In the ATM domain, SOA has been regarded by both ATCOs (Goeters et al., 2004; Kissing and Eißfeldt, 2014) and aircraft pilots (Goeters et al., 2004; Eißfeldt et al., 2011) as one of the most important cognitive abilities for operational competence. Performance-wise, experimental studies involving spatial ability assessment have shown that higher levels of SOA were associated with (i) higher situation awareness of downlinked messages among professional ATCOs (Yang and Zhang, 2009), (ii) improved management of ground level traffic among ATCO applicants (Grasshoff et al., 2015), and (iii) better aircraft handling performance during different flight phases (i.e., departure, en route, arrival) among university students and employees who can become ATCOs (Ackerman and Kanfer, 1993; Ackerman et al., 1995; Ackerman and Cianciolo, 2002).

Previous ATM studies that involved psychometric testing mainly involved spatial orientation tests (SOTs) that required: (i) judgments of lateral heading directions (i.e., differentiating between left and right turning directions when following a schematically drawn route) (Yang and Zhang, 2009), and (ii) perspective-taking focused on a schematically drawn 3D object (e.g., visualizing how a 3D cube-like object would appear from different viewpoints Ackerman and Kanfer, 1993; Ackerman et al., 1995; Ackerman and Cianciolo, 2002). Notably, the SOT currently administered by EUROCONTROL (“spot the side”) focuses on lateral directional judgments, requiring test takers to take the perspective of an on-screen avatar and pinpoint the location of a geometric object (left or right) that is held by one of his/her outstretched hands¹. This test is conceptually similar to the aforementioned test requiring lateral heading judgments (Yang and Zhang, 2009) and can be seen theoretically as an elementary test of egocentric PTA (Kozhevnikov and Hegarty, 2001).

These aforementioned SOTs showed that alternative types of SOTs involving a wider array of objects and presenting higher orders of task complexity were absent in the ATM domain. This could be explained by the fact it has taken almost two decades of intensive psychometric research, involving large samples of participants, to provide strong support for the reliability and validity of SOTs featuring multiple objects (or landmarks) (Kozhevnikov and Hegarty, 2001; Hegarty and Waller, 2004; Kozhevnikov et al., 2006; Lun et al., 2013; Zhong, 2013; Weisberg et al., 2014; Zhong and Kozhevnikov, 2016; Holmes et al., 2017; Hegarty et al., 2018; Gunalp et al., 2019; Muffato and Meneghetti, 2019; Friedman et al., 2020; Gunalp, 2020). In general, these SOTs require participants to execute pointing responses toward multiple objects (or landmarks) arranged in an array with reference to an initial heading direction (Kozhevnikov and Hegarty, 2001; Hegarty and Waller, 2004; Kozhevnikov et al., 2006; Lun et al., 2013; Zhong, 2013; Zhong and Kozhevnikov, 2016; Friedman et al., 2020). Some of these SOTs were programmed to provide sensitive recordings of pointing performance on computers and such computerized SOTs have been shown to correlate significantly with real-world landmark pointing performance that reflected stored environmental knowledge acquired from route learning (Lun et al., 2013; Zhong, 2013, 2016).

On the other hand, studies from the ATM domain showed that performance on a classical paper-based version of these computerized SOTs (Kozhevnikov and Hegarty, 2001) did not predict how successful ATCOs were in handling aircraft for safe landings (Jong, 2015). Likewise, performance on a paper-based SOT involving the judgments of left and right heading changes along a 2D route did not predict ATCOs' conflict detection performance (Yang and Zhang, 2009). These negative findings could be explained by the fact that aircraft handling (i.e., monitoring and sequencing aircraft for safe departures, hand-overs, and arrivals) relies more on other cognitive abilities like spatial attention and mental rotation than on SOA (Jong, 2015) and that the cognitive processes involved in making lateral directional judgments (Yang and Zhang, 2009) did not match well with higher-order attentional and spatial reasoning processes required for air traffic conflict detection (Rantanen and Nunes, 2005; Loft et al., 2009). Conceptually, these negative findings were crucial for supporting the notion (mentioned above in “Introduction”) that there needs to be a convergence or overlap of similar mental processes (perceptual and/or cognitive) between a spatial ability test and an ATC task in order for the former to predict performance on the latter.

