The upper-hemifield advantage for masked face processing: Not just an attentional bias
- 697 Downloads
Recent evidence suggests that face processing may be more robust in the upper visual field (UVF) than in the lower visual field (LVF). We asked whether this UVF advantage is due to an upward bias in participants’ visuospatial attention. Participants classified the sex of a UVF or LVF target face that was preceded by a congruent or incongruent masked prime face. We manipulated spatial attention within subjects by varying the predictability of target location across sessions (UVF:LVF ratio of 50:50 on Day 1 and 20:80 on Day 2). When target location was unpredictable, priming emerged earlier in the UVF (~165 ms) than the LVF (~195 ms). This UVF advantage was reversed when targets were more likely to be presented in the LVF. Here priming arose earlier for LVF targets (~53 ms) than UVF targets (~165 ms). Critically, however, UVF primes were processed to the same degree regardless of whether spatial attention was diffuse (Day 1) or deployed elsewhere (Day 2). We conclude that, while voluntarily directed spatial attention is sufficient to modulate the processing of masked faces in the LVF, it is not sufficient to explain the UVF advantage for masked face processing.
KeywordsAttention Faces Vertical asymmetry Upper visual field Lower visual field
Human visual perception is not uniform across the retinal field, but, rather, is characterized by perceptual asymmetries arising from the brain’s preferential response to particular stimulus types at different retinal locations. Investigations of vertical asymmetry in visual perception have historically favored the lower visual hemifield relative to the upper visual hemifield. For example, spatial resolution (Talgar & Carrasco, 2002) and contrast sensitivity (Carrasco, Penpeci-Talgar, & Cameron, 2001; Skrandies, 1987) are enhanced in the lower visual field (LVF) compared to the upper visual field (UVF), with this asymmetry being most pronounced at the vertical meridian (Carrasco et al., 2001; Talgar & Carrasco, 2002). Performance is also typically better in the LVF on tasks of hue discrimination (Levine & McAnany, 2005), perception of illusory contours (Rubin, Nakayama, & Shapley, 1996), motion processing (Amenedo, Pazo-Alvarez, & Cadaveira, 2007; Edwards & Badcock, 1993; Levine & McAnany, 2005), global processing (Christman, 1993), and perception of coordinate spatial relationships (Niebauer & Christman, 1998). There are select tasks which exhibit a UVF advantage, however, including visual search (Chaiken, Corbin, & Volkmann, 1962; Fecteau, Enns, & Kingstone, 2000; Previc & Blume, 1993; Previc & Naegele, 2001; Yund, Efron, & Nichols, 1990), local processing of hierarchical stimuli (Christman, 1993), perception of apparent distance (Levine & McAnany, 2005), and categorical judgments of position (Niebauer & Christman, 1998).
While vertical visual asymmetries are well documented for a range of low-level perceptual stimulus types, it is only recently that researchers have extended this line of enquiry to include higher level visual representation. Kravitz, Kriegeskorte, and Baker (2010) used functional magnetic resonance imaging (fMRI) to show that object representation in several object selective regions is constrained by vertical position. Critically for the present case, they observed a UVF bias for object representation in the left and right posterior fusiform areas, in that split-half correlations (Haxby et al. 2001) for these regions were strongest for object line drawings presented in the upper hemifield. Consistent with this report, there is now a burgeoning body of work which suggests the processing of faces – arguably one of the most important types of object we encounter in the visual world – may enjoy an upper-hemifield advantage. For example, Felisberti and McDermott (2013) have reported that participants are able to recognize previously seen faces better when those faces are initially encoded in the upper hemifield rather than the lower hemifield. Within the neurophysiological literature, Liu and Ioannides (2010) have shown that magnetoencephalography (MEG) peaks elicited by emotional faces in regions such as the medial prefrontal cortex (MPFC), left fusiform face area (FFA), and left occipital face area (OFA) arise earlier for faces presented in the UVF compared to the LVF. Most recently, we used masked priming to show that that the UVF exhibits a clear behavioral advantage for face-sex categorization (Quek & Finkbeiner, 2014a). Participants in our study categorized the sex of a target face by reaching to either the left or right edge of a computer monitor. The target face could appear in either the UVF or LVF and was preceded by a nonconscious masked prime face of either the same or the opposite sex (i.e., congruent or incongruent). Despite being outside participants’ conscious awareness, these masked prime faces nevertheless exerted a strong influence on participants’ classification of the target faces, in that the participants reached towards the correct response panel faster when the prime and target were associated with the same sex-categorization response (i.e., the congruent condition). Importantly, we found that this index of nonconscious face processing (termed the masked congruence effect, or MCE) arose earlier in time when the prime and target faces appeared in the UVF compared to the LVF. Moreover, we observed a differential benefit of attentional cueing between the vertical hemifields. Where the MCE elicited by faces in the lower hemifield depended on a valid spatial cue (i.e., no priming without spatial attention), the emergence of the MCE in the upper hemifield was unaffected by spatial cueing, suggesting that masked prime faces presented above-fixation were processed irrespective of the spatial attention’s locus (see Finkbeiner & Palermo, 2009 for a similar finding with masked UVF primes).
