The year 2023 marks the 50th anniversary of the publication of the first issue of Memory & Cognition. The editorial on the first page of that issue stressed that the range of topics was to be broad, excluding only manuscripts devoted to sensory processes and psychophysics, which were more appropriate for Perception & Psychophysics (Briggs & Schulz, 1973). The first article in the inaugural issue was “On the Selection of Signals,” by Michael I. Posner, Raymond Klein, Jeffrey Summers, and Stephen Buggie. The first topic addressed in the experiments reported in that article was alertness, of which, Posner et al. (1973) concluded,

Alertness does not affect the buildup of information within the memory system but only the rate at which a later system responds to that information. Thus, in standard reaction-time tasks, increased alertness produces a reduction in reaction time but no decrease in errors. (p. 2)

The research reported in that article and others from Posner’s research group played a major role in getting cognitive psychologists to focus on alertness and other aspects of attention. The experiments reported in this article are grounded in that research.

Alertness (or temporal preparation), as well as orienting and executive function, are referred to as three basic components of a human’s attention system (Petersen & Posner, 2012). In a simple- or choice-reaction task, presenting a neutral warning signal prior to an imperative stimulus can reduce the reaction time (RT). This effect is called phasic alertness in some studies (e.g., Schneider, 2018a) and temporal preparation in others (e.g., Los et al., 2014). One critical factor in this line of research is the temporal interval between termination of the warning signal and onset of the target stimulus (foreperiod). The foreperiod effect is defined as how RT and error percentage (EP) are influenced by the foreperiod duration.

Other than duration, the contingency (or predictability) of foreperiod can also be manipulated to produce different levels of temporal uncertainty. In the constant-foreperiod paradigm, the same foreperiod is used in all the trials within a trial block, rendering the duration predictable. In the variable-foreperiod paradigm, different foreperiods are randomly mixed within a block, which makes the duration of the next foreperiod unpredictable. In the latter paradigm, the effect of the current foreperiod is modulated by the foreperiod of the preceding trial (called the sequential foreperiod effect; Los et al., 2001; Steinborn et al., 2008; Vallesi & Shallice, 2007). In contrast, constant-foreperiod studies mainly focus on how the foreperiod effect is affected by different foreperiod durations (McCormick et al., 2019; Müller-Gethmann et al., 2003; Posner et al., 1973) and how the foreperiod effect interacts with other factors (Los & Schut, 2008; Schneider, 2018a, 2018b, 2019; Seibold, 2018; Weinbach & Henik, 2011, 2012).

When plotting RT as a function of the constant foreperiod, as foreperiod increases RT decreases, reaching its shortest point around 250 ms, and then increases as the foreperiod gets longer (e.g., Müller-Gethmann et al., 2003; Niemi & Näätänen, 1981). A critical issue regarding the effect of short foreperiods is whether the change in RT reflects more efficient information processing or adjustment of response criteria (speed–accuracy trade-off). Previous research on this topic shows divergent findings. Some studies found that the decrease in RT was accompanied by an increase in EP (Bertelson, 1967; McCormick et al., 2019; Posner et al., 1973), indicative of a change in speed–accuracy criterion. In other studies, this change in EP was not significant, leading to the conclusion that the warning signal could improve information-processing efficiency (Bertelson & Tisseyre, 1968; Los & Schut, 2008; Müller-Gethmann et al., 2003).

Han and Proctor (2022a) argued that providing trial-level RT feedback could contribute to finding a speed–accuracy trade-off pattern for the constant-foreperiod effect. They conducted two experiments with the same spatial two-choice reaction task as in Posner et al. (1973). Participants responded to the left-right locations of a capital letter “X” with spatially compatible or incompatible left-right key presses. There were five different foreperiod conditions in which a warning tone was presented 50, 100, 200, 400, or 800 ms before each imperative stimulus. A no-alert condition was also included for demonstrating whether the warning signal caused a speed–accuracy trade-off. In one experiment, RT feedback was provided after each trial, as in Posner et al. (1973), whereas the other experiment did not have this feedback. The results showed not only faster overall responses but also evidence supporting the existence of speed–accuracy trade-off with RT feedback but not without it.

In Han and Proctor’s (2022a) Experiment 2 (with RT feedback), the 50-ms foreperiod led to shorter RT and higher EP compared with the no-alert condition. In contrast, faster responses and no worse accuracy was found at the 200-ms foreperiod, as compared with the 50-ms foreperiod or the no-alert condition. Han and Proctor regarded these findings as evidence supporting a dynamic warning signal effect at short foreperiods. An extremely short foreperiod (e.g., 50 ms) could only cause a shift of response criteria, but longer ones (e.g., 200 ms) allow improved efficiency of information processing, implying different underlying mechanisms and distinct timeframes for their activation.

Another finding in Han and Proctor’s study (2022a) is related to the stimulus–response (S–R) compatibility effect, which refers to task performance being better when stimuli are spatially compatible with the responses than when they are not (see Proctor & Vu, 2006, for a review). This effect reflects that information builds up faster for the compatible S–R mapping compared with the incompatible one (Sanders, 1977; Yamaguchi & Proctor, 2012). Han and Proctor found an interaction between S–R mapping and foreperiod duration, which indicated a larger mapping effect at the 50-ms foreperiod than at the other foreperiod conditions. This finding was absent in most previous research (e.g., Frowein & Sanders, 1978; Los & Schut, 2008; McCormick et al., 2019; Posner et al., 1973). Han and Proctor argued that this interaction could have been caused by a longer activation needed by the incompatible mapping (p. 761), but they took a conservative attitude towards the reliability of this pattern.

