Pilot fatigue is a critical reason for aviation accidents related to human errors. Humanrelated accidents might be reduced if the pilots’ eye movement measures can be leveraged to predict fatigue. Eye tracking can be a nonintrusive viable approach that does not require the pilots to pause their current task, and the device does not need to be in direct contact with the pilots. In this study, the positive or negative correlations among the psychomotor vigilance test (PVT) measures (i.e., reaction times, number of false alarms, and number of lapses) and eye movement measures (i.e., pupil size, eye fixation number, eye fixation duration, visual entropy) were investigated. Then, fatigue predictive models were developed to predict fatigue using eye movement measures identified through forward and backward stepwise regressions. The proposed approach was implemented in a simulated shorthaul multiphase flight mission involving novice and expert pilots. The results showed that the correlations among the measures were different based on expertise (i.e., novices vs. experts); thus, two predictive models were developed accordingly. In addition, the results from the regressions showed that either a single or a subset of the eye movement measures might be sufficient to predict fatigue. The results show the promise of using nonintrusive eye movements as an indicator for fatigue prediction and provides a foundation that can lead us closer to developing a near realtime warning system to prevent critical accidents.
Fatigue is a critical reason for humanerrorrelated aviation accidents [
Fatigue can hinder the pilot’s ability to stay alert and be attentive during a shorthaul flight consisting of multiple consecutive flight missions. Shorthaul flights usually include 4–5 legs per day, whereas longhaul flights usually include 20 or more hours of nonstop single leg flight [
Some approaches have focused on observing physiological and subjective data including PVT, electroencephalography (EEG), electrooculogram (EOG), Samn–Perelli fatigue scale (SPS), and the Karolinska sleepiness scale (KSS) [
Building a pilot fatigue prediction model is crucial for developing timely alerting or scaffolding methods to prevent fatigueinduced aviation accidents. The effectiveness of a fatigue prediction model depends on the methodology used to evaluate pilot fatigue. For example, intrusive fatigue evaluation methods (requiring the task to be paused for assessing fatigue) will hinder the fatigue prediction model’s adaptation for realtime fatigue prediction. Hence, before developing the pilot fatigue prediction model, we explored the limitations of the various pilot fatigue evaluation methods so that we could implement the most appropriate method (or a combination of eye tracking and PVT) in our case. The various pilot fatigue evaluation methods implemented in earlier studies can be broadly categorized into subjective methods [
Subjective methods consist of evaluating pilot fatigue through selfassessment scores, for example, fatigue rating and sleepiness scales, that allow us to understand a pilots’ opinions and fatigued feelings [
In detail, PVT evaluates pilot fatigue by measuring the change in their taskbased performance [
The EEG method evaluates pilot fatigue by analyzing their brain waves, and it does not require the task to be paused [
The eye tracking method can be used to evaluate fatigue by analyzing pilots’ eye movements collected using a small eye tracker placed beneath a monitor or anywhere within the pilots’ visual field. Prior studies [
To the best of our knowledge, there has been no research on investigating the following research questions: (a) how fatigue increase might differ based on a pilot’s expertise for a multileg flight mission; (b) how fatigue levels increase after each flight leg; (c) how eye movement measures are correlated with fatigue levels (measured using PVT that proved to measure accurate fatigue levels as briefly explained above); (d) whether a fatigue predictive model can be created using only eye movement measures. Furthermore, PVT has several measures (such as reaction times, false alarms, and number of lapses), and there is no research on how to combine those into a single fatigue assessment measure. Motivated by our preliminary research efforts [
In this section, an indepth literature review based on the fatigue evaluation methods within the aviation and other relevant domains is provided below.
In the absence of any direct fatigue measurement approach, PVT has been proven to be the most effective and widely used fatigue evaluation method. PVT evaluates fatigue by assessing changes in an individual’s performance for a particular buttonpressing task when a visual stimulus (e.g., lights) is sequentially presented one by one on a display at random time intervals [
Note that RT increases with the rise in fatigue level. Furthermore, for RT < 150 ms, the button pressing action is considered a false alarm, which implies that either the onset of the visual stimulus was anticipated or the action was performed without seeing the stimulus. Conversely, the button pressing act having an RT > 500 ms is counted as a lapse, thus implying a temporary failure of concentration due to the fact of fatigue. Another possible measure, also used by researchers to assess fatigue, is the response speed, which is defined as the inverse of RT (i.e., 1/RT). For example, if RT is 200 ms, then the response speed is 0.005 ms^{−1}.