Although not all studies of face perception across the visual field have observed stable biases across individuals (Afraz, Pashkam, & Cavanagh, 2010), on balance the evidence suggests that the visual system is able to process face information more efficiently at locations above fixation compared to below. What is not yet established, however, is why this might be the case. A possible explanation that we explore in the present paper is that the upper-hemifield advantage in face processing arises due to an upward bias in voluntarily directed spatial attention. We are not the first to appeal to the tendency to divide attention unevenly across the visual field as a potential explanation for vertical asymmetry in visual perception. Rezec and Dobkins (2004) proposed that an “attentional weighting” favoring the lower hemifield might explain the LVF advantage they observed in motion and orientation discrimination. Interestingly, however, there is some suggestion that biases in spatial attention may depend on stimulus type. For example, studies involving vertical line bisection (Bradshaw, Nettleton, Nathan, & Wilson, 1985; Drain & Reuter-Lorenz, 1996; van Vugt, Fransen, Creten, & Paquier, 2000), object matching (Chambers, McBeath, Schiano, & Metz, 1999), and mental scene representation (Drummond & Tlauka, 2012) suggest an upward bias in spatial attention. Importantly for the present case, there are good reasons to think participants may be particularly prone to favoring the UVF in the context of face perception, as this region of space often carries important face information during real-world interactions. We tend to encounter human faces more frequently in the UVF, as this region corresponds to extrapersonal space (Previc, 1990). Eye gaze, a potent source of social information which guides adaptive behavior (Hood, Willen, & Driver, 1998; Langton, Watt, & Bruce, 2000), is also typically contained within the top half of our visual field. Thus, it is reasonable to think that participants in our previous study might have voluntarily attended to the upper hemifield even while maintaining central fixation, resulting in both the earlier emergence of priming in the (attended) UVF, and the lack of an exogenous cueing effect in this region (Quek & Finkbeiner 2014a).
If an upward bias in voluntarily directed spatial attention does indeed explain the UVF face processing advantage, then manipulating participants’ endogenously oriented spatial attention away from the UVF (i.e., towards the LVF) should attenuate the UVF advantage evident in the timecourse of the MCE for faces. To test this hypothesis, we adapted our previous design known to yield both robust masked priming effects and a clear UVF advantage for face-sex categorization (Quek & Finkbeiner 2014a). We manipulated participants’ endogenous attention to the vertical hemifields by varying the probability of UVF and LVF targets across days. On Day 1, the ratio of UVF targets to LVF targets was 50:50, such that target location was entirely unpredictable for participants. In contrast, on Day 2, targets appeared in the LVF on 80 % of trials (20:80 UVF to LVF ratio). We reasoned that participants would be sensitive to this increased probability of LVF targets and would consequently direct their spatial attention towards the lower hemifield after a period of learning. Our predictions were as follows: For Day 1 (50:50 target location ratio), we expected the MCE would be superior for UVF prime-target pairs than for LVF prime-target pairs, replicating the UVF advantage for face-sex categorization we have shown previously (Quek & Finkbeiner 2014a). For Day 2 (20:80 target location ratio), we predicted that masked face processing below fixation would be facilitated by participants voluntarily directing spatial attention towards this hemifield. That is, we expected the priming effect for the LVF would be equivalent to, or perhaps even superior to, the UVF priming effect from the same day. The question of critical interest was what impact our manipulation of endogenous attention on Day 2 would have on the priming effect for the UVF. If the documented UVF advantage for subliminal face processing should be attributed to an upward bias in participants’ spatial attention, then the MCE elicited by UVF faces should suffer when participants direct their spatial attention to a different location. That is, in the measure to which the UVF advantage is due to focused spatial attention, then the redeployment of that focused spatial attention to the LVF should yield both increased performance in the LVF and a cost to the priming effect in the UVF.
Before undertaking this planned design, however, we wanted to first take account of the fact that our manipulation of target location probability across days would not allow us to counterbalance the order of conditions across participants. All participants would need to complete the conditions in a fixed order (50:50 on Day 1, 20:80 on Day 2), to ensure that the location probability learned during the 20:80 condition would not carry over and influence participants’ approach to the 50:50 version of the task. Consequently, we are faced with the possibility of an artefact in this design: namely, participants would be highly practiced during the 20:80 condition on Day 2, and less so during the 50:50 condition on Day 1. Thus, if we were to observe the predicted enhancement of the priming effect in the 20:80 condition on Day 2, one could argue that this was due to participants' performance simply improving over time as opposed to the redeployment of spatial attention. To rule out this possible explanation of our predicted results, before embarking on the experiment proper we ran an initial control experiment in which the target location probability was held constant at 50:50 across both testing days. If the properties of the masked congruence priming in the LVF are sensitive to the effects of practice, then we would expect to see differences in either the magnitude or onset of the MCE (or both) across days. To anticipate our results of this initial control experiment, the MCE did not vary across days in either hemifield, suggesting this index of face processing to be robust to the effects of practice.