The current study had three purposes. The first was to confirm whether the findings of Han and Proctor (2022a) could be replicated in a scenario with more temporal uncertainty. In that study, the speed–accuracy trade-off at the 50-ms foreperiod seems to have been driven by a relatively automatic mechanism, as argued by Posner et al. (1973) and McCormick et al. (2019). That mechanism is like the exogenous type of temporal attention, which should not be affected by the contingency of foreperiod (Lawrence & Klein, 2013Footnote 1). In contrast, at the 200-ms foreperiod, response speed became even faster than at the 50-ms foreperiod, but with a decrease in EP. That this effect emerged later than the one caused by the 50-ms foreperiod points to a more controlled mechanism underlying the 200-ms foreperiod effect, which prioritizes task-relevant rather than automatic pathways.Footnote 2 Thus, it is more likely that this effect relies on the contingency of foreperiod. In summary, the patterns found in Han and Proctor’s (2022a) study indicate the two effects’ different sensitivity to the manipulation of temporal uncertainty.

To decrease the predictability of the target stimulus’s onset, we used a variable-foreperiod paradigm. Although when longer foreperiods are used, the foreperiod effect in the variable foreperiod paradigm has a different direction than that in the constant-foreperiod paradigm, in shorter foreperiod scenarios, the foreperiod effect does not differ and the influence of sequential modulations is limited (Bertelson & Tisseyre, 1968; Han & Proctor, 2022b; Lawrence & Klein, 2013). For example, in Han and Proctor (2022b), the sequential effect at a 50-ms foreperiod was larger when that foreperiod was paired with 400 ms (less similar) than when it was paired with 200 ms (more similar). Also, the main effect of foreperiod sequence was significant with the mixture of 50 and 400 ms but not with that of 50 and 200 ms. These results imply that mixing short foreperiods with long foreperiods (>200 ms) increases the probability of sequential foreperiod modulation. To avoid the influence of long foreperiods and focus on the foreperiods of interest, we included only foreperiods of 50, 100, and 200 ms in the present study.

The second aim of the study was to investigate the reliability of the interaction between S–R mapping and foreperiod in a variable-foreperiod context. Han and Proctor (2022a) proposed that this interaction is related to the different speeds in processing spatially compatible and incompatible information. If this explanation were true, then a similar interaction should be expected even with increased temporal uncertainty, if the same foreperiods are used.

The insensitivity to whether foreperiods are blocked or mixed has been observed in literatures that focused on how the effect of phasic alertness interacts with cognitive control. For example, in the constant foreperiod paradigm, the presence of a warning signal enlarges the size of flanker effect at the 400-ms foreperiod (e.g., Schneider, 2018a, 2018b). Weinbach and Henik (2013) used a variable foreperiod and found a similar effect appearing at the 500-ms foreperiod. Because the response congruency effect in flanker tasks and the S-R mapping in two-choice spatial reaction tasks both involve the processing stage of response selection, additive factors logic (Sternberg, 1998) leads to prediction of an interaction between mapping and foreperiod in the variable foreperiod paradigm.

Lastly, our study intended to confirm the role that RT feedback plays in replicating the foreperiod-related patterns. The only difference between the two experiments in Han and Proctor’s (2022a) study was that their Experiment 2 provided trial-level RT feedback, whereas their Experiment 1 did not. Based on the results of those experiments, the presence of RT feedback speeded reactions and caused more errors, which seems to align with Hines’s (1979) point that RT feedback causes an adjustment in response criteria. At the same time, the RT feedback in Han and Proctor’s study enlarged several effects in their Experiment 2, including the mapping and foreperiod effects on EP and the interaction between mapping and foreperiod on RT. This pattern indicates that RT feedback not only adjusts people’s response criteria to a more liberal level but also can increase the sensitivity of RT and EP data to some experimental manipulations. In the current study, Experiments 1 and 2 provided the trial-level RT feedback, whereas Experiment 3 did not, giving us another opportunity to investigate the effect of RT feedback.

Experiment 1

Experiment 1 is a variable-foreperiod variant of Han and Proctor’s (2022a) second experiment. The 400-ms and 800-ms foreperiods were omitted from the design to avoid sequential modulation by long foreperiods and to focus more on the foreperiods of interest. The three foreperiods (50, 100, and 200 ms) were randomly mixed within “alert” blocks—that is, those with warning signals. There was also one no-alert block without warning signals in each mapping condition. The planned analyses were the comparison between the alert (averaged across foreperiods) and no alert conditions and the comparison among different foreperiods within the alert conditions.

Method

Participants

A total of 36 students (15 male, 24 female, age range: 18–22 years, mean = 18.58 years, SD = 1.0 years) participated. This sample size is identical to that in Han and Proctor (2022a), which was determined based on a power analysis of Los and Schut (2008) for a similar experiment showing that a sample size of 24 would provide a power of 0.9 to detect a mapping × foreperiod interaction. All participants in this experiment and the remaining ones (a) were enrolled in an introductory psychology course at Purdue University and received research credits, (b) reported having normal or corrected-to-normal vision and audition, and (c) were naïve to the study’s purpose. All three experiments were conducted in accord with a protocol approved by the Purdue University Institutional Review Board and the ethical principles of the American Psychological Association, and all participants signed an informed consent form prior to participating.

