High visual salience of alert signals can lead to a counterintuitive increase of reaction times

It is often assumed that rendering an alert signal more salient yields faster responses to this alert. Yet, there might be a trade-off between attracting attention and distracting from task execution. Here we tested this in four behavioral experiments with eye-tracking using an abstract alert-signal paradigm. Participants performed a visual discrimination task (primary task) while occasional alert signals occurred in the visual periphery accompanied by a congruently lateralized tone. Participants had to respond to the alert before proceeding with the primary task. When visual salience (contrast) or auditory salience (tone intensity) of the alert were increased, participants directed their gaze to the alert more quickly. This confirms that more salient alerts attract attention more efficiently. Increasing auditory salience yielded quicker responses for the alert and primary tasks, apparently confirming faster responses altogether. However, increasing visual salience did not yield similar benefits: instead, it increased the time between fixating the alert and responding, as high-salience alerts interfered with alert-task execution. Such task interference by high-salience alert-signals counteracts their more efficient attentional guidance. The design of alert signals must be adapted to a “sweet spot” that optimizes this stimulus-dependent trade-off between maximally rapid attentional orienting and minimal task interference.


Supplementary Material, Part 1: Salience models
To verify that our experimental manipulation of visual contrast indeed affects saliency as intended, we use 3 common models of low-level salience to compute salience across the stimuli for the 8 different contrast levels.Specifically, we computed the saliency map S1 , graph-based visual salience (GBVS S2 ) and the spectral salience S3 .For the former two models we used the implementation of Harel and colleagues S2 , for the latter the implementation of Schauerte and Stiefelhagen S4 .As the stimuli are symmetric relative to the image center and none of the models has any off-center spatial bias, we restricted this analysis to alert squares appearing on the left-hand side open to the top and primary task squares open to the right.Models were used in their default settings and maps were scaled back to the original image size for analysis.For the spectral model, the map was computed on a 10 th of the image size, in accordance with the typical settings of the other models.In line with the typical procedure when using salience maps, the actual stimuli were put into the models (i.e., the pixel values used for presentation), such that the values were not linear in luminance, but scaled with the screen's gamma.This may affect the precise functional form of the salience values' dependence on "contrast", but a monotonic relation will stay monotonic.As by the Harel et al. implementation, saliency maps and GBVS maps were scaled to a dynamic range from 0 to 1, while spectral salience was scaled to the maximum across all maps for display.We compared the salience of the alert-task square SA to the primary-task square SP by computing the mean across the 16x16 pixels of the map corresponding to the respective square.Using these values, we computed the global salience index (GSI S5 ), which has been suggested earlier S6 to compare targets with distractors, as (SA-SP)/(SA+SP).This value is bound between -1 and 1, where 1 means maximal salience at the alert task square, -1 maximal salience at the primary task square.For all models, the alert-task square and the primary-task square are more salient than the background (Fig. S1).Importantly, with increasing contrast the alert-task square becomes relatively more salient compared to the primary-task square, and the GSI increases monotonically with contrast (except for the lowest levels of the spectral salience measure).This shows that the intuitive definition of increasing salience by increasing contrast is compatible with standard models of low-level salience.

Supplementary Material, Part 2: Tables for Follow-up Analyses
For variables for which a main effect of contrast or sound pressure level was observed at a 5% significance level in Experiment 1, pairwise follow-up tests were conducted, which are reported in Table S1.
Consistent across experiments, we observe more alert-task intrusions for lower salience (visual or auditory) -that is, more alerts are missed if salience is low.However, once the alert is successfully detected1 , there is little to no effect of salience on the correctness of the response to the alert.

Alert task
To estimate how the performance in alert trials develops within a block and over the course of the experiment, we aggregated all alert trials across all experiments irrespective of condition.We consider how the probability to execute an alert trial completely correctly (i.e., no intrusions, no fixation error, responses to alert task and primary task correct; 100% minus "any error" of Table 1) varies over blocks and over trials within a block.The fraction of correct alert trials showed a dependence on block number (F(9,702) = 16.10,p < .001)with an overall performance increase from early to late blocks (Fig. S4a).
Alert reaction time, in turn, depended on block number (F(9,702) = 3.83, p < .001)with a near monotonic speed-up over the course of the experiment (Fig. S4b).Hence, while the speed-up follows the same trend as the primary tasks, there is no speed-accuracy trade-off over the course of the experiment.
There is a decline in correctly executed alert trials over the course of a block, although the first alert trial in a block is considerably more error-prone than all the remaining alert trials (Fig. S4c).For the alert reaction time we observe a similar pattern as for the primary task: the first 2 to 3 alert trials in a block are substantially slower than the following ones, while there is a slight slowing over the course of the block (Fig. S4d).Note that -as in the main analyses -only completely correct alert trials are included in the analysis.The qualitatively observed patters are confirmed by computing correlations between trial number and the performance measure in each individual (excluding the first alert trial in each case).
Despite considerable variability across individuals (Fig. S4e), the mean Spearman correlation (-.080) is shifted slightly to the negative on average with the mean across individuals significantly smaller than 0 (t(78) = 4.13, p < .001).Similarly, reaction times tended to increase over the course of a block with a mean Pearson correlation of .09across individuals (Fig. S4f; t(78) = 3.69, p < .001).

Table S1C :
: Experiment 1, p-values of all pair-wise follow-up tests for variables for which a significant main effect of contrast or sound pressure level was observed.Red p-values indicate significance at a 5% level after Bonferroni-Holm correction for the 28 tests conducted in the respective table; blue values indicate uncorrected significance.RT alert, auditory

Table S2 :
Experiment 2, p-values of all pair-wise follow-up tests for variables for which a significant main effect of duration was observed.Red p-values indicate significance at a 5% level after Bonferroni-Holm correction for the 28 tests conducted in the respective table; blue values indicate uncorrected significance.