 25+ years of my own research). It reviews how others and I have harvested covariance to a) develop computational models of structures and processes underlying human and animal vision, b) analyze and delineate the developing visual system, c) compare typical and abnormal visual systems, d) relate visual behavior, anatomy, physiology and molecular biology, e) interrelate sensory processes and cognitive performance, and f) develop efficient (non-redundant) tests. Some examples are from my factor-analytic research on spatiotemporal, chromatic, stereoscopic, and attentional processing. Introduction Here I report research examining individual differences (IDs) in a diverse variety of vision studies, and demonstrate how these differences contain systematic variability that elucidates sensory, cognitive, neural and genetic processes. Most vision science uses “experimental” paradigms, focusing on average differences across stimulus conditions, while treating IDs as “random error variance.” But data from many of these experiments contain a separate type of information relevant to studying visual mechanisms. Far less vision science has focused on “correlational” or “factor analytic” approaches which treat normal IDs as systematic and meaningful, reflecting the true variability of underlying processes more than random error. I propose a more extensive exploration of systematic IDs in visual data to identify “factors” of the visual mind, eye, nervous system and genome. In fact, attempts to understand vision by using IDs, and to understand IDs by modeling visual processes, have been pursued since the 19th century, when Bessel and others [1][4] studied IDs in temporal detection among astronomers, and Galton [5]-[6] attempted to link visual performance to general cognitive abilities. This “manifesto” coincides with scattered, oftneglected prescriptions for a covariance-based methodology for vision, and with recent studies that interrelate IDs in functional organization, anatomy, physiology, heredity, psychophysics, optics, cognition, and multiple visual functions, using normal, clinical, developing and aging populations [7]-[31], [64]-[178], including my own work [32]-[53]. But the primary emphasis here is on analyzing IDs using small samples obtained from typical experiments and archival data, to discover and confirm visual processes. Figure 1. Luminance spatial contrast sensitivity functions (CSFs) for a few individuals selected from larger samples. Upper panel: adults, photopic [38]. Middle panels: infants, photopic [31]-[32], [36]. Lower panel: adults, scotopic [44]-[45], [54]. Solid lines without points show means for complete samples. Orange, purple, green, and yellow delineate separate “factors.” Even for small samples, IDs at one SF correlate with IDs at neighboring but not distant SFs. ©2016 Society for Imaging Science and Technology DOI: 10.2352/ISSN.2470-1173.2016.16HVEI-112 IS&T International Symposium on Electronic Imaging 2016 Human Vision and Electronic Imaging 2016 HVEI-112.1 1. The variability in visual data is often systematic, not random. This principle becomes clear when inspecting the spatial contrast sensitivity functions (CSFs) in Figure 1, which contains: (A) Top panels: three CSFs obtained from a larger sample of human adults under photopic conditions [38]. (B) Middle panels: a longitudinal sample of data obtained at photopic light levels from 4 exceptionally consistent infants at 4-, 6, and 8 months of age [31]-[32], [36]. (C) Lower panel: three CSFs obtained from a larger sample of human adults under scotopic conditions [54], [44]-[45]. Even in these small samples, and without correlational or factor analytic statistics, a clear “intuitive factor analysis” is possible. These data were selected to show that IDs at a particular spatial frequency are correlated with IDs at neighboring but not distant spatial frequencies. For instance, in the top panel (orange region), observer KB shows the highest sensitivity for all four spatial frequencies below 1 c/deg, while DT is near the mean, and HK is below the mean for all four. But in the rest of the top panel (purple region), KB’s sensitivity regresses to the mean, DT drops below the mean, and HK rises above the mean. While results within a region (orange or purple) inter-correlate, results across larger regions seem not to inter-correlate (between orange, purple). And so, this typical (as I’ve found) example shows that even with very small samples containing negligible measurement error, systematic and potentially informative differences reside in IDs. The patterns evident in small samples are often consistent with what is found in much larger sets. For