Our goal in Experiment 1 was to determine whether the masked congruence priming effect (particularly in the LVF) would improve as a function of exposure to the task. Participants attended two separate testing sessions within a 3-day period (i.e., the second session occurred either 1 or 2 days after the first). The task on each day was identical: classify the sex of the target face by reaching out to touch a response panel at the left or right edge of the screen. We used a fully crossed factorial design with the factors Visual Field (UVF vs. LVF) and Prime Type (congruent vs. incongruent). Importantly, this design was identical to the one planned for Experiment 2 (e.g., same number of trials, etc.), save that the predictability of the target’s location did not vary across days. That is, the ratio of UVF to LVF targets was 50:50 on both Day 1 and Day 2. If the MCE is indeed sensitive to the effects of practice, then we should observe a stronger priming effect on Day 2 compared to Day 1.
Experiment 1 methods
A group of 14 individuals (seven males) gave informed consent to participate in Experiment 1 for financial compensation. All participants had normal or corrected-to-normal vision, and were identified as strong right-handers using the Edinburgh Handedness Inventory (Oldfield, 1971). The mean age was 26.5 years.
Stimuli and apparatus
Targets were five neutral faces of each sex drawn from the Psychological Image Collection at Stirling database (PICS, http://pics.psych.stir.ac.uk/), cropped to exclude the facial contour. No face contained obvious sex indicators (e.g., facial hair). Distractors were ten animal images presented within a similar cropped oval, one of which was selected at random on each trial. We used the SHINE toolbox written for Matlab to adjust the ten target and ten distractor items so that their mean luminance and contrast values were comparable (Willenbockel, Sadr, Fiset, Horne, Hosselin, & Tanaka, 2010). We then used the same procedure to prepare an additional two faces (one male, one female) and two additional animal images to serve as masked primes and masked foils, respectively. These stimuli were considered “novel” in that they never appeared under unmasked conditions. The backward mask on each trial was randomly selected from a group of ten possible scrambled face images of the same mean luminance and contrast. All stimuli were 75 × 100 pixels and subtended 3.44 × 4.58° of visual angle from a viewing distance of 65 cm. Response panels marked “M” and “F” remained at the left and right edge of the screen throughout the trial; we counterbalanced the order of these positions across participants. We recorded each participant’s reaching trajectories by fitting the their right hand with two light-emitting diode (LED) markers and sampling the position of these markers every 5 ms with an OptotrakCertus® motion capture system.
Procedure and design
On each trial, the participant had over 3 s to freely adjust their reaching trajectory and finalize their classification of the target by touching either the left or right response panel. However, although participants could complete the reaching movement in their own time, we imposed strict parameters regarding when they should initiate the movement. We required participants to begin reaching in response to an auditory go-signal on each trial – the third tone in a series of three ascending beeps (see Finkbeiner, Coltheart, & Coltheart, 2014; Quek & Finkbeiner, 2014a). The position of this go-signal varied in time with respect to target onset, such that on any trial the third beep could occur at one of five randomly selected durations following target onset (0 ms, 75 ms, 150 ms, 225 ms, or 300 ms). We defined Movement Initiation Time (MIT Latency) as the time in milliseconds from target onset until the finger’s velocity reached 10 cm/s. We allowed MIT latencies up to 100 ms before and 200 ms after the beep signal. If the participant began their reaching movement outside this critical window, they received negative auditory (a loud buzz) and visual feedback (e.g., “Too Early!”), and the trial was aborted and cached for re-presentation at the end of the block. Trials with response-window failures during the re-presentation stage were removed from all subsequent analyses (11.37 % of all trials).
As in our previous study, we assessed the visibility of the masked prime items using a two-alternative forced-choice (2AFC) prime-detection task (conducted at the conclusion of the Day 2 testing session). This prime-visibility test included 160 trials which were identical to those used in the experiment proper, save that after reaching out to classify the target face, participants saw two faces (the real prime and a lure) and had to indicate which of them had been the prime on that trial. This prime-identification response was untimed, and position of the real prime and lure was counterbalanced across trials. Importantly, we instructed participants to maintain the same strategy for these prime-identification trials as they used in the experiment proper.
To prepare the reaching trajectories for analysis, we selected the 450 xyz samples between the points corresponding to 100 ms before movement onset, and 2150 ms after movement onset.1 At each sample within this epoch, we calculated x-velocity – a signed value which reflects the finger’s velocity along the x-axis. Because participants indicate their sex-categorization decision along this left-right dimension (e.g., left for male; right for female), x-velocity at any given sample represents the finger’s velocity in the correct direction. Positive x-velocity values indicate the finger is moving towards the correct response panel, and negative x-velocity values indicate that the finger is heading away from the correct response panel (i.e., in the incorrect direction). We smoothed the x-velocity profiles using a modified version of the Orthogonal Polynomial Trend Analysis (OPTA) procedure (Karayanidis, Provost, Brown, Paton, & Heathcote, 2011; Woestenburg, Verbaten, & Slangen, 1983), the details of which have been extensively reported elsewhere (Finkbeiner, Coltheart, & Coltheart, 2014; Quek & Finkbeiner, 2013, 2014a). As a final step, we averaged across the initial 300 ms of the reaching movement to produce a single representative value for each trial termed initial x-velocity. We limited our dependent measure to the initial segment of the trajectory because this initial portion best reveals the motor plan that participants had formulated just prior to movement initiation. This is important since we are interested in comparing the way in which participants’ responses develop with stimulus-viewing time across different experimental manipulations. In the present case, the masked congruence effect (MCE) is reflected in higher initial x-velocities on congruent trials than incongruent trials. This pattern is analogous to the MCE reflected in response-time (RT) data (e.g., Finkbeiner & Palermo, 2009; Naccache, Blandin, & Dehaene, 2002), in that just as participants in button-press priming studies tend to press the correct button faster on congruent trials than incongruent, here participants move in the correct direction faster when the prime and target are associated with the same classification response.