Apparatus and stimuli

Stimulus presentation and response recording were achieved by means of E-Prime software (Version 2.0, Psychology Software Tools, Inc.) installed on a PC workstation. Participants were seated in front of a 76-cm-high table on which an E-Prime response box with a row of five response buttons was placed. Instructions, visual imperative stimulus, and response feedback were presented on a 17-in. LCD monitor in front of the participant, with an unconstrained viewing distance of approximately 63 cm in a dimly lit room. The response box was centrally aligned with the display, and participants responded with their left and right index fingers on the leftmost and rightmost buttons of the box.

The background color of the monitor was black during the whole experiment with instructions, feedback, and stimuli displayed in white. The imperative stimulus was a capital letter “X” that appeared at the left or right side of the display. The size of the stimulus was 0.5° × 0.5°. The possible positions of “X” were both 10.7° in visual angle from the center of the screen. The warning signal was an 80-dBA tone of 500 Hz transmitted through a pair of SONY headphones.

Procedure

The design was a factorial combination of foreperiod and mapping. The four foreperiod conditions were: no alert block and 50-, 100-, and 200-ms foreperiod in the alert block. Each participant performed four compatible and four incompatible trial blocks, each set containing one no-alert block and three identical alert blocks. Each block contained 30 trials, 15 for each stimulus position. The experiment started with all the compatible blocks or all the incompatible blocks, and participants were allowed to take a 5-min rest between the two mapping conditions, although most participants chose to continue without a break. The sequence of mapping conditions was counterbalanced, and the order of foreperiod conditions was randomized. The experiment took about 40 min to complete.

Each trial began with an intertrial interval that randomly varied between 2 s and 5 s, after which the warning tone was presented for 50 ms. In the no-alert condition, the warning tone was replaced by a 50-ms blank slide without any auditory or visual stimuli. In alert blocks, offset of the warning signal was followed by a variable foreperiod. Three equally probable foreperiods—50, 100, and 200 ms—were randomly intermixed. At the foreperiod’s expiration, the imperative stimulus “X” was presented equally likely at the left or right location on the display. With the spatially compatible mapping, participants pressed the right key when “X” appeared at the right location, and the left key when it appeared at the left location. The incompatible mapping was a reversed version of the compatible mapping. The imperative stimulus stayed on the display until a response was made by pressing the left or right key. The response was then followed by the 1,500-ms feedback slide, which provided both the RT and accuracy information of that trial. Offset of the feedback slide initiated the next trial.

Before the experiment, participants were instructed to maintain their index fingers on the corresponding keys and to use only them to respond. Mapping information was provided at the beginning of the whole experiment, before the first trial of each block, and at the end of the first four trial blocks. All participants were told to respond as fast as possible. The experimenter stayed in the room with the participant for all the trials.

Results

Median RTs of correct responses and EPs were computed for each participant at each foreperiod for both compatible and incompatible mappings, which was the same method as in Han and Proctor’s (2022a) study. To conduct the comparison between the alert and no-alert conditions, we calculated the mean RTs and EPs of the alert conditions with either mapping for each participant. Figure 1 (left) shows RTs of the correct responses as a function of foreperiod and mapping. The numerical values for this and the other experiments are shown in Appendix Tables 1 and 2. A repeated-measures analysis of variance (ANOVA) tested the significance of both the alertness (no alert vs. alert) and mapping (compatible vs. incompatible) effects, and their interaction, at an α level of .05. In this and all the other experiments, the reported values for F and p correspond to the Greenhouse–Geisser corrections for violations of the sphericity of the variance-covariance matrix (although the reported degrees of freedom have not been adjusted). The ANOVA revealed that responses were 54 ms faster with the compatible mapping than with the incompatible mapping, F(1, 35) = 73.00, p < .001, \({\upeta}_p^2\) = .68, and 55 ms faster in the alert condition (with warning signal) than in the no-alert condition, F(1, 35) = 224.80, p < .001, \({\upeta}_p^2\) = .87. No interaction was found between mapping and alertness, F(1, 35) = .03, p = .872. The ANOVA for comparing different foreperiods within the alert conditions showed the compatibility effect for RT, F(1, 35) = 88.30, p < .001, \({\upeta}_p^2\) = .72. More critically, RT decreased as the foreperiod increased, F(2, 70) = 54.86, p < .001, \({\upeta}_p^2\) = .61, and, again, there was no interaction between mapping and foreperiod, F(2, 70) = 1.42, p = .249.

Fig. 1
figure 1

Experiment 1, with trial-level feedback: RT as a function of foreperiod (left); error percentage as a function of foreperiod (right). Error bars plot the adjusted standard errors for within-subject factors using the method described in Cousineau et al. (2021)

Full size image

Figure 1 (right) shows EP as a function of foreperiod and mapping. The ANOVA that focused on comparing the no-alert condition with the alert conditions revealed that responses were 1.8% more accurate with the compatible mapping, F(1, 35) = 12.43, p < .001, \({\upeta}_p^2\) = .26. Also, the participants made more errors with the presence of the warning signal (3.5%) compared with the no alert condition (2.2%), F(1, 35) = 4.26, p = .046, \({\upeta}_p^2\) = .11. No interaction was found between mapping and alertness, F(1, 35) = 2.10, p = .156. The ANOVA for comparing different foreperiods within the alert conditions showed lower EPs with the compatible mapping (2.8%) than with the incompatible mapping (4.1%), F(1,35) = 4.36, p = .044, \({\upeta}_p^2\) = .11. Also, EP increased as the foreperiod increased, F(2, 70) = 7.53, p = .003, \({\upeta}_p^2\) = .18. There was no interaction between mapping and foreperiod, F(2, 70) =.44, p = .622.