instance, Figure 2 shows results for larger samples for spatial CSFs obtained for adults using luminance modulated gratings (as in Figure 1). The upper and lower panels are for photopic and scotopic CSFs, respectively. Each square within the matrix of squares represents a scatterplot. Each scatterplot plots log contrast sensitivities obtained for many individuals at one spatial frequency as a function of log contrast sensitivities contrast obtained for many individuals at another spatial frequency. For instance, in the scotopic data [44][45], [54], in the second to left top square, the sensitivities of 50 observers obtained using .2 c/deg gratings are plotted as a function the sensitivities of these 50 observers obtained using .4 c/deg gratings, with visibly high correlation. The correlation between .2 c/deg and 1.2 c/deg is also positive, but not as strongly correlated. As such, regions of inter-correlation can be seen in these data. As with Figure 1, regions of inter-correlation are marked by colored boxes, with distinct regions evident. And so again, in these comparatively large samples, a clear “intuitive factor analysis” is possible, even without correlational or factor analytic statistics. Although the example provided is for spatial contrast sensitivity functions, it is worth noting that such systematic variability with clearly delineated underlying factors is typical of high quality visual data collected using psychophysics (large numbers of trials per point, and relatively bias-free methods such as 2AFC) and many electrophysiological measures (e.g., electroretinograms and visual evoked potentials). (See historical section). This is certainly the case in my own work and collaborations involving human spatial and temporal CSFs, using luminance and chromatic gratings, photopic and scotopic light conditions, psychophysics and VEPs methods, and adults and infants. It is also the case for many other types of data I’ve investigated, including spectral sensitivity functions in man and (genetically modified) mouse, binocular corrugated gratings, VEP contrast response functions, infant visual attention data, and color naming [32]-[53]. Figure 2. Scatterplot matrices for full sample of individuals’ photopic and scotopic CSF data. 2. Systematic variability is usually visible and interpretable in terms of underlying processes, even in data from a few individuals. When one inspects CSFs from adults, infants, and non-human species, patterns of individual variability consistent with spatiotemporal channels become evident. Figure 3, for instance, shows spatial frequency tuning curves for six foveal, scotopic processes postulated by Wilson and Gelb after modeling psychophysical masking data [55], [56]. The orange and purple regions from Figures 1 and 2 correspond to regions primarily detected by mechanisms A and B, respectively. ©2016 Society for Imaging Science and Technology DOI: 10.2352/ISSN.2470-1173.2016.16HVEI-112 IS&T International Symposium on Electronic Imaging 2016 Human Vision and Electronic Imaging 2016 HVEI-112.2 That is, it looks like the “sources of variability” (a term used in factor analysis) in Figures 1 and 2 are explained by, or consistent with, mechanisms A and B from the model of Wilson and Gelb. Similar indications of underlying mechanisms are visible in IDs for other published functions (e.g. spectral sensitivity, luminous efficiency, color matching, sensitivity of horizontal and vertical corrugations defined by binocular disparity). But visual and perceptual scientists typically focus on average differences across experimental conditions (e.g., average differences in contrast sensitivity for different spatial or temporal conditons, or different ages). 3. Analyses of IDs can be used to confirm what is already “known” about underlying processes, or to discover previously unknown visual processes. If one has a priori knowledge or a strong theory of the processes underlying a visual function, then confirmatory factor analyses should succeed in recovering those processes from IDs in that function. This type of “confirmatory” analysis is contrasted to “exploratory” analyses, in which one uses individual differences and factor analyses to discover previously unknown processes or sources of variability. The difference between confirmatory and exploratory factor analyses is illustrated in Figure 4, for CSF data. In a confirmatory analysis (upper section), one begins (Panel 1) with an a priori model of underlying spatially-tuned processes, and attempts to predict (blue arrow) the pattern of IDs obtained in empirical data (Panel 4, i.e., the types of patterns presented in Figures 1 and 2). In other