We analyzed all data using custom software written in R (www.r-project.org), implementing linear mixed-effects modelling (LMM; cf. Baayen, Davidson, & Bates, 2008; Bates, 2005) using the lmer4 package (http://lme4.r-forge.r-project.org, Bates, Maechler, & Bolker, 2011). We evaluated the reliability of each effect of interest using an incremental model comparison procedure in which we used goodness-of-fit statistics (AIC, BIC, and Log Likelihood values; see Akaike, 1974; Schwarz, 1978) to determine which of two models fit our data better – a model which included the term under inspection, or the same model without this term. For each comparison, we selected the model which (a) minimized AIC and BIC, and (b) maximized the Log Likelihood. Below we report the results of this Likelihood ratio test and, where appropriate, also report the coefficients, standard errors (SE), and t-values for terms included in the final model selected. Our criterion for significance for individual fixed effects was an absolute t ratio of 2.0, as per Kliegl, Masson, and Richter (2010). Although the degrees of freedom are not known exactly in LMM, the very large number of observations in the datasets used here and elsewhere mean that the t distribution converges to the normal distribution. Thus, the criterion cutoff of two SEs corresponds well to the .05 significance criterion (see Finkbeiner et al., 2014; Kliegl et al, 2010; Masson & Kliegl, 2013; Quek & Finkbeiner, 2013, 2014a).
Experiment 1 results
To assess prime visibility, we used the 2AFC prime detection data to calculate a hit rate, false alarm rate, and d’ value for each participant. The mean d’ value was 0.07. A one-sample t-test confirmed that this was not significantly different from zero (t(13) = 0.86, p = .404, Pearson’s r = .23), suggesting participants in Experiment 1 had minimal awareness of the masked prime faces.
Experiment 1 discussion
We observed a strong MCE in Experiment 1, in that participant’s classification reaching movements were both more accurate and more efficient for targets preceded by congruent primes compared to incongruent primes (i.e., same rather than opposite sex). We also found clear evidence of a UVF advantage for face-sex classification, in that participants were more accurate in categorizing the sex of UVF targets compared to LVF targets. Moreover, although both UVF and LVF prime-target pairs yielded a significant MCE in participants’ reaching data, the magnitude of this priming effect was reliably larger in the UVF. This suggests participants were able to process the sex information carried by the masked prime more efficiently when the prime appeared in the upper hemifield compared to the lower hemifield. Taken together, these results provide strong evidence for the emerging position in the literature that face processing, and in particular face-sex categorization, is superior in the upper hemifield relative to the lower hemifield (Liu & Ioannides, 2010; Quek & Finkbeiner, 2014a). In addition, and importantly for Experiment 2, we also found that the priming effect reflected in both participants’ accuracy and initial reaching movements did not differ significantly between testing days for either hemifield. This suggests that participants’ ability to extract the task-relevant sex information from masked faces did not improve as a function of exposure to the task (i.e., from the first to the second testing day).
Armed with the knowledge that the MCE in our paradigm was not sensitive to the effects of practice across days, we were free to pursue the principal goal of the present study – to determine how voluntarily directed spatial attention would modulate the effect of vertical hemifield on masked face priming. Experiment 2 used an identical design to Experiment 1, save that here we manipulated target location predictability across days. On Day 1, the ratio of UVF to LVF targets was 50:50, such that target location was unpredictable to participants (just as in Experiment 1). On Day 2, however, we increased the ratio of LVF to UVF targets so that the target appeared below fixation on 80 % of trials. We reasoned that participants would be sensitive to this increased probability of LVF targets and, after a period of learning, would direct their attention to the lower hemifield as a consequence. We predicted that covertly attending to the LVF would facilitate the priming effect in this hemifield, resulting in a stronger MCE for the LVF on Day 2 than on Day 1. Because we had already confirmed that the MCE in the LVF did not improve with task exposure in Experiment 1, we felt we would be able to attribute any improvement in the MCE in the LVF across testing days to our manipulation of target location probability, rather than simple practice effects. Of critical interest was what impact this manipulation would have on participants’ ability to process masked prime faces above fixation. If the UVF advantage we and others have observed for face processing really is underpinned by an attentional bias towards the upper hemifield, then a redeployment of spatial attention to the LVF should yield an enhanced MCE in the LVF and a diminished MCE in the UVF.