In view of the aforementioned SOTs in the ATM psychometric literature that presented small numbers of objects requiring judgments of changes in orientation and that there are currently no multiobject-based SOT that can predict air traffic conflict detection performance, we aimed to assess conflict detection through the use of a computerized SOT that was developed recently by Friedman et al. (2020) and made freely accessible to researchers from all disciplines. This new SOT was designed with respect to the notion of SOA as a multiobject-directed and navigationally relevant construct that involves judging orientation or perspective changes (Kozhevnikov et al., 2006; Lun et al., 2013; Zhong, 2013; Zhong and Kozhevnikov, 2016; Hegarty et al., 2018). Compared with its paper-based predecessor (Kozhevnikov and Hegarty, 2001), this computerized SOT offers greater ease of stimuli presentation and higher precision in data recording (i.e., recording of pointing responses to the accuracy of two decimal places, see Methods, for more details).

To our knowledge, the current study is the first that applied this new SOT to air traffic conflict detection. Prior to this study, several studies had directed this new test successfully to investigations of individual differences in SOA and environmental learning (Gunalp, 2020; He and Hegarty, 2020; Simpson, 2020). In this study, we investigated the extent to which SOA, as conceptualized and measured in the recent spatial navigation literature, could predict conflict detection performance among a sample of university students and employees who are eligible for application as ATCOs. For a detailed examination of this predictive relationship, we introduced a mental workload component of air traffic density (Remington et al., 2000) to explore if this association would change in magnitude with varying numbers of aircraft in a simulated, controlled airspace (see Methods, for details). Like what Friedman et al. (2020) did, we also examined the convergent validity of the new SOT (Friedman et al., 2020) in relation to a classical paper-based SOT, the Money Road Map Test (Money et al., 1965), that requires quick judgments of lateral heading changes. By investigating all these relationships, we aimed to provide novel findings to argue for the relevance of this new SOT for predicting conflict detection performance in a simulated FRA environment among prospective ATCOs. By directing a SOT whose design has been well-conceptualized by spatial navigation researchers to ATM research, we also aimed to promote more communication and collaboration between researchers from the spatial navigation and ATM domains.

20 young adults (11 females, nine males), ranging from 19 to 31 years of age (M = 23.65, SD = 2.96), participated in the study². This participating group comprised university students (undergraduates and postgraduates) and employees. They were recruited through email invitations and advertisements posted on notice boards in the university campus. All participants possessed normal or corrected vision (20/40 vision or better) and were found to be right-handed. The majority of participants (n = 17) majored in the hard sciences (i.e., Aerospace Engineering, Electrical Engineering, Computer Science). The remaining participants (n = 3) majored in the social sciences (i.e., Business Administration, Psychology, Sociology). Age and educational qualifications of all participants met the entry requirements for ATCO application specified by the Civil Aviation Authority of Singapore (CAAS)³. Informed consent for study participation was approved by the Institutional Review Board (IRB) of Nanyang Technological University (NTU) [Reg. no.: 200604393R]. Signed consent was obtained from each participant prior to his/her participation. All tasks and procedures mentioned below were carried out in accordance with relevant ethical and institutional guidelines.

The SOT used in this study was a computerized test developed by Friedman et al. (2020). It was packaged as a Java applet made freely available for download and research use under the auspices of the international Creative Commons 4.0 (CC BY 4.0) license provided by the Open Science Framework (OSF), Center for Open Science, Charlottesville, VA, USA (https://osf.io/t58ab/).