Discussion

We observed several informative results from Experiment 1. First, when comparing the no-alert condition with the alert condition (averaged across foreperiods), the results indicated that the presence of the warning signal speeded reactions but also caused more errors. This pattern implies that the warning signal led to a more liberal response criterion. As a further confirmation of this general pattern, the comparison between foreperiod conditions revealed a consistent speed–accuracy trade-off. As the foreperiod increased (from 50 ms to 200 ms), RT decreased, and EP increased. In Han and Proctor (2022a), the 200-ms foreperiod showed a distinct effect from that of the 50-ms foreperiod. The results in the current experiment imply that the contingency of foreperiod is necessary for the 200-ms foreperiod to improve the efficiency of information processing.

The lack of significant interaction between mapping and foreperiod indicates that there was no reliable evidence of a larger mapping effect at any foreperiod, even though the difference between the RTs of the two mappings was numerically larger at the 100-ms foreperiod. In short, it is either that the contingency is critical for observing the interaction between mapping and foreperiod, or that the interaction (larger mapping effect at 50-ms foreperiod compared with other foreperiod conditions) observed in Han and Proctor (2022a) was small or unreliable.

Experiment 2

The purpose of conducting Experiment 2 was to replicate the findings of the alert blocks in Experiment 1 without inclusion of the no-alert blocks. Los et al. (2017) and Crowe and Kent (2019) revealed that the alerting states from one trial block can be transferred to a subsequent block even when the temporal context is changed. In Experiment 1, we randomly mixed alert and no-alert trial blocks, allowing the latter to possibly affect the former. To rule out this possible confound in evaluating the foreperiod effect within the alert conditions, in Experiment 2 we excluded the no-alert blocks so that all the blocks would share the same structure and alerting states. The planned analyses in Experiment 2 were the comparison among foreperiods and a between-experiment comparison with Experiment 1.

Method

Participants

Thirty-six students (13 male, 23 female, age range: 18–21 years, mean = 18.5 years, SD = 0.7 years) from the same participant pool took part, none of whom had participated in the prior experiment.

Apparatus, stimuli, and procedure

The apparatus, stimuli, and procedure were the same as those of Experiment 1, except for the following changes. First, the no-alert condition was omitted. Second, each mapping session contained six identical blocks with 30 trials in each.

Results

Median RTs of correct responses and EPs were computed for each participant at each foreperiod for both compatible and incompatible mappings. Figure 2 (left) shows RTs of the correct responses as a function of foreperiod and mapping. The repeated-measures ANOVA with foreperiod (50 , 100, 200 ms) and mapping (compatible vs. incompatible) as two within-subject factors showed that responses were 54 ms faster with the compatible mapping than with the incompatible mapping, F(1, 35) = 107.79, p < .001, \({\upeta}_p^2\) = .76. Also, RT decreased as the foreperiod increased, F(2, 70) = 108.94, p < .001, \({\upeta}_p^2\) = .76. Unexpectedly, we found a significant interaction between mapping and foreperiod, F(2, 70) = 6.20, p = .004, \({\upeta}_p^2\) = .15. Orthogonal contrasts on the interaction showed a quadratic component, F(1, 35) = 11.19, p = .002. The mapping effect at the 100-ms foreperiod (59 ms) was larger than that at the 50 ms and 200 ms foreperiods (52 ms for each). The linear component (i.e., comparing the mapping effects between the 50-ms and 200-ms foreperiods) was nonsignificant, p = .896.

Fig. 2
figure 2

Experiment 2, with trial-level feedback: RT as a function of foreperiod (left); error percentage as a function of foreperiod (right). Error bars plot the adjusted standard errors for within-subject factors using the method described in Cousineau et al. (2021)

Full size image

Figure 2 (right) shows EP as a function of foreperiod and mapping. The repeated-measures ANOVA with foreperiod (50, 100, 200 ms) and mapping (compatible vs. incompatible) as two within-subject factors showed lower EPs with the compatible mapping (2.3%) than with the incompatible mapping (4.4%), F(1,35) = 24.63, p < .001, \({\upeta}_p^2\) = .41. Also, EP increased as the foreperiod increased, F(2, 70) = 14.20, p < .001, \({\upeta}_p^2\) = .29. There was no interaction between mapping and foreperiod, F(2, 70) = .02, p = .97.

Between-experiment comparison

A repeated-measures ANOVA on RT included experiment (1 vs. 2) as a between-subjects factor and mapping and foreperiod as within-subject factors. The no-alert condition in Experiment 1 was excluded from this analysis. We report here the overall interaction between mapping and foreperiod and the terms involving the experiment factor. The ANOVA showed a significant Mapping × Foreperiod interaction, F(2, 140) = 5.21, p = .008, \({\upeta}_{\textbf{p}}^{\textbf{2}}\) = .07. Consistent with the results of Experiment 2, the mapping effect was 6.8 ms larger at the 100-ms foreperiod than in other foreperiod conditions. None of the experiment-related effects was significant. Like the RT results, the ANOVA on EP showed no significant experiment-related effects.

Discussion

Most of the findings of Experiment 2 were identical to those of Experiment 1. From 50-ms to 200-ms foreperiod, speeded reactions were accompanied by higher chances of error responses. Also, shown in the between-experiment comparison, neither the main effect of experiment nor the experiment-related interactions were significant, which means that excluding the no-alert conditions did not have much impact on the performance in the alert conditions.

