Rosenholtz (Reference Rosenholtz2024) argues that summary statistics, processed through peripheral vision, can account for various attentional phenomena, such as change blindness and object-based attention. While we broadly agree, we challenge the notion that summary statistics and selective attention are alternative mechanisms necessitating a “paradigm shift.” Instead, we propose that they are inherently interdependent – extracting summary statistics from a visual display often requires selective attention, depending on task demands and display complexity.

For instance, de Fockert and Marchant (De Fockert & Marchant, Reference De Fockert and Marchant2008) found that when observers estimated the average size of a set of circles, they did not always integrate information across the entire set. Instead, their estimates were often biased by the features of the more attended item or based on a smaller subset of items. Similarly, Myczek and Simons (Reference Myczek and Simons2008) demonstrated that apparent evidence for mean size perception could be explained by various focused-attention strategies, such as sampling only one or two items from the set or averaging the largest and smallest items. Therefore, just as summary statistics can explain visual perception beyond focused attention, selective attention contributes to the formation of summary statistics under limited capacity.

Summary statistic processing, guided by attention, operates across diverse contexts, suggesting a generalized mechanism in which selective attention plays a key role in extracting summary information. For example, when participants estimated the average color of a set of colored objects, their judgments exhibited a “robust averaging” effect, in which the impact of the most extreme colors in the spectrum of the set on the average estimation was underweighted – potentially as a strategy to enhance the signal over noise (de Gardelle & Summerfield, Reference De Gardelle and Summerfield2011). In contrast, the opposite pattern emerged when participants estimated the average of numerical values: extreme values received greater weight (Rosenbaum, de Gardelle, & Usher, Reference Rosenbaum, de Gardelle and Usher2021; Spitzer, Waschke, & Summerfield, Reference Spitzer, Waschke and Summerfield2017; Tsetsos, Chater, & Usher, Reference Tsetsos, Chater and Usher2012; Vanunu et al., Reference Vanunu, Hotaling and Newell2020; Vanunu, Pachur, Usher, Reference Vanunu, Pachur and Usher2019), presumably because they capture more attention (Kunar et al., Reference Kunar, Watson, Tsetsos and Chater2017; Vanunu et al., Reference Vanunu, Hotaling, Le Pelley and Newell2021). Further, a similar pattern has been observed in spatial perception, where visually salient objects disproportionately influence centroid estimations, shifting perceived locations toward them (i.e., “The Weighted Average Illusion”; Hong, Witt, & Szafir, Reference Hong, Witt and Szafir2021).

We argue that while summary statistics reduce information load, they are often derived through selective attention, with their interplay shaped by task demands that would cause processing to exceed capacity if all information were to be used. For example, recent studies have shown that when discriminating between object numerosities – such as determining whether an intermingled array contains more blue or yellow dots – people tend to rely on a subset of information from the display’s center, a phenomenon termed the “proximity-to-center bias” (Vanunu & Ratcliff, Reference Vanunu and Ratcliff2025; Vanunu & Donnelly, Reference Vanunu and Donnelly2024). As a result, there is a bias to report objects near the center as more numerous, even when they are less prevalent across the entire display. Notably, the proximity-to-center bias was absent when participants compared the numerosity of two spatially separated arrays, suggesting that this bias is specific to tasks requiring high visual sensitivity. Because visual sensitivity peaks in the central region of the visual field, participants likely rely on a subsample from the center to distinguish between object types in briefly presented, intermingled displays.

To conclude, we challenge Rosenholtz’s call for a “paradigm shift” and argue that summary statistics cannot replace selective attention (or inattention); rather, they function as complementary mechanisms operating under limited capacity. That is, while summary statistics can explain perception in the absence of focal attention, they are themselves constrained by capacity limits. Thus, rather than serving as an alternative to selective attention, summary statistics and attention work in tandem, shaping visual processing in an interdependent manner.