A 15.6-inch LCD monitor with an aspect ratio of 16:9 (13.60 x 7.65 inches, equivalent to 34.51 x 19.42 cm) was used for presenting the SOT. The task was shown in full screen at a resolution of 1,920 x 1,080 pixels and a refresh rate of 60 Hz. Each participant sat 20 inches (50.8 cm) away from the center of the screen while performing the task. The task comprised three practice trials and 12 test trials. On each trial, participants viewed an array of seven inanimate objects whose locations were kept fixed across all trials. A drum was depicted at the center of the array surrounded by six other objects, rendering a spatial layout that appeared quasi-hexagonal (see Figure 2). To eliminate pointing biases in any direction, these inanimate objects were intentionally designed by Friedman et al. (2020) to have no inherent directionality (i.e., a salient initial heading direction).

The object array was shown on the left half of the computer screen, subtending a visual angle of 16.2, while a circle for making pointing responses was shown on the right half of the screen, subtending a visual angle of 11.2°. Following on-screen instructions located at the bottom of the object array, participants were told to imagine themselves standing at the location of an initial object, facing a second object, and then point to a third object based on that imagined perspective. On each trial, the imagined initial heading direction was shown on the adjacent response circle and each participant responded by positioning a straight line that rotated about the center of the response circle in a desired pointing direction. The straight line represented a pointing arrow and showed up when a participant made a left-button mouse click at any position within the response circle. To confirm one's decision on a pointing direction, a participant must press the “Enter” key on the keyboard, upon which data recording occurred. Participants' responses were recorded in the form of angular pointing errors to the accuracy of two decimal places. This error is unsigned and refers to the deviation in degrees from a point-to-target angle (i.e., target angle) that was pre-computed as correct. For instance, if the target angle was pre-computed as 40°, and a participant's response angle was 30°, he/she would have registered a pointing error of 10°.

Each participant performed three practice trials first before performing the test trials. To each participant, a total of 12 test trials, arranged in a randomized sequence that differed between participants, were presented. The target angles ranged from a minimum of 25° to a maximum of 333° (M = 191.83°, SD = 103.98). The number of trials completed, mean and standard deviation of pointing errors were computed automatically by the program at the end of the task. For any trial that was not completed by a participant, the program automatically registered a pointing error of 90°. To ensure accuracy in our data analysis, such data from uncompleted trials were removed in the computation of mean and standard deviation errors for each participant.

The MRM-T, designed originally by Money et al. (1965), was the second test used for assessing participants' SOA. It is a paper-based test and has been used previously by Friedman et al. (2020). We used it to assess the convergent validity of Friedman et al.'s 2020 SOT, that is, we wanted to examine how well pointing errors from the new SOT correlated with accurate performance on the MRM-T. In this study, we presented an open-source MRM-T used previously by Elman et al. (2013). It shows a paper map of an imaginary town, with a route running through its streets in the form of a continuous dashed line. The map shows 32 turning points in total and participants were told to imagine traveling along the continuous route and to indicate at each turning point whether the route turned left or right. At each turning point, each participant was told to write a capital letter “L” besides it if he/she perceived the route as turning left and to write a capital letter “R” if he/she perceived the route as turning right. A short route for pre-test practice was located at the bottom-right corner of the map.