Experiment 2 methods
We recruited a different group of 16 Macquarie University undergraduate students (seven males) to participate in Experiment 2 for course credit. All participants had normal or corrected-to-normal vision, and were identified as strong right-handers using the Edinburgh Handedness Inventory (Oldfield, 1971). The mean age was 18.81 years.
Procedure and design
We used the same stimuli, apparatus, trial structure, and testing procedure as described for Experiment 1. As before, we required participants to initiate their reaching movement in response to an auditory go-signal (the third beep in a series) which could occur either 0 ms, 75 ms, 150 ms, 225 ms, or 300 ms after target onset. Participants completed 240 trials on Day 1 in which the prime and target faces appeared with equal probability (50 %) in either the upper or lower panel. On Day 2, we included the same number of UVF trials as on Day 1 (i.e., 60 UVF-congruent, 60 UVF-incongruent), and included 480 LVF trials to yield a 20:80 ratio of UVF to LVF trials. There were 80 practice trials excluded from later analyses on Day 1, and 40 on Day 2. We assessed prime visibility at the end of the experiment proper on Day 2 using the same 2AFC task described for Experiment 1, in which we maintained the 20:80 UVF to LVF target ratio.
Experiment 2 results
As in Experiment 1, we calculated initial x-velocity by averaging the first 300 ms of each smoothed x-velocity profile. Below we subject these initial x-velocities to statistical analysis, both (a) averaged across target viewing time and then (b) as a function of target-viewing time.
Analyses averaging across target-viewing time
To model initial x-velocity collapsing across target viewing times, we firstly verified that a model including both random intercepts for Participant and random slopes between Participant and Prime Type fit the data better than a model with only random intercepts χ2(2) = 2354.22, p < .001. There was a significant fixed effect of Prime Type, χ2(1) = 13.97, p < .001, in that initial x-velocity was higher for congruent trials compared to incongruent trials (b = −27.82, SE = 7.33, t = −3.80). The fixed effects of Visual Field, χ2(1) = 83.00, p < .001, and Day, χ2(1) = 243.48, p < .001, also improved the model’s fit, with initial x-velocity being higher on average for UVF trials compared to LVF trials (b = −3.05, SE = 2.08, t = −1.46), and on Day 2 compared to Day 1 (b = 14.67, SE = 2.10, t = 7.00). Regarding two-way interactions, including the Prime Type × Visual Field interaction did not improve the model’s fit, χ2(1) = 1.82, p = .177; however, both the Prime Type × Day and Visual Field × Day interactions did, χ2(1) = 42.29, p < .001 and χ2(1) = 5.81, p < .05, respectively. Critically, and in contrast to Experiment 1, including the three-way interaction between Prime Type, Visual Field, and Day also significantly improved the model of Experiment 2 initial x-velocities.
For interest’s sake, we also examined another aspect of the three-way interaction between Prime Type, Visual Field, and Day by modelling the data separately for Day 1 (50:50) and Day 2 (20:80). To summarize these analyses briefly, when target location was unpredictable (i.e., Day 1), there was a significant interaction between Prime Type and Visual Field which indicated the MCE to be larger in the UVF than the LVF (28 mm/s vs. 20 mm/s), χ2(1) = 7.91, p < .005. In contrast, on Day 2 this interaction was not reliable, χ2(1) = 1.52, p = .218, indicating that the magnitude of the MCE did not differ significantly between the hemifields on Day 2 when participants could expect the target to appear more frequently in the LVF (34 mm/s in the UVF vs. 36 mm/s in the LVF).
Analyses taking target-viewing time into account
LMM analysis for Day 1 indicated significant fixed effects of MIT Quantile and Prime Type (p < .001 in both cases), but not Visual Field, χ2(1) = 0.58, p = .448. We nevertheless retained this factor in the model so as to inspect its interaction with other factors. Next, we verified the inclusion of the two-way interaction between Prime Type and MIT Quantile and determined that this was further qualified by Visual Field (p < .001 in both cases). Because we obtained the critical Prime Type × Quantile × Visual Field interaction, we were able to conduct a paired t-test between the congruent and incongruent conditions at each MIT Quantile, separately for the upper and lower hemifields. In the UVF (see Fig. 9A), the FDR-corrected p-values were significant from the 6th to the 20th MIT Quantile – indicating that a participant’s initial reaching movements reflected a significant MCE from 165 ms of target-viewing time onwards. In contrast, primes presented in the LVF did not elicit a reliable MCE until the 8th MIT Quantile, at around 195 ms of target-viewing time (see Fig. 9C). Thus, on Day 1, when target location was unpredictable for participants, we observed a clear UVF advantage for masked face processing, with the MCE emerging some 30 ms earlier in stimulus-processing time for UVF face stimuli than for LVF face stimuli.