Different researchers have used one or a collection of the four PVT measures mentioned above to evaluate pilot fatigue. For example, the number of lapses was found to increase, whereas response speed decreased with a rise in pilot fatigue levels for shorthaul flight operations [
Regarding the fatigue prediction model development, prior studies (which used a hybrid approach of combining PVT and subjective measures) found different results for long and shorthaul flights. For example, for a shorthaul flight, the number of lapses was found to be important for pilot fatigue prediction [
Note that all these previous studies found similar trends, in which response times, number of false alarms, and number of lapses increased as fatigue increased. Response speed (being the inverse of reaction times) decreased with an increase in fatigue. However, to the best of our knowledge, each PVT measure was evaluated separately when we reviewed the previously published papers on PVT. If we used the PVT measures separately, then would we need to develop 3 times the regression models and would not know which one to apply for prediction. Hence, we would need a single unified PVT measure to develop a single prediction model. Therefore, we devised a simple unified measure which is the weighted linear combination of the three basic PVT measures (i.e., reaction times, number of false alarms, and number of lapses). More details are provided in
The abovementioned studies, due to the PVT method’s implementation, require the current task to be paused, which might not be either favorable or feasible when performing a piloting task. In addition, they include flight operations that only have single takeoff and landing tasks for both short and longduration flights. Thus, the results obtained cannot necessarily be transferred to our case of multiple takeoffs and landing tasks. Furthermore, these studies did not consider the effect of pilot expertise on their fatigue level.
Eye tracking is nonintrusive and can provide measures such as eye fixation position (or location), duration, pupil dilation, visual scanpath (i.e., the timeorder of the eye fixations that occurred on display), saccade, blink, and eyelid closure [
In more detail, previous studies utilizing only the eye tracking method for pilot fatigue evaluation suggested that expert pilots’ saccadic movements decreased with an increase in timeinduced fatigue for a single takeoff and landing operation [
We currently do not know whether similar results can be obtained for a longduration aircraft piloting task with multiple takeoffs and landing operations and whether a fatigue predictive model (i.e., regression model) can be developed for the same. In addition to the traditional measures, eye movement data, especially the saccadic eye movements, can be further processed to evaluate the pilots’ overall eye movement transition behavior using visual entropy [
Before calculating two visual entropy measures, we first needed to evaluate the eye fixation transition probability matrix, also called area of interest (AOI) transition probability using the design principles [
Visual entropy can be divided into transition entropy and stationary entropy [
Transition entropy:
Stationary entropy:
In
Note that having more eye fixations (leading to more eye movement transitions) does not mean that the entropy is higher. The entropy was calculated based on transition probabilities; therefore, we can have a higher entropy value with a fewer number of eye fixations.
Our proposed method consisted of two steps. In the first step, we evaluated pilot fatigue using PVT measures and investigated the correlation between the PVT measures and various eye movement measures. This correlation study helped us to evaluate the eye movement measures’ validity in assessing pilot fatigue for the given flight scenario. The second step involved developing a fatigue prediction model using a stepwise regression model where only the normalized eye movement measures were treated as predictor variables. Note that fatigue was assessed by normalizing and aggregating three PVT measures into a single fatigue measure. In other words, we assumed that PVT measures were very accurate in providing fatigue levels based on previous research [
The detailed analysis steps are as follows:
Assess fatigue level through PVT after each task. Measures are (a) reaction times (RT), (b) number of false starts (FS), and (c) number of lapses (L).
Collect eye tracking data, analyzing the data using the contextspecific areas of interest (AOIs). Measures are:
Mean eye fixation number on AOIs;
Mean eye fixation duration on AOIs;
Mean pupil size on AOIs;
Visual entropy (calculation process explained below).
The visual entropy evaluates the amount of randomness associated with the visual scanning strategy of the pilots. Higher visual entropy value means that relatively more randomness (in eye movements) exists. We hypothesized that experts’ visual entropy would be lower than that of the novices, meaning that the novices’ eye movements might show more randomness, especially when fatigued. Although we followed the procedures provided in [
Plot the relationships of the variables and investigate the correlations between the PVT measures and the eye tracking measures. The measures were all those provided in Steps 1 and 2. Step 3 was needed to first see whether linear correlations could be observed prior to applying multiple regression. In other words, different regression models should be applied based on the relationships. For example, if the relationship among the variables were quadratic, then a quadratic regression should be applied.