All participants were tested individually over single experimental sessions that lasted approximately 1 h per participant. Each participant completed the SOT and MRM-T before performing the AT-CDT. The experimenter administered the first two tests in a counter-balanced order. Half of the participants performed the SOT first while the other half performed the MRM-T first. Before formal testing, participants were given some practice trials. In the MRM-T, participants practiced indicating three turning directions correctly on the short route located on the bottom-right corner of the paper maze. Importantly, the experimenter fixed the paper map in a portrait orientation in front of each participant during this practice phase and the subsequent 1-min test phase. A participant was only allowed to proceed to the test phase upon correct indication of all three turns on the practice route. In the presence of any errors, the particular participant was told to repeat the practice until no errors were made. Likewise, on the SOT, each participant performed three practice trials prior to the test phase. Before performing these trials, to ensure that all participants knew the identities of each object comprising the object array, the experimenter pointed to each object and mentioned their names to each participant. On each practice trial, automatic feedback showing the correct target direction was shown immediately after the registration of a response. In the event of large angular error (≥90°) on any practice trial, the particular participant was told to repeat that trial until there was a noticeable reduction in error. After passing the practice phase of each test, each participant was given maximum time-limits of 1 and 5 min, respectively, to complete the MRM-T and SOT in the formal testing phases. Before the start of each testing session, the experimenter told each participant to respond as fast and as accurately as he/she could and recorded each participant's test completion time with a stopwatch.

After completing the two spatial tests, all participants were introduced to the AT-CDT. They were given detailed descriptions of the main characteristics of a conflict event (mentioned above in “Conflict Event Generation” subsection) and told explicitly that the conflicts under examination in the current task referred only to lateral conflicts that involved a loss of minimum separation between a pair of aircraft at a distance of 5.0 nm or less. Before starting a practice trial, the experimenter presented a demonstration trial explaining the visual stimuli presented by NARSIM. This trial featured conflict event model B and involved three aircraft. With the simulation in progress on this trial, the experimenter introduced the radar blips representing each moving aircraft, the waypoints, and the data blocks containing information about each aircraft's callsign (first row), flight level (second row), true airspeed (third row), and landing runway (last row). First, they were told that all aircraft cruised at the same true airspeed of 450 knots and would land at the same runway, and that they did not need to pay attention to the information indicated on the third and last rows of the data block (Figure 3, right panel). Second, they were told to attend to the flight level data and informed that any pair of aircraft that were separated by 10 units (representing 1,000 feet) would not be in conflict. Third, they were told to pay close attention to aircraft heading changes and to all aircraft cruising at the same flight level.

In addition, all participants were shown a scale located at the bottom-right of the screen that indicated the lateral conflict distance threshold of 5.0 nm on the radar screen (Figure 3, left panel). This scale spanned 20 mm in length and subtended a visual angle of 1.8°. Together with this scale, the experimenter also brought up a measuring line by clicking on the scroll wheel of the mouse that enabled an automatic measurement of the lateral distance separating any pair of aircraft. Specifically, when the conflicting aircraft pair reached a distance approximating the length of the scale, the experimenter activated the measuring line, showing the magnitude of lateral distance separation. This procedure was done to give each participant an initial percept of what a loss of minimum lateral separation entailed and was not repeated in the practice and test trials. More importantly, all participants were told to indicate their detection of the conflicting aircraft pair by double-clicking their callsigns in quick succession. To demonstrate this, the experimenter double-clicked on the correct conflicting pair. With each series of double-clicks, time elapsed from trial onset was recorded to millisecond accuracy. This form of temporal data recording applied to all subsequent trials.

After the demonstration, the experimenter started a practice trial. This trial featured six aircraft and two conflict events based on models A and B. Each conflict event occurred in a discrete sector of the controlled airspace and involved three aircraft cruising together on a flight level (25,000 feet) that was distinct from the flight level of the other trio (30,000 feet). Each participant was given control of the mouse and told to indicate two pairs of conflicting aircraft in the same way demonstrated by the experimenter. A participant was only allowed to commence on the test trials after the correct detection of both pairs of conflicting aircraft. In the event of any incorrect detection, participants were re-briefed about the task and re-attempted the practice trial from the start. After completing the practice trial, all participants proceeded to the test trials. There was a total of nine test trials arranged in three trial blocks. Table 1 shows the flight parameters (i.e., ATD, aircraft numbers, flight levels) associated with these practice and test trials.