Individual differences in visual science: What can be learned and what is good experimental practice?

ABSTRACT We all pass out our lives in private perceptual worlds. The differences in our sensory and perceptual experiences often go unnoticed until there emerges a variation (such as ‘The Dress’) that is large enough to generate different descriptions in the coarse coinage of our shared language. In this essay, we illustrate how individual differences contribute to a richer understanding of visual perception, but we also indicate some potential pitfalls that face the investigator who ventures into the field.

Variations in normal color vision. VII. Relationships between color naming and hue scaling.

HighlightsColor normal observers vary in how they name stimuli and scale their color appearance.Factor analyses show differences in both tasks depend on multiple processes.Some factors reveal a common influence on both hue scaling and hue naming.Differences in color naming may partly reflect differences in color perception. ABSTRACT A longstanding and unresolved question is how observers construct a discrete set of color categories to partition and label the continuous variations in light spectra, and how these categories might reflect the neural representation of color. We explored the properties of color naming and its relationship to color appearance by analyzing individual differences in color‐naming and hue‐scaling patterns, using factor analysis of individual differences to identify separate and shared processes underlying hue naming (labeling) and hue scaling (color appearance). Observers labeled the hues of 36 stimuli spanning different angles in cone‐opponent space, using a set of eight terms corresponding to primary (red, green, blue, yellow) or binary (orange, purple, blue‐green, yellow‐green) hues. The boundaries defining different terms varied mostly independently, reflecting the influence of at least seven to eight factors. This finding is inconsistent with conventional color‐opponent models in which all colors derive from the relative responses of underlying red‐green and blue‐yellow dimensions. Instead, color categories may reflect qualitatively distinct attributes that are free to vary with the specific spectral stimuli they label. Inter‐observer differences in color naming were large and systematic, and we examined whether these differences were associated with differences in color appearance by comparing the hue naming to color percepts assessed by hue scaling measured in the same observers (from Emery et al., 2017). Variability in both tasks again depended on multiple (7 or 8) factors, with some Varimax‐rotated factors specific to hue naming or hue scaling, but others common to corresponding stimuli for both judgments. The latter suggests that at least some of the differences in how individuals name or categorize color are related to differences in how the stimuli are perceived.

Variations in normal color vision. VI. Factors underlying individual differences in hue scaling and their implications for models of color appearance

HighlightsColor‐normal observers vary widely in their color percepts.Factor analyses of the differences suggest they depend on multiple, narrowly‐tuned factors.These factors are inconsistent with conventional models of color opponency.The factors may reflect a population code for color appearance. ABSTRACT Observers with normal color vision vary widely in their judgments of color appearance, such as the specific spectral stimuli they perceive as pure or unique hues. We examined the basis of these individual differences by using factor analysis to examine the variations in hue‐scaling functions from both new and previously published data. Observers reported the perceived proportion of red, green, blue or yellow in chromatic stimuli sampling angles at fixed intervals within the LM and S cone‐opponent plane. These proportions were converted to hue angles in a perceptual‐opponent space defined by red vs. green and blue vs. yellow axes. Factors were then extracted from the correlation matrix using PCA and Varimax rotation. These analyses revealed that inter‐observer differences depend on seven or more narrowly‐tuned factors. Moreover, although the task required observers to decompose the stimuli into four primary colors, there was no evidence for factors corresponding to these four primaries, or for opponent relationships between primaries. Perceptions of “redness” in orange, red, and purple, for instance, involved separate factors rather than one shared process for red. This pattern was compared to factor analyses of Monte Carlo simulations of the individual differences in scaling predicted by variations in standard opponent mechanisms, such as their spectral tuning or relative sensitivity. The observed factor pattern is inconsistent with these models and thus with conventional accounts of color appearance based on the Hering primaries. Instead, our analysis points to a perceptual representation of color in terms of multiple mechanisms or decision rules that each influence the perception of only a relatively narrow range of hues, potentially consistent with a population code for color suggested by cortical physiology.