Create a “unified” PVT measure by combining the PVT measures of RT, FA, and L. The unified measure (S) is expressed as follows:
Discover an optimal regression model that can predict fatigue using one or more eyetracking measures. Stepwise regression approach was applied (both forward and backward) to discover the optimal regression model. We assumed that the unified PVT measure accurately represented one’s fatigue level, and we found eye tracking measures that could predict fatigue level. All eye tracking measures were normalized, meaning that the minimum and maximum values obtained from all the experiment participants were mapped to 0 and 1. The full model and associated variable for the backward regression is:
A moderate fidelity flight simulation environment was created that involved the shorthaul multiphase flight. Details are as follows.
A total of twenty pilots participated in the experiment. Ten participants were defined as “novices” who had moderate expertise (pilot experience: mean of 18 months and SD of 2.4), less than 36 months of experience, and at least met the minimum requirements of 40 h of actual or simulated IFR flights. The other ten participants were defined as “experts” who had more expertise (pilot experience: mean of 42 months SD of 4.5), more than 36 months of experience, and expressed they completed substantially more IFR flight hours (at least more than twice) than the minimum requirement of 40 h. Unfortunately, all pilots were not able to exactly recall their IFR flight hours; therefore, the statistics are not provided.
The power analysis indicated that the sample sizes provided reasonable power of 0.91 for the mixed design of withinsubjects design related to the tasks and betweensubjects design related to the expertise. In addition, other research papers related to evaluating pilots’ performance had a mean sample of ten pilots [
Microsoft flight simulator software (i.e., FSX) was used for generating the Boeing B52 aircraft and the flight scenarios. B52 aircraft was selected to possibly induce more visual attention from the pilots. However, since a moderate fidelity flight simulator was used, piloting a simulated B52 should be not as difficult as piloting the actual B52.
The PVT measures were assessed using the Psychology Experiment Building Language (PEBL) software version 0.13 [
The four consecutive tasks (tasks 1–4, which were equivalent to each leg) are provided in
IFR flights mean that the pilot does not have visibility out the window and has to rely on the information obtained from the flight instruments. After completing each flight task, pilots underwent the PVT, which lasted for approximately 5 min (providing 30 stimuli during the 5 min), following the guidelines offered [
The response variables extracted from the PVT task were mean reaction times, mean number of lapses (i.e., number of reaction times greater than 500 milliseconds), and mean number of PVT false starts (i.e., defined as the number of reaction times less than 150 milliseconds). In addition, the unified PVT measure (see Step 4 within the proposed analysis approach above) was calculated using the three PVT measures by assigning equal weights. Contextdependent important AOIs for an IRF flight were identified as shown in
Twoway mixed model analysis with repeated measures were applied to consider tasks (i.e., tasks 1–4) and the expertise (i.e., novices vs. experts). After, the relationships among the variables were plotted followed by correlation analysis. After identifying the linear relationships, stepwise regressions were conducted using Equation (4).
Descriptive statistics (i.e., means and standard errors) are plotted in
The mixedmodel (i.e., mix of betweensubjects design of expertise and withinsubjects design of task) analysis results are provided in
In
Descriptive statistics for the overall tasks are plotted in
Overall, the mean eye fixation number trended downwards as the task number increased. Moreover, expert pilots showed higher mean number of eye fixations than those of the novice pilots for all tasks. On the other hand, mean eye fixation duration showed an increasing trend over the course of the flight for both experts and novices (see
In addition,
The results from the mixedmodel analysis show significant effects of both pilot experience and task number, and their interactions on all eye movement measures (see
The increasing and decreasing trends were quantified through the correlation analysis shown in
After identifying that high correlation exist among the variables, two types of regression models (i.e., full models and optimized models) were investigated as provided below. Note that we chose to conduct the regression analysis using the unified PVT measure instead of using each measure separately. Details of the reason are provided in
Multiple linear regression results: The multiple linear regression analysis using the unified PVT measure (
Novice pilots (full model):
Expert pilots (full model):
Stepwise regression results: The results of the stepwise regressions are provided in Equations (7) and (8). For the novice pilots, to predict
Novice pilots (optimized model):
Expert pilots (optimized model):
Steps of the stepwise regression analysis are provided in
Note that it is possible to have positive correlation but obtain a negative regression coefficient in a multiple regression model due to the effect of other variables [
In summary, the increase in fatigue was verified through the PVT measures of reaction time, number of lapses, and number of false starts, and the results accord with many previous research efforts in aviation [