In each trial block, to ensure that there was a gradual exposure to trials of increasing difficulty, the trial with low ATD was always shown first, followed by medium and high ATD trials (see Table 1, column 2). As shown in Figure 5, three pairwise combinations of discrete collision sites were presented in each block of three trials. We configured them using a Latin square design, meaning that they appeared in a distinct and non-overlapping sequence within each trial block and ATD condition. This ensured that no two collision sites were in the same air sectors between trials and reduced the chances for any participant to predict the whereabouts of a conflict on any trial based on knowledge acquired from the trial that immediately preceded it. Across all trials, inclusive of the demonstration and practice trials, the flight levels of all aircraft ranged between 21,000 and 32,000 feet (M = 28, 111.11, SD = 2895.77) (see Table 1, last column). On the low, medium, and high ATD test trials, each conflict event occurred within clusters of three, four, and five aircraft cruising at the same flight level, respectively (see Table 1, column 4). Between trials, the flight levels assigned to each conflict event involving these clusters of aircraft differed by a margin ranging from 2,000 to 5,000 feet. Additional non-conflicting aircraft in the medium and high ATD trials cruised at flight levels that are safely separated from the flight levels of the quartets and quintets of aircraft designed for conflict detection by a minimum margin of 2,000 feet. Crucially, there was no repetition of the same flight level information across the test trials. This means that that no aircraft cruised at the same flight level on two consecutive trials. This deterred participants from making responses based on attentional biases or memory of aircraft with repeated flight level information.

On each test trial, like on the practice trial, participants responded by double-clicking on the callsigns of the aircraft that were perceived as coming into conflict. They were told that their responses were non-amendable and that they must think carefully before making each response. A time-limit of 3 min was given for the detection of both conflict events and each participant was told to respond as fast and as accurately as he/she could.

From each of the three tests, we selected relevant accuracy and response time (RT) measures for use in the correlation- and regression-based data analyses. We detailed their recording and computation in the subsections below.

Table 2 shows the descriptive statistics of all relevant test measures obtained from the three tests performed by the total sample of 20 participants. Total accuracy scores from the MRM-T and AT-CDT were converted into percentages for ease of comparison. For each participant, SOT mean RTs and and AT-CDT mean tLMS were computed to the accuracy of two decimal places. Consequently, the means of these temporal measures shown in Table 2 reflect the averages of means taken from each participant.

An examination of the data distribution of Table 2 test measures showed that that most participants achieved perfect or close to perfect accuracy scores on the AT-CDT (Mean % accuracy = 95). Specifically, a scrutiny of AT-CDT accuracy data using histograms, box-plots, and Q-Q plots showed a negatively skewed distribution (skewness = −1.28, SE = 0.51), with most participants scoring full marks (n = 11) or close to full marks (n = 8, accuracy score ≥ 15). Only one participant achieved a “low” score of “14.” A test of the normality assumption using the Shapiro Wilk's test further showed that the distribution of these accuracy scores deviated prominently from normality (Shapiro-Wilk's W = 0.750, p < 0.001). In consideration of the observed skewness in accuracy score distribution and normality violation, as well as the fact that accuracy had already been factored into the computation of mean tLMS, we removed AT-CDT accuracy in the correlational analysis presented below⁵.

With respect to AT-CDT mean tLMS computed based on correct detections, we applied a general measure of mean tLMS, averaged across both models, in the analyses below. To examine whether participants had a general tendency to respond before or after the start of a conflicting aircraft's heading change in each model, we computed two sets of mean tLMS values derived from conflict events A and B, respectively (see last two rows in Table 2). One-sample t-tests showed that the mean tLMS values derived from Models A and B, respectively, were higher than the mean reference tLMS values of –37 s that denoted heading change onset in Model A [t₍₁₉₎ = 4.31, p < 0.001, M(SE)_Difference= 6.77 (1.57)] and –43 s that denoted heading change onset in Model B [t₍₁₉₎ = 4.38, p < 0.001, M(SE)_Difference= 9.78 (2.23)] (see “Conflict Event Occurrence” subsection above, for details of computing these reference tLMS values). This showed that participants generally responded after the start of the conflicting aircraft's heading change in either model. Importantly, this finding suggested that most participants attended to aircraft heading changes and used such spatial orientation information to guide their correct detections of conflicting aircraft pairs.