The only difference between the results of the two experiments is the detection of a Mapping × Foreperiod interaction in Experiment 2. The mapping effect was larger at the 100-ms foreperiod, which is consistent with the numerical pattern in Experiment 1. The nonsignificant three-way interaction in the between-experiment comparison indicates that the significant interaction found in Experiment 2 cannot be simply regarded as a fortuitous result, as it revealed some shared characteristics in the data of both experiments. However, because Experiment 1 did not show a significant Mapping × Foreperiod interaction, the size of this effect likely is smaller than what was measured in Experiment 2. The size of the interaction revealed in the between-experiment comparison should be regarded as a better representation. In short, Experiment 2 provided evidence supporting Han and Proctor’s (2022a) argument that the warning signal does lead to different activation processing speeds between spatially compatible and incompatible information.

Experiment 3

The purpose of conducting Experiment 3 was to confirm the role that RT feedback plays in replicating the foreperiod-related patterns. In Han and Proctor (2022a), providing RT feedback led to faster responses and higher chances of making errors, consistent with adjusting response criteria. RT feedback also enlarged the mapping and foreperiod effects on EP and the Mapping × Foreperiod interaction on RT. Therefore, it is necessary to confirm whether it would affect the replicability of foreperiod-related effects in a variable foreperiod scenario. The planned analyses in Experiment 3 were the comparison among different foreperiods and a between-experiment comparison with Experiment 2.

Method

Participants

Thirty-six students (eight male, 28 female, age range: 18–20 years, mean = 18.5 years, SD = 0.7 years) from the same participant pool participated, none of whom had participated in the prior experiments.

Apparatus, stimuli, and procedure

The apparatus, stimuli and procedure of Experiment 3 were identical to those of Experiment 2, except that RT feedback was omitted. In other words, the participants in Experiment 3 were only told about whether the response was correct but not about the RT in each trial.

Results

Median RTs of correct responses and EPs were computed for each participant at each foreperiod for both compatible and incompatible mappings. Figure 3 (left) shows RTs of the correct responses as a function of foreperiod and mapping. The repeated-measures ANOVA with foreperiod (50, 100, 200 ms) and mapping (compatible vs. incompatible) as two within-subject factors showed that responses were 57-ms faster with the compatible mapping, F(1, 35) = 61.49, p < .001, \({\upeta}_p^2\) = .64. Also, RT decreased as the foreperiod increased, F(2, 70) = 97.83, p < .001, \({\upeta}_p^2\) = .74. The interaction between mapping and foreperiod was not significant, F(2, 70) = .47, p = .619.

Fig. 3
figure 3

Experiment 3, without RT feedback: RT as a function of foreperiod (left); error percentage as a function of foreperiod (right). Error bars plot the adjusted standard errors for within-subject factors using the method described in Cousineau et al. (2021)

Full size image

Figure 3 (right) shows EP as a function of foreperiod and mapping. The repeated-measures ANOVA with foreperiod (50, 100, 200 ms) and mapping (compatible vs. incompatible) as two within-subject factors showed lower EPs with the compatible mapping (0.3%) than with the incompatible mapping (0.9%), F(1, 35) = 13.89, p < .001, \({\upeta}_p^2\) = .28. Unlike Experiments 1 and 2, foreperiod duration did not significantly affect EP, F(2, 70) = .42, p = .654. More surprisingly, there was a significant interaction between mapping and foreperiod, F(2, 70) = 3.93, p = .025, \({\upeta}_p^2\) = .10. Separate analyses indicated that the mapping effect was significant at the 200-ms foreperiod (p < .001) and larger than that at the 50-ms foreperiod (p = .843). However, the foreperiod effect was not significant in either the compatible (p = .094) or incompatible (p = .136) condition alone.

Between-experiment comparison

A repeated-measures ANOVA on RT included experiment (2 vs. 3) as a between-subjects factor and mapping and foreperiod as within-subject factors. We report only the terms involving the experiment factor. The ANOVA showed a main effect of experiment, F(1, 70) = 28.60, p < .001, \({\upeta}_{\textbf{p}}^{\textbf{2}}\) = .29. Responses were 73 ms faster in Experiment 2 than in Experiment 3. The two-way within–between Foreperiod × Experiment interaction was significant, F(1, 70) = 6.82, p = .005, \({\upeta}_{\textbf{p}}^{\textbf{2}}\) = .09, indicating a larger foreperiod effect when RT feedback was absent. In contrast, the presence of RT feedback did not change the mapping effect, F(1, 70) = .11, p = .744. The three-way interaction of those variables was significant, F(2, 140) = 3.21, p = .044, \({\upeta}_{\textbf{p}}^{\textbf{2}}\) = .04. The Mapping × Foreperiod interaction was larger when RT feedback was present.

The ANOVA on EP showed a main effect of experiment, F(1, 70) = 29.78, p < .001, \({\upeta}_{\textbf{p}}^{\textbf{2}}\) = .30. The participants in Experiment 2 made 2.7 percentage point more errors, overall, than those in Experiment 3. The two-way interactions of mapping and experiment, F(1, 70) = 12.10, p < .001, \({\upeta}_{\textbf{p}}^{\textbf{2}}\) = .15 , and foreperiod and experiment, F(2, 140) = 10.88, p < .001, \({\upeta}_{\textbf{p}}^{\textbf{2}}\) = .13, were both significant, implying that the mapping and foreperiod effects on EP were more pronounced with the presence of RT feedback. The three-way interaction was not significant, F(2, 140) = .85, p = .426.