LMM analysis for Day 2 confirmed the same fixed effects and interactions as observed for Day 1 (p < .001 in all model comparison cases). Most importantly, the significant three-way interaction indicated that emergence of the MCE across MIT Quantile depended on Visual Field. As for Day 1, we further examined this interaction by conducting a paired t-test comparison of the congruent and incongruent conditions at each MIT Quantile, separately for each hemifield. As can be seen in Fig. 9B, the MCE in the UVF was reliable from the 6th to the 20th MIT Quantile, or from approximately 165 ms of target-viewing time onwards. In contrast, the MCE elicited by prime-target pairs presented in the LVF was reliable at all MIT Quantiles, from around 53 ms of target-viewing time onwards (see Fig. 9D). Thus, on Day 2, in which participants could reliably anticipate the target’s location and direct their attention appropriately in response, we observed an LVF advantage for masked face processing. Here primes presented in the LVF elicited a priming effect at an earlier stage of stimulus processing than primes presented in the UVF – the opposite pattern to that observed on Day 1, when target location was unpredictable. Of critical importance for present purposes is the comparison between Day 1 and Day 2 for the UVF. Looking across Figs. 9A and B, it is clear that face processing efficiency in the UVF did not suffer when participants covertly attended to the LVF. That is, the MCE in the UVF emerged at the same stage of stimulus-processing on both Day 1 and Day 2 (i.e., around 165 ms of target-processing time), suggesting that directing participants’ covert attention away from the upper hemifield did not interfere with participants’ ability to process masked faces in this region.
As in Experiment 1, to assess prime visibility we calculated d’ from each participant’s 2AFC prime detection data. Here the mean d’ value was 0.11 (SD = 0.13), which a one-sample t-test determined was significantly greater than zero (t(15) = 3.37, p < .01, Pearson’s r = .66). This suggests that the participants in Experiment 2 may have had some partial awareness of the masked prime items, likely owing to our manipulation of target location predictability. That is, because participants voluntarily attended to the lower hemifield in response to the increased probability of targets in this region, masked primes presented in the LVF were likely to have been subject to focused spatial attention.
Experiment 2 discussion
Just as in Experiment 1, we observed a strong MCE in both participants’ accuracy and initial reaching movements. We also replicated the UVF advantage for face-sex categorization in that when participants could not reliably predict the target’s location (i.e., on Day 1), the magnitude of the MCE reflected in the initial stages of their reaching movement was reliably larger for UVF prime-target pairs than for LVF pairs. Moreover, the analysis that took target-viewing time into account revealed that the MCE under unpredictable target location conditions emerged earlier during stimulus-processing time when the face stimuli appeared in the UVF compared to the LVF (~165 ms vs. ~195 ms of stimulus-processing time). That masked prime items influenced the target classification response at an earlier stage of stimulus-processing when presented above – compared to below – fixation would suggest that face-sex categorization is more efficient in the UVF than the LVF.
The effect of vertical hemifield on masked-face processing was very different, however, when participants could reliably expect the target to appear in the lower half of the display and could direct their spatial attention appropriately in response (i.e., on Day 2). Here accuracy was better for LVF targets than for UVF targets. Moreover, the magnitude of the MCE in the LVF was significantly larger on Day 2 than on Day 1. This suggests that voluntarily directing their spatial attention towards the LVF improved participants’ ability to process masked prime faces presented in this region. This interpretation is further supported by the finding that LVF primes influenced the initial stages of target classification at a much earlier point during stimulus-processing when participants voluntarily attended to the lower hemifield (i.e., on Day 2 compared to Day 1). By itself, this result is perhaps not altogether surprising – indeed, there is good reason to expect that attending to the prime’s location should facilitate processing and increase its ability to the response to the target (see Quek & Finkbeiner, 2013). What is surprising, however, is that the increase in performance for LVF faces was not coupled with a commensurate decrease in performance for UVF faces. Target classification accuracy for UVF targets was comparable across testing days, and the MCE in the UVF was neither diminished nor delayed on Day 2 compared to Day 1, emerging at the same time during stimulus processing on both testing days (~165 ms). Taken together, these results suggest that while directing covert spatial attention to the lower hemifield clearly confers a benefit on the processing of LVF faces, face processing in the UVF appears to be unaffected by participants’ deployment of spatial attention to the opposite hemifield.
This paper establishes three important findings. First, we have demonstrated that under unpredictable target location conditions, face-sex categorization exhibits a clear UVF advantage. We observed this robust advantage in two separate groups of participants in both the magnitude and the timecourse of the MCE. Participants in Experiment 1 also showed this effect in sex-categorization accuracy. These results replicate an earlier study (Quek & Finkbeiner, 2014a) and they are also highly consistent with findings from the neurophysiological literature showing that faces presented above fixation activate regions such as the left fusiform face area (FFA), left occipital face area (OFA), and medial prefrontal cortex (mPFC) earlier than faces presented below fixation (Liu & Ioannides, 2010). Taken together, these findings strongly suggest that the visual system processes faces more efficiently in the upper hemifield than in the lower hemifield.
Second, we have shown that voluntarily directed spatial attention reliably modulates the ability of masked faces to influence the overt response to the target face. Masked primes in our study influenced the target classification response to a greater extent, and at an earlier stage of stimulus-processing, when participants could predict the likely location of the upcoming target and direct their attention to this location. This finding represents an important contribution to the masked priming literature, in that as far as we are aware, it is the first demonstration that endogenously-directed spatial attention modulates the masked priming effect for faces. The finding also echoes and complements our previous demonstrations that exogenously-oriented spatial attention facilitates the processing of masked face primes (Quek & Finkbeiner, 2013, 2014a).