Table 3 shows results from multiple bivariate correlations between MRM-T percent accuracy, SOT pointing errors (mean and SD), SOT mean RTs (raw and corrected), and AT-CDT mean tLMS. In particular, we computed four sets of mean tLMS values: an overall mean that encompassed all nine AT-CDT test trials and three categorical means associated with the three ATD conditions.

We found three sets of noteworthy findings. First, there were significant and moderately high negative correlations between MRM-T percent accuracy and the SOT pointing errors [r₂₀ = −0.50, p = 0.023 (M error); r₂₀ = −0.51, p = 0.021 (SD error)], These findings were important for showing that the SOT possessed convergent validity with respect to the MRM-T. Second, there were significant and moderately high positive correlations between AT-CDT mean tLMS obtained from the high ATD condition and the two types of SOT mean RTs [r₂₀ = 0.61, p = 0.005 (raw); r₂₀ = 0.57, p = 0.008 (corrected)]. These findings were crucial to the central aim of this study and we further examined them these using linear regression analysis in the section below. Third, there were significant and positive correlations between the four sets of mean tLMS measures that ranged from moderate to high (0.53 ≤ r₂₀ ≤ 0.95, p_s ≤ 0.015). Mean tLMS from the high ATD condition did not correlate highly with those from low (r₂₀ = 0.57, p = 0.008) and medium (r₂₀ = 0.53, p = 0.015) ATD conditions, suggesting that performance on the high ATD trials were not highly consistent with performance on the low and medium ATD trials (see ANOVA findings below, for details on the distribution of data points).

Continuing from the correlational analysis above, we regressed each of the four sets of AT-CDT mean tLMS values on SOT mean RT (raw and corrected, respectively) to address the main aim of this study, that is, the extent to which SOT performance could predict AT-CDT performance. By doing so, we examined the extent to which SOT mean RT affected AT-CDT mean tLMS across all ATD trials and across three subsets of trials, each showcasing a distinct level of ATD. Table 4 shows the results of these eight linear regression models. All beta coefficients matched their respective correlational coefficients shown in Table 3. Figures 7, 8 show the scatter-plots tied to these regression models. They show four scatter-plots apiece with SOT mean raw RT and corrected RT, respectively, as the predictor.

With alpha set at 0.017 (Bonferroni-corrected), both sets of SOT mean RTs emerged as significant predictors of AT-CDT mean tLMS in the high ATD condition [SOT raw RT: B = 0.58, t₍₁₈₎ = 3.24, p = 0.005 (see Figure 7A); SOT corrected RT: B = 0.60, t₍₁₈₎ = 2.96, p = 0.008 (see Figure 8A)]. Notably, around a third of the variance in AT-CDT mean tLMS was explained by variation in SOT mean RTs [R² = 0.37 (raw); R² = 0.33 (corrected)]. The remaining relationships between SOT mean RTs and AT-CDT mean tLMS were found to be low and non-significant [SOT raw RT: 0.21 ≤ B ≤ 0.36, 0.14 ≤ β_s ≤ 0.35, 0.130 ≤ p_s ≤ 0.551 (see Figures 7B–D); SOT corrected RT: 0.09 ≤ B ≤ 0.30, 0.05 ≤ β_s ≤ 0.27, 0.248 ≤ p_s ≤ 0.824 (see Figures 8B–D)].

With alpha set at 0.017 (Bonferroni-corrected), comparisons of the beta values from the three ATD conditions using Fisher's Z test of dependent correlations (Ramseyer, 1979) showed that the correlation between SOT mean RTs and AT-CDT mean tLMS was significantly stronger in the high ATD condition than in the low ATD condition [Z = 2.38, p = 0.009 (SOT raw RT); Z = 2.56, p = 0.005 (SOT corrected RT)].