Discussion

In Han and Proctor (2022a), neither the foreperiod effect on EP nor the Mapping × Foreperiod interaction on RT was observed when RT feedback was not provided. Their between-experiment comparison showed that the presence of RT feedback not only speeded responses and increased errors, but also enlarged the effects of both mapping and foreperiod on EP. The current experiment revealed a similar pattern of the RT feedback effect. In Experiment 3, the absence of RT feedback led to longer RTs and lower EPs, which reinforces the idea that the setting of the response criterion is highly sensitive to the presence of RT feedback. Both the foreperiod effect on EP and the Mapping × Foreperiod interaction on RT were absent, and the comparison between Experiments 2 and 3 showed that RT feedback led to larger mapping and foreperiod effects on EP in Experiment 2. Regarding the RT data, the absence of RT feedback decreased the Mapping × Foreperiod interaction. In other words, RT feedback caused a similar effect in the current study as in Han and Proctor’s (2022a).

There were two unexpected findings in this experiment. First, the foreperiod effect on EP seems to have been modulated by mapping. Given the low overall EP in Experiment 3 and the nonsignificant foreperiod effect on both mapping conditions, this interaction could reflect a floor effect on the EP data, and its reliability is questionable. Combined with the nonsignificant three-way interaction in the between-experiment comparison on EP, the importance of this Mapping × Foreperiod interaction on EP is limited and does not deserve further discussion.

Second, based on the results of the between-experiment comparison, the absence of RT feedback increased the size of the foreperiod effect on RT. Han and Proctor (2022a) did not find this interaction with the constant foreperiod paradigm. Implications of this and the other effects of RT feedback are discussed further in the next section.

General discussion

Lawrence and Klein (2013) dissociated the two different modes of temporal attention, which refers to the shift of attention caused by a neutral warning signal. Temporal attention can be oriented exogenously by an abrupt change in the environment which only has a random temporal relation with onset of the imperative stimulus, or it can be oriented endogenously by an inconspicuous but meaningful warning with a constant foreperiod. From their results, Lawrence and Klein concluded that the exogenous mode of temporal attention speeded responses with no cost on accuracy whereas the endogenous mode enhanced both speed and accuracy. In Posner et al. (1973) and McCormick et al. (2019), a conspicuous neutral warning with a constant foreperiod led to shorter RT but higher EP compared with the no-warning condition. Based on the endogenous-exogenous modes of temporal attention, the warning signal in those studies should have caused a combination of the two temporal attention modes. Therefore, as argued by McCormick et al., the data indicated that combining the modes of temporal attention led to a speed–accuracy trade-off.

In Lawrence and Klein’s (2013) results, however, this speed–accuracy trade-off was not reflected by the effect of foreperiod, as neither type of temporal attention had an effect of causing more errors with the increase of foreperiod. Instead, it was reflected by the effect of increasing the intensity of the warning signal. This intensity effect increased speed and decreased accuracy with a maximum magnitude at around 100-ms stimulus-onset asynchrony (SOA). This time-course characteristic was also found by McCormick et al. (2019), who indicated that the difference between endogenous phasic alertness and the combination of the two modes was largest at 100-ms SOA and diminished at 250-ms SOA. Interestingly, what was overlooked by Posner et al. and McCormick et al. was that in both studies EP reached its maximum level at around 100-ms SOA and started to decrease at 250-ms SOA. These results seem to imply that the dominant component of phasic alertness changed from 100 to 250 ms.

With the purpose of clarifying whether a conspicuous warning signal with a constant foreperiod would only lead to one (shift of response criteria) or two types of effects, Han and Proctor (2022a) replicated Posner et al.’s (1973) study and conducted separate comparisons between different foreperiod conditions. They found that the 50-ms foreperiod (corresponding to a 100-ms SOA) condition did lead to a speed–accuracy trade-off compared with the no-alert condition. However, at the 200-ms (corresponding to a 250-ms SOA) foreperiod, both speed and accuracy improved compared with the 50-ms foreperiod. The importance of this finding is that it implies the existence of two distinct underlying processes in the constant-foreperiod warning signal effect, which is relatively more consistent with what Lawrence and Klein (2013) argued. What remained unclear was which attention mode, exogenous or endogenous, was responsible for the distinct patterns observed at 50-ms and 200-ms foreperiods.

In the current study, different foreperiods were randomly mixed such that the participants were unable to know in advance how long the next foreperiod would be. This situation leads to a decreased temporal contingency and therefore is more like that for exogenous temporal attention described in Lawrence and Klein (2013). The results of Experiments 1 and 2 showed that without the predictability of foreperiod, a longer foreperiod (up to 200 ms) led to a more liberal response criterion. In other words, response speed was enhanced with the cost on accuracy. A similar pattern was observed only at the 50-ms foreperiod in Han and Proctor (2022a), which implies that this effect may have been overridden by another process that improved information processing in terms of both speed and accuracy at the 200-ms foreperiod. This process, however, was absent in the current study, which decreased the predictability of foreperiod relative to Han and Proctor (2022a). Therefore, it is reasonable to infer that this process, which was detected in Han and Proctor’s 200-ms foreperiod condition and also present (but not noted) in Posner et al.’s (1973) and McCormick et al.’s (2019) studies, relies on the predictability of foreperiod, which is more consistent with the characteristics of the endogenous mode of temporal attention. In contrast, the foreperiod effect in the current study, which resembled the speed–accuracy trade-off caused by the 50-ms foreperiod effect in Han and Proctor’s study, implies a more exogenous mode of temporal attention.