Third, and most importantly, the present paper dispels the possibility that the UVF advantage for face processing is underpinned by an upward bias in voluntarily-directed spatial attention. If this were the case, then masked face processing in the UVF would have been negatively affected when participants voluntarily attended to the opposite hemifield, as they did on Day 2 of our second experiment. Instead, we found that participants’ ability to extract the task-relevant sex information from masked faces in the upper hemifield did not vary as a function of our manipulation of voluntarily directed spatial attention. This key finding strongly undermines the possibility that the UVF advantage for face-sex processing we report here and elsewhere (Quek & Finkbeiner, 2014a) is driven by an upward bias in spatial attentional orienting (see Bradshaw et al., 1985; Drain & Reuter-Lorenz, 1996; van Vugt et al., 2000). The implications of these findings for the masked priming, face perception, and broader object recognition literatures are discussed below.
The upper-hemifield advantage for face processing: Not a product of attentional bias
Although vertical asymmetry in face perception has received far less attention than the well-documented laterality effects for faces, there is nonetheless increasing evidence to suggest that face processing may be superior in the upper hemifield relative to the lower hemifield (Coolican, Eskes, McMullen, & Lecky, 2008; Kessler & Tipper, 2004; Liu & Ioannides, 2010). In a recent study, we provided what might be considered the strongest evidence to date for a UVF advantage in face perception, demonstrating that masked faces affect the sex-categorization of a target face at an earlier stage of stimulus processing when presented in the UVF compared to the LVF (Quek & Finkbeiner, 2014a). In the present study, we have replicated this finding in two separate groups of individuals, providing strong support for the existence of an upper hemifield advantage in face-sex processing. The discovery has significant implications for the face-perception literature, as where researchers have long tried to take account of laterality artefacts in their designs, thus far there appears to have been little consideration given to the impact of vertical anisotropy in face-perception tasks. The finding also qualifies the characterization of sex categorization as a highly robust instance of object recognition (Bruce & Young, 1998; Wild et al., 2000). While this process is unquestionably efficient, we have highlighted that this aspect of face perception is nevertheless still subject to modulation by factors such as spatial location and spatial attention. A natural extension of this finding, and one that is already hinted at in the literature, would be to examine whether other aspects of face perception, such as identity or emotion recognition, exhibit a similar upper hemifield advantage (Felisberti & McDermott, 2013; Kessler & Tipper, 2004; Liu & Ioannides, 2010). Given that some have suggested that visual field biases for face processing are distinct to individual participants (Afraz et al., 2010), another intriguing line of future enquiry is how individual differences in visual field shape (Fortenbaugh, Silver, & Robertson, 2015) might influence the UVF advantage for faces we have documented here.
Importantly, although the mechanism underlying this vertical asymmetry for faces remains unclear, we have provided strong evidence here to discredit the possibility that the upper hemifield advantage for face processing is driven by an upward bias in voluntarily directed spatial attention. When participants’ directed their spatial attention away from the upper hemifield, as they did when targets appeared in the lower visual field on 80 % of the trials, their ability to process faces presented in the UVF was unaffected. It is worth noting, however, that voluntarily directed spatial attention is not the only top-down process capable of modulating basic perceptual processing. Where endogenously directed spatial attention prioritises stimulus processing on the basis of task demands, expectation can constrain visual interpretation on the basis of prior knowledge (Summerfield & Egner, 2009) by sharpening sensory representations (Kok, Jehee, & de Lange, 2012). It is well established, for example, that object detection is facilitated by a congruent context (e.g., a coffee cup in a kitchen scene will be found faster than the same cup in a beach scene) (Davenport, 2007; Fize, Cauchoix, & Fabre-Thorpe, 2011). Moreover, Biederman has singled out both probability and spatial position as important relations which affect object recognition in scenes (Biederman, Mezzanotte, & Rabinowitz, 1982). Thus, it is reasonable to think that our wealth of experience encountering faces in extrapersonal space could well have engendered in us a very strong expectation that face stimuli are more likely to appear above fixation than below. Just as expectation of object category is known to increase the efficiency with which category exemplars are processed (Puri & Wojciulik, 2008), presumably a perceptual “prior” which also encompasses spatial location might be expected to facilitate exemplar processing in an anisotropic manner. Importantly, while attention and expectation are often intertwined, it could be the case that a perceptual prior for faces is overlearned and thus relatively less flexible. With this possibility, the probabilistic manipulation of target location used in our second experiment might have effectively guided participants’ voluntarily directed spatial attention, while simultaneously having no effect on their pre-existing expectation. The possibility of a relatively inflexible prior expectation about face location could explain the UVF advantage observed when participants have no cause to preferentially direct their spatial attention to either hemifield (i.e., when target location is unpredictable). This possibility could also explain why face processing remains unaffected in the UVF in experimental contexts in which participants voluntarily direct their attention to the lower visual field. That is, whereas spatial attention can be flexibly controlled and directed to facilitate stimulus processing in predicted target regions, perhaps the well-learned perceptual prior regarding faces remains unchanged, allowing face processing in the expected region to proceed similarly with or without spatial attention. Intriguingly, there is already some indication that the strength of face and body representations in different parts of the visual field appears to be modulated by the frequency with which the stimuli occur in those locations (Chan, Kravitz, Tryuong, Arizpe, & Baker, 2010).
