The N170, not the P1, indexes the earliest time for categorical perception of faces, regardless of interstimulus variance

Abstract

A negative event-related potential (ERP) at occipito-temporal sites peaking around 150–170 ms after stimulus onset (N170) is typically larger for faces than other object categories. Most theories interpret this finding as due to face-selective processing in occipito-temporal and temporal cortex. However, a controversial account recently attributed the N170 effect to differences in interstimulus variance (ISV) among the images typically used for face and object conditions and proposed that the earlier P1 instead indexes the categorical processes generally attributed to the N170. This ERP study aimed to test this account definitively by using conditions in which the same face and object were shown repeatedly, eliminating both physical and perceptual ISV. Fourier amplitude spectra of faces and objects were matched to equate basic low-level visual properties that may affect early ERPs such as the P1. Results demonstrate that i) face selectivity of the N170 is largely preserved across many object categories after abolishing ISV, and ii) stimulus category does not modulate the P1. This conclusively refutes the ISV account while strongly supporting category as a critical factor driving N170 face selectivity.

Introduction

The N170 is a negative event-related brain potential (ERP) between 120 and 200 ms that peaks around 150–170 ms post-stimulus and is usually larger for images of faces than other objects (Itier and Taylor, 2002, Jeffreys, 1989, Jeffreys, 1996, Joyce and Rossion, 2005, Kovacs et al., 2006, Schendan et al., 1998). Most theories propose that the N170 (and its magnetoencephalographic counterpart, the M170) indexes perceptual processes, such as the perceptual categorization of faces, implemented in ventrolateral occipito-temporal regions (e.g., Bentin et al., 1996, Gao et al., 2012, Jeffreys, 1996, Joyce and Rossion, 2005, Schendan et al., 1998). However, the interstimulus variance (ISV) account recently challenged these theories, arguing that faces elicit larger average N170s because the stimuli used in the object conditions are visually more variable than those used in the corresponding face conditions (Thierry et al., 2007a). The main proposed mechanism is that such variability would increase inter-trial jitter in the latency of the N170, decreasing its average size for objects (Regan, 1989). This account also proposed that the earlier P1 component, not the N170, is sensitive to categorical differences between faces and objects.

Although the ISV account has been critiqued thoroughly (e.g., Bentin et al., 2007, Rossion and Jacques, 2008), to date there has not been a direct empirical test, in part because quantifying interstimulus variance is difficult. Pixel-by-pixel correlation between images (Thierry et al., 2007a) is only one measure of physical variance and ignores perceptual variance, that is, how different stimuli in a set are perceived to differ from each other (Rossion and Jacques, 2008). Thus, findings of any experiment matching specific measures of ISV between stimulus sets could always be criticized for not using the “correct” variance measure. Further, ISV can be either within-participant, the only type considered so far, or between-participant. Within-participant ISV is how stimuli in each category vary for each participant. However, between-participant ISV could also be a factor when different participants see different stimuli, for example because of counterbalancing requirements or because the question of interest involves autobiographical stimuli that differ across participants: Stimuli could be less variable for faces than objects, resulting in different between-participant ISV and in an artifactually larger group N170 for faces than objects.

We bypassed the definitional issues and tested the ISV account directly by eliminating ISV: Participants saw a single face and a single object multiple times. The minimal assumption is that the ISV (physical and perceptual) of a set of identical stimuli must be zero, regardless of how it is measured. In experiment 1, the first condition replicated previous N170 findings using multiple faces and objects from many categories. The second condition used the same stimuli, but participants categorized only one object and one face, which were unique for each participant, and did so multiple times. This condition removes within-participant ISV. However, since each person sees a different face and object, ISV across different participants is still greater for objects than faces: between-participant ISV. Thus, the ISV account predicts that the N170 time courses for the particular face and object seen by each participant will vary more for objects than faces between subjects. In experiment 2, all participants categorized the same object and face, removing both within- and between-participant ISV. Importantly, Fourier amplitude spectra of faces and objects were matched to eliminate low-level visual properties that may affect early ERPs.

The ISV account predicts no N170 amplitude difference between faces and objects when the stimulus sets have zero ISV. Furthermore, it predicts a larger P1 to faces than objects in all conditions. In contrast, all other N170 theories predict N170 face selectivity, regardless of ISV, and no P1 category effect.