Combining the results of the current study and those of Han and Proctor (2022a), we provided evidence for two different types of temporal attention (or phasic alertness, as argued by Schneider, 2018a, 2018b). Our results indicate that the stimulus-driven mode of temporal attention gets activated quickly but speeds response with a cost of accuracy, whereas the expectancy-driven mode of temporal attention is activated slower but can override the effect of the exogenous mode with improvement of both speed and accuracy. Although this distinction differs from Lawrence and Klein’s (2013) regarding the effect of the exogenous temporal attention, the difference in activation speed indicated by Han and Proctor’s results and the present ones is consistent with the results of Lawrence and Klein’s study. Their results showed the maximum of the exogenous mode at the 100-ms foreperiod but the maximum of the endogenous mode at the 400-ms foreperiod (Lawrence & Klein, 2013, pp. 567–568).

One question that could be raised when comparing the current study to Lawrence and Klein’s (2013) is why the speed–accuracy trade-off was reflected by the main effect of foreperiod in the current study but not in Lawrence and Klein’s results. The two experiments do share some important characteristics, including the presentation of RT feedback and decreasing the predictability of the temporal relation (this was done to a greater extent by Lawrence & Klein, 2013). Also, it should be noted that a speed–accuracy trade-off was observed as the intensity effect in Lawrence and Klein’s contingent condition. The absence of the foreperiod effect on EP in the noncontingent condition in their data could have been due to the inclusion of longer foreperiods (>200 ms).

According to the multiple trace theory of temporal preparation (Los et al., 2014; Salet et al., 2022), a long foreperiod in the preceding trial will decrease response activation when the current foreperiod is short. The longer is the preceding foreperiod, the more response inhibition will be added to the current short foreperiod. Los (2013) revealed that the sequential modulation of a long foreperiod in the preceding trial is like a preceding no-go trial in a go/no-go task. In Han and Proctor (2022b), when 50-ms and 200-ms foreperiods were mixed with catch trials (presenting the warning signal without the target stimulus), the RTs at both foreperiods were increased by about 40 ms after a preceding catch trial. Also, comparing the results from their Experiments 1 and 3, the RT at the 400-ms foreperiod was longer when that foreperiod was paired with the 1,400-ms foreperiod (452 ms) than when it was paired with the 50-ms one (427 ms). In both Han and Proctor (2022a) and the current study, the foreperiod effect on EP was more robust in the RT-feedback condition, in which the overall response speed was faster. Therefore, it is possible that mixing short and long foreperiods in a trial block might have decreased the chance of detecting the foreperiod effect on EP at short foreperiods in Lawrence and Klein’s (2013) study. Alternatively, it is possible that the more reliable “trial” structure in the current experiments (the target stimulus always appeared shortly after the warning signal in each trial) led to this different foreperiod effect on EP compared with the less reliable warning-target temporal relation in Lawrence and Klein’s study. Testing these two possibilities is beyond the scope of our experiments but should be explored in future research.

Considering sequential modulation in the variable-foreperiod paradigm, one might also question whether the patterns in the current study could have been affected by the sequential foreperiod effect. Han and Proctor (2022b) did observe a sequential effect at the 50-ms foreperiod when it was mixed with the 200-ms foreperiod. However, it was a tiny effect (4 ms), and the sequential effect at the 200-ms foreperiod was not significant. A larger sequential effect (9 ms) at the 50-ms foreperiod was found when the other foreperiod was prolonged to 400 ms, which implies again the import of not including longer foreperiods if the purpose is to focus on the effect of the current foreperiod.

The current study also showed evidence for the existence of the Mapping × Foreperiod interaction in a variable-foreperiod scenario. Han and Proctor (2022a) indicated that the larger mapping effect at the 50-ms foreperiod might be related to faster processing speed for the spatially compatible information. Because we did not change the warning, the imperative stimuli, or the mappings, a similar pattern was expected in the current context, although the temporal uncertainty of the target stimulus’s onset was increased. The finding of a Mapping × Foreperiod interaction in a variable-foreperiod scenario supports Han and Proctor’s argument and is in line with other studies on the interaction between phasic alertness and cognitive control (e.g., Schneider 2018a, 2018b).

Fischer et al. (2010) found a larger Simon effect with the presence of the warning signal. They attributed this result to an enhanced automatic S–R pathway. This argument can be used to explain the Mapping × Foreperiod interaction in both the current study and Han and Proctor (2022a). When the warning signal was presented, it provided more enhancement to the spatially compatible S–R pathway. This difference in the activation speed caused a larger decrease in RT, which led to a larger mapping effect at the shorter foreperiods, which was 50 ms in Han and Proctor’s study and 100 ms in the current experiments. At the longer foreperiod (200 ms), the activation of the task-relevant and spatially incompatible pathway caught up and reached its optimum, leading to a less pronounced mapping effect. Our results support the argument in Schneider (2019) and Schneider (2020) that the effect of phasic alertness is closely related to the processing of spatial information.

Although finding the Mapping × Foreperiod interaction in a variable-foreperiod scenario supports its insensitivity to the change of temporal uncertainty, the interaction in the current study did occur at a later timepoint (100 ms) from onset of the warning signal compared with that in a constant-foreperiod scenario (50 ms). This small delay might indicate that increasing temporal uncertainty could have slowed the activation of the overall warning signal effect. However, the current study does not have enough evidence to support this argument.