Thresholds for sine-wave corrugations defined by binocular disparity in random dot stereograms: Factor analysis of individual differences reveals two stereoscopic mechanisms tuned for spatial frequency

Graphical abstract Figure. No Caption available. HighlightsDisparity threshold data for sinusoidal corrugations contained two factors.Factors loaded onto low (<0.4cpd) and high (0.8cpd) spatial frequencies.Each factor had nearly identical tuning for horizontal and vertical patterns.The factors were correlated, implying separate but interdependent mechanisms. ABSTRACT Threshold functions for sinusoidal depth corrugations typically reach their minimum (highest sensitivity) at spatial frequencies of 0.2–0.4cycles/degree (cpd), with lower thresholds for horizontal than vertical corrugations at low spatial frequencies. To elucidate spatial frequency and orientation tuning of stereoscopic mechanisms, we measured the disparity sensitivity functions, and used factor analytic techniques to estimate the existence of independent underlying stereo channels. The data set (N=30individuals) was for horizontal and vertical corrugations of spatial frequencies ranging from 0.1 to 1.6cpd. A principal component analysis of disparity sensitivities (log‐arcsec) revealed that two significant factors accounted for 70% of the variability. Following Varimax rotation to approximate “simple structure”, one factor clearly loaded onto low spatial frequencies (≤0.4cpd), and a second was tuned to higher spatial frequencies (≥0.8cpd). Each factor had nearly identical tuning (loadings) for horizontal and vertical patterns. The finding of separate factors for low and high spatial frequencies is consistent with previous studies. The failure to find separate factors for horizontal and vertical corrugations is somewhat surprising because the neuronal mechanisms are believed to be different. Following an oblique rotation (Direct Oblimin), the two factors correlated significantly, suggesting some interdependence rather than full independence between the two factors.

Factor analysis of individual differences in retinal (PERG) and cortical (VEP) visual contrast responses reveals two retinal and two cortical processes in adults with and without depression.

Recent studies of contrast response functions have found: 1) Separate cortical processes mediate amplitudes at high and low contrasts, as revealed by principal component analysis (PCA) of individual differences in visual evoked potentials (VEPs) (Hamer, Souza et al. 2013). 2) There is a strong reduction in contrast perception and retinal contrast gain in patients with major depression, which normalizes after antidepressive therapy and remission of depression (Bubl et al. 2009; 2010; 2012). The current analysis examines processes (statistical factors) underlying cortical and retinal contrast response functions. Further, it examines which of these processes change with depression. Thirty-five patients with a diagnosis of major depression (26 with and 9 without medication) and 21 healthy subjects participated. Pattern electroretinograms (PERGs) were recorded from both eyes. To quantify PERG and VEP based contrast responses, a sequence of five checkerboard stimuli was presented with 0.5° check size, contrast-reversing at 12 reversals per second, and Michelson contrasts of 3.2%, 7.3%, 16.2%, 36%, and 80%. Individual differences in left and right eye responses were highly intercorrelated, and thus averaged. Principal components were computed from log amplitudes and rotated to approximate simple structure using a Varimax rotation. (1) Two retinal (PERG) and two cortical (VEP) factors were found. The two cortical processes mediate high and low contrasts, consitent with Hamer et al. The two retinal factors also mediate high and low contrasts, but are independent of (uncorrelated with) cortical factors. The four factors were distinct from three additional factors found for VEP noise and PERG noise in left and right eyes. (2) The only factor of the four that changes significantly with depression is a retinal factor tuned to high contrasts. Our analysis identifies probable contrast-sensitive mechanisms, and shows a surprising independence of retinal and cortical gains. Changes in contrast perception in depression may be linked to a single retinal process. Meeting abstract presented at VSS 2015.