Alternatively, it may be that a structural account is best equipped to explain why face processing exhibits a UVF advantage. As others have already alluded to (Kravitz et al., 2010; Kravitz, Saleem, Baker, Ungerleider, & Miskin, 2013), since the representation of the upper hemifield in early visual cortex is predominantly contained below the calcarine sulcus (Sereno et al., 1995; Zeki, 1969), it may be that presenting faces in the UVF enables more efficient transfer of information to the ventral object recognition systems contained in the temporal lobe, such as the posterior fusiform (Ishai, Ungerleider, Martin, Schouten, & Haxby, 1999; Kravitz et al., 2010; Martin, Wiggs, Ungerleider, & Haxby, 1996). Conversely, transfer to these ventral regions may take slightly longer from the dorsal regions of striate and extrastriate cortex, where information presented to the LVF is initially projected. We note, however, that at present this account of the UVF advantage in face perception remains speculative – it is clear that there are still important steps to be taken in furthering our understanding of vertical asymmetries in object-recognition. Perhaps the foremost question of interest is whether the UVF advantage is specific to face processing, or whether it may in fact extend to other forms of object recognition, as early reports suggest it could (see Quek & Finkbeiner, 2014b).
Endogenous attention modulates masked face processing
Where the modulatory effects of voluntarily directed spatial attention on the neural response to face stimuli are well described in the neurophysiological literature (Holmes, Kiss, & Eimer, 2006; Holmes, Vuilleumier, & Eimer, 2003; Pessoa, McKenna, Gutierrez, & Ungerleider, 2002; Vuilleumier, Armony, Driver, & Dolan, 2001; Wojciulik, Kanwisher, & Driver 1998), behavioral investigations to date have often failed to observe an impairment in task performance resulting from directing spatial attention away from a face’s location. For example, Reddy and colleagues showed that neither face identification nor face-sex classification of peripheral masked faces suffer when spatial attention is engaged centrally by a demanding discrimination task (Reddy, Reddy, & Koch, 2006; Reddy, Wilken, & Koch, 2004). At first glance, it may appear that our findings do not support the conclusions of Reddy and colleagues. This is because we have shown that face perception does indeed benefit from endogenously directed spatial attention, just as it does from exogenously captured attention (Quek & Finkbeiner, 2013). But we suggest, in fact, that our results are consistent with those of Reddy and colleagues. Just as participants in their studies were able to reliably identify both identity and sex even when attention was directed elsewhere, participants in our study remained perfectly able to extract the sex-information from masked face primes in the UVF when attention was directed the LVF.
There is increasing evidence to suggest that face perception is supported better at above-fixation locations than at below-fixation locations. The present study further establishes this finding by showing that face-sex categorization is more efficient in the UVF than the LVF. It has been suggested that the UVF advantage might be due to participants possessing a strong upward attentional bias in categorization tasks. The purpose of this study was to test this possibility. Our primary conclusion is that the UVF advantage is not due to an upward bias in voluntarily directed spatial attention. Specifically, masked faces presented in the upper hemifield engaged cognitive processes to the same degree even when spatial attention had been successfully deployed elsewhere.
We included the 100 ms leading up to movement onset to ensure that we considered the very start of the reaching movement on each trial. For instances in which the participant’s reaching movement concluded sooner than 2,150 ms after movement onset, we repeated the xyz coordinates from the final sample to make up the full number of samples for that trial.
The authors are grateful to Irene Chork for her assistance in collecting behavioral data. This work was supported in part by an Australian Research Fellowship (FT120100830) to MF from the Australian Research Council.
- Bates, D. (2005). Fitting linear mixed models in R. R News, 5, 27–30.Google Scholar
- Bates, D., Maechler, M., & Bolker, B. (2011). Linear mixed-effects models using S4 classes. R package version 0.999375-42, http://lme4.r-forge.r-project.org/ .
- Bruce, V., & Young, A. (1998). In the eye of the beholder. Oxford: Oxford University Press.Google Scholar
- Previc, F. H. (1990). Functional specialization in the lower and upper visual fields in humans: Its ecological origins and neurophysiological implications. Behavioral and Brain Sciences, 13, 519–575.Google Scholar
- Sereno, M. I., Dale, A. M., Reppas, J. B., Kwong, K. K., Belliveau, J. W., Brady, T. J., . . . Tootell, R. B. H. (1995). Borders of Multiple Visual Areas in Humans Revealed by Functional Magnetic Resonance Imaging. Science, 268(5212), 889–893.Google Scholar
- Wild, H. A., Barrett, S. E., Spence, M. J., O’Toole, A. J., Cheng, Y. D., & Brooke, J. (2000). Recognition and sex categorisation of adults’ and children’s faces: Examining performance in the absence of sex-stereotyped cues. Journal of Experimental Child Psychology, 77(4), 269–291.PubMedCrossRefGoogle Scholar