Another difficult-to-answer question is why this interaction between mapping and foreperiod was only found in Han and Proctor’s (2022a) Experiment 2 and in Experiments 1 and 2 of the current study. The experimental design in Posner et al. (1973) and McCormick et al. (2019) involved having the participants switching between the spatially compatible and incompatible mappings, the purpose of which was to increase the overall EP. Han and Proctor argued that this consecutive switch could have led to a larger memory-based interference when the participants shifted from one mapping to the other, which could have influenced the measurement of the mapping effect, turning it into a combination of the regular mapping effect and the interference effect. The current study and Han and Proctor’s did not adopt this design with the purpose of excluding the potential modulation of mixing mappings (see Vu & Proctor, 2004; Yamaguchi & Proctor, 2006, for examples). However, the comparison between Han and Proctor’s (2022a) study and Posner et al.’s should not be regarded as direct evidence supporting this argument, although the overall mapping effect in the former is numerically larger than that in the latter.

The same question can be raised regarding the results of Los and Schut (2008), who did not require the participants to switch between different mappings either and did not find an interaction between mapping and foreperiod. From our viewpoint, the only critical difference in experimental design is that block-level RT feedback was provided in Los and Schut’s study, whereas in the current study and Han and Proctor’s (2022a) Experiment 2, RT feedback was provided after each response. The between-experiment comparisons in our prior and current studies showed consistently that the presence of the trial-level RT feedback enlarged the interaction between mapping and foreperiod. However, the reason behind it is unclear.

In the current study, RT feedback not only modulated the Mapping × Foreperiod interaction. Consistent with the findings in Han and Proctor (2022a), the presence of RT feedback increased the overall response speed and induced more errors. This result agrees with the conclusion of Hines (1979) that providing RT feedback let participants adopt a more liberal response criterion. Apart from this general effect, RT feedback in the current study also raised the sensitivity of the EP data to the effects of both mapping and foreperiod. We think this interaction between RT feedback and other factors could be a byproduct of the general effect on performance. If we assume that trial-level RT feedback only changes the participants’ response criteria but not the efficiency of information processing, then according to research on the speed–accuracy trade-off, using RT feedback will increase the sensitivity of EP data to changes in RT because in a speed–accuracy trade-off curve, accuracy does not vary much with the change in RT when RTs are very long (Heitz, 2014; Pachella, 1974).

An implication of our second experiment is that, as foreperiod increased, the response criterion was also adjusted to a more liberal level. Following this rationale, it is reasonable to infer that a more liberal baseline produced by providing RT feedback could have been the reason for the increased sensitivity of the EP data to the change in RT caused by the warning signal. If this is true, then the interaction between experiment and foreperiod on RT could be explained by assuming that adjusting the response criterion to a more liberal level becomes more difficult as the current response criterion becomes more liberal. Although we emphasized the importance of response speed in the oral instruction, each time participants made an erroneous response, they were reminded about the mistake and were not given RT feedback of that trial, which should have discouraged them from making more mistakes. This could have set a rough limit of the participants’ response criterion adjustment, beyond which making too many mistakes will become unavoidable. One way to breach this limit and make responses even faster is to have more efficient information processing, which was what happened in Han and Proctor’s (2022a) Experiment 2. This could explain why RT feedback did not interact with foreperiod in their between-experiment comparison. However, attributing the effect of RT feedback to the adjustment of response criteria also raised the question why RT feedback did not modulate the magnitude of mapping effect in our studies, given that a larger S–R compatibility effect is expected when a more conservative response criterion is adopted (e.g., Sanders, 1977). Therefore, stronger and more direct evidence is needed to support the above arguments related to the effect of RT feedback.

Finally, we stress that the arguments we made about the foreperiod effect are based on the results collected from a short-foreperiod scenario (≤200 ms). Although the effect of short foreperiods is a key part of the general warning signal effect, the conclusions should not be simply applied to other foreperiod scenarios because sequential modulation and other factors could play a more important role in them (Los et al., 2014). Also, the auditory warning signal in the present study was perceptually salient. Therefore, our findings related to its effect should not be simply generalized to a visual or less intense auditory warning signal. However, the warning signal in the current study was similar to the one used in Posner et al. (1973) and the exogenous (and combination of exogenous and endogenous modes) mode of temporal attention in Lawrence and Klein (2013), thus should have been appropriate for the current investigation. The results from the current study and Han and Proctor’s (2022a) experiments also suggest the value of applying RT feedback in precisely measuring and investigating the effect of a neutral warning signal, which should be taken into consideration in future research on similar topics.

Conclusion

The present study provided evidence that a neutral warning signal with unpredictable foreperiods shorter or equal to 200 ms makes people respond faster without improving information processing, so long as trial-by-trial RT feedback is provided (although this does not exclude the possibility that improving information processing and adjusting response criteria co-occur). This evidence agrees with that provided by Posner et al. (1973) in the inaugural issue of Memory & Cognition, but with more precise specification of the conditions under which this relation will occur. We also found that the warning signal enhances the activation of spatially compatible and incompatible S–R pathways differently at short foreperiods, regardless of temporal uncertainty. Consistent with the findings of Han and Proctor (2022a), the results verify the essential role that RT feedback plays in revealing the foreperiod effect on RT and EP. We argue that RT feedback is closely related to shifting the response criterion, which demands further investigation in future research.