It is tempting to agree that top-down effects on perception (such as our own, moral pop-out effect; Gantman & Van Bavel Reference Gantman and Van Bavel2014) constitute a radical reinterpretation of a foundational issue (Firestone & Scholl Reference Firestone and Scholl2015b). Unfortunately, we cannot get excited about a notion of perception that excludes attention, inference, prediction, or expectations; that whittles the fascinating and broad domain of perception to sawdust. It is one thing for Firestone & Scholl (F&S) to argue that the entire field of visual neuroscience is irrelevant to understanding human perception. It is quite another to dismiss evidence of re-entrant processing precisely because it is well-established (“whatever one thinks of the relevance,” sect. 2.2, para. 2). That makes F&S's cognitive penetrability argument circular.

F&S define perception to “encompass both (typically unconscious) visual processing and the (conscious) perceptions that result” (sect. 1, para. 3) as they “have the broader aim of evaluating evidence for top-down effects on what we see as a whole” (sect. 1.2, para. 2, emphasis theirs). Yet, they only consider perception that is separable from attention and occurs prior to – and independently from – memory, judgment, and social and physical context. It is difficult to understand how this definition might include “what we see as a whole.” Moreover, their dismissal of unconscious inferences in vision, crossmodal effects, evidence of re-entrant processing in neuroscience, and changes in perceptual sensitivity over time, carves the mind at false joints. Their model of the mind seems to reify the administrative structure of psychology departments, manufacturing natural kinds of perception, cognition, and social processing.

The architecture of the mind does not recognize these distinctions. After perceptual input reaches the retina, multiple brain regions operate on the input – selecting the significant from the mundane (Lim et al. Reference Lim, Padmala and Pessoa2009), often by emotion (Anderson & Phelps Reference Anderson and Phelps2001) or motivation (Egner & Hirsch Reference Egner and Hirsch2005), and some via top-down re-entrant processes (Gilbert & Li Reference Gilbert and Li2013) – to construct perceptual experience. We suggest that the more pertinent question for future research is not whether top-down influences penetrate perception, but rather which components of the perceptual processing stream are sensitive to top-down influence. But perhaps this is a matter of preference.

Thankfully, the crux of F&S's argument lies in their empirical re-explanations of a handful of case studies. These are falsifiable. We invite readers to take a closer look at their case studies, which make strong claims about psychological process. We evaluate F&S's empirical claims regarding the moral pop-out effect and ask whether this exposes a fundamental problem with their case study approach.

We previously reported that moral words were more frequently detected than nonmoral words (matched for length and frequency), only when presented at the threshold for perceptual awareness (termed the moral pop-out effect, Gantman & Van Bavel Reference Gantman and Van Bavel2014; Reference Gantman and Van Bavel2015). The moral pop-out effect occurred over and above measured differences in valence, arousal, or extremity. We concluded that moral words were detected more frequently than control words.

F&S recently claimed that semantic memory must be solely responsible for the moral pop-out effect because the moral words were more related to each other than the control words were. As proof, they claimed to find “entirely analogous” fashion and transportation pop-out effects. Unfortunately there are fundamental flaws in the design and reporting of these studies that make it difficult for us to see how they could draw such strong conclusions.

First, there are some surprising problems with the experimental design of F&S's studies. For example, participants were never randomly assigned to experiments. Without this linchpin of experimental design, it is difficult, if not impossible, to draw any firm inferences about the similarities or differences between moral versus fashion/transportation pop-out effects.

Even more surprising, F&S's inferences rely on merely looking at the studies side by side and judging whether summaries of the results appear similar. Yet, there is good reason to believe these similar behavioral effects (moral vs. fashion/transportation word detection) arise by different processes even at the semantic level. We suspect that moral words are not explicitly encoded in semantic memory as moral terms or as having significant overlapping content. For example, kill and just both concern morality, but one is a noun referring to a violent act and the other is an adjective referring to an abstract property. Category priming is more likely when the terms are explicitly identifiable as being in the same category or at least as having multiple overlapping semantic features (e.g., pilot, airport).

Unfortunately, Firestone and Scholl (Reference Firestone and Scholl2015b) do not present the necessary data to evaluate whether a similar process (specifically, repetition priming, p. 42–43) underlies moral versus fashion/transportation experiments. Although they reported the relevant means for the fashion/transportation experiments – which support their claims that fashion/transportation words show repetition priming effects – they do not report the analogous means for their morality experiment That unfortunate omission seems particularly problematic because the moral perception case study – like all of the presented case studies – is presumably an argument about process.

Any observed behavior can be explained by multiple processes intervening between perceptual input and motor response. A single process rarely explains any behavior, and possible explanations are not always mutually exclusive. Accordingly, it is trivially true that semantic memory plays some role in the moral pop-out effect (how else would our participants know words like kill and die), yet this does not rule out that the motivational relevance of morality (or any motivationally relevant construct) boosts related content to awareness. Accordingly, we do not think morality is modular, either (Van Bavel et al. Reference Van Bavel, FeldmanHall and Mende-Siedlecki2015). For example, we would expect a “pop-out” effect for any motivationally salient construct, such as food-related words when participants are food-deprived (Radel & Clément-Guillotin Reference Radel and Clément-Guillotin2012). We suspect that other case studies suffer from this same failure to consider the multiple determinants of behavior.

Arguments about process – and especially mediation – must examine it directly. Otherwise, dismissing any effect with a 1:1 model of cause-to-behavior seems presumptuous (and even naive). These problems plague F&S's reinterpretation of the moral pop-out effect, and they may well be embedded in their reinterpretation of other case studies.