Finally, to ascertain that bivariate normality was present and that our regression findings were not skewed by the presence of outliers, we further performed one-sample tests of normality (Shapiro-Wilk and Kolmogorov-Smirnov) on the standardized residuals emanating from the eight linear regression models. Joint assessments of normality test statistics and Q-Q plots showed that none of these distributions of standardized residuals deviated significantly from normality with alpha set at 0.05 (0.932 ≤ Shapiro-Wilk's W ≤ 0.974, 0.165 ≤ p_s ≤ 0.827; 0.095 ≤ Kolmogorov-Smirnov's D ≤ 0.153, 0.680 ≤ p_s ≤ 0.986).

Owing to the fact that the AT-CDT is not a psychometric test but a controlled experimental task that we designed to assess SOA, we performed a 3 × 3 Repeated Measures Analysis of Variance (RM ANOVA) on the mean tLMS values to examine whether or not our experimental manipulations exerted significant effects. With mean tLMS specified as the dependent variable (DV), we examined the main effects of ATD (low, medium, high) and trial block (TB; triplets of trials presented over three cycles), as well as the ATD × TB interaction effect. We applied ATD as the first independent variable (IV) to check for the presence of mental workload effect, that is, we wanted to know if an increment in ATD elicited more positive mean tLMS values that signified longer delays before responses. We applied TB as the second IV to check for the presence of spatial learning effect, that is, we wanted to know if participants acquired some conflict geometry-related knowledge as the trials progressed, becoming faster in response on trials presented later in the sequence than on trials presented earlier.

Through RM ANOVA, we found significant main effects of both ATD [F_{(2, 38)} = 21.32, p < 0.001, partial η² = 0.529, observed power = 1.00] and TB [F_{(2, 38)} = 9.91, p < 0.001, partial η² = 0.343, observed power = 0.976], plus a significant ATD × TB interaction effect [F_{(2, 76)} = 5.07, p = 0.001, partial η² = 0.211, observed power = 0.955]. Figure 9 shows this significant interaction effect by plotting out participants' mean tLMS distribution and averages on each test trial while Figure 10 shows the two significant main effects by plotting out participants' mean tLMS distribution and averages in each ATD condition and TB.

To examine the simple effects emanating from these omnibus test effects, two-tailed dependent t-tests were performed with alpha set at Bonferroni-corrected thresholds of 0.0014 for between-trial comparisons (0.05/36 possible comparisons) and 0.017 for categorical comparisons (between the three ATD conditions and between the three TB). Table 5 shows all pairwise differences that were found to be significant based on these post-hoc comparisons. The findings showed that participants responded the slowest on the second (M = −19.81, SE = 2.65) and sixth (M = −19.41, SE = 2.63) test trials, on which they experienced the medium ATD condition for the first and the hard ATD condition for the second time, respectively. Specifically, responses on these two trials were significantly slower than responses on the fifth and seventh trials (p_s < 0.001, see Table 5, rows 3, 4, 9, 10 after header), which showed the medium ATD condition for the second time and the low ATD condition for the third time, respectively. On the other hand, participants responded the fastest on the seventh trial (M = −39.41, SE = 2.31). Responses on this trial was not only significantly faster than those from the second and sixth trials (as aforementioned), but also significantly faster than those from the first, third, eighth, and ninth trials (p_s < 0.001, see Table 5, rows 1, 7, 5, and 11 after header).

In addition, categorical comparisons between the three ATD conditions showed significant differences for all three possible comparisons. On average, participants responded much slower on the high ATD trials than on the medium (p = 0.005) and low ATD trials (p < 0.001). They also responded slower on the medium ATD trials than on the low ATD trials (p < 0.001). On the other hand, participants responded the slowest within the first TB, and comparatively faster within the second and third TB (p_s = 0.001).