We learned that, unlike novices, the expert pilots had a greater number of eye fixations and shorter eye fixation duration on the context dependent AOIs throughout the flight mission. The results accord with a previous research [
Furthermore, in the case of the pupil size, it became progressively smaller (for both expert and novices) as fatigue increased over the course of the flight mission. The results are similar to existing research [
Regarding the visual entropy outputs, both the stationary entropy and the transition entropy increased with higher fatigue levels. One possible reason might be that, with higher fatigue levels, pilots’ visual search strategy became more random in nature resulting in higher entropy values. Expert pilots showed significantly lower visual entropies (both stationary entropy and transition entropy) compared to novice pilots (see
The regression results show that, depending on the level of expertise of the pilot (experts vs. novices), a different set of eye tracking measures can be used for predicting fatigue. Furthermore, limited to our experiment conditions in a multiphase consecutive flight mission, the optimized models show that some eye movement measures can be more effective at predicting fatigue than others. Specifically, when observing the optimized models (Equations (7) and (8)), eye fixation duration was a significant predictor variable for both pilot groups, whereas the eye fixation number and the visual entropies can be additionally useful when assessing the fatigue of the novice pilots.
In addition, note that only eye fixation duration (FD) was sufficient in the optimized model for the expert pilots even though the eye fixation numbers (FN) seemed to be equally highly correlated. The reason is that FN was highly correlated with reaction times (RT) and number of lapses (L), but not highly correlated with false starts (FS). Since we used the unified PVT measure that considers all three measures of RT, L, and FS, the stepwise regression analysis resulted in not including FN as a predictor and only using FD was sufficient, in which the results are limited to our experiment conditions. We do not recommend the stakeholders to simply use the computed coefficients and the predictors in their unique environment. We do believe that the stakeholders could benefit by applying our developed research methods. We would be very interested in any insight other researchers could provide if they obtained similar or different regression models.
In detail, the results make us question whether we should only use the optimized models to predict fatigue. The important contribution of this research is that we were able to discover that all eye movement measures introduced in this paper are somewhat strongly correlated with fatigue, and some eye movement measures might better predict fatigue over other measures. The optimized models can vary based on individual differences, experiment settings, and/or the flight task types. Therefore, we recommend that the proposed research approach should be used as a foundation that can be further customized based individual needs and flight environment.
Furthermore, it will be possible to use each eye movement measure separately or in different combinations to provide multiple evidence (or accumulated evidence) to better detect and verify fatigue levels. To the best of our knowledge, the general guidelines are recommending a break after piloting an aircraft for a certain number of hours or legs. The multimodal analysis approach provided in this paper can be utilized to develop near realtime fatigue detection models that can be used as a tool to manage fatiguerelated risk by proactively detecting fatigue of pilots.
One of the limitations of this research is that we collected and analyzed the data based on each flight phase (or leg) rather than across a continuous flow of time. The reason that we chose the above option was to compare the eye movement measures against the discrete PVT measures. Therefore, future research involves devising methods to evaluate fatigue continuously using only eye movement measures. The continuous evaluation can be done based on time (i.e., seconds, minutes, hours) or based on detailed events during takeoff, cruising, and landing.
Another limitation in on defining an expert and a novice. How to define an expert and a novice has always been an issue raised by the research communities in all applications. Although we have used the thresholds based on the inputs of the flight instructors, opinions can differ, and unfortunately, the participants were not able to precisely recall their IFR flight hours. However, we believe that out classification of the participants into two groups were somewhat successful, as we did obtain distinctive differences between the two groups. We are planning to apply a set of carefully constructed criteria for followup research.
In addition, this research is concentrated on providing aggregated outputs. It is possible that individual differences can exist. Therefore, future research involves investigating whether individual eye movement characteristics, especially the individual’s visual scanning patterns, differ as fatigue levels increase. The analysis of the visual scanning patterns involves developing algorithms to effectively characterize and compare those differences.
In terms of the methodology, we have proposed the concept of the unified PVT measure, but more indepth analysis is required on how to assign an optimal weight to each PVT measure. In this research, we assumed equal weights, but our assumption might be incorrect. Discovering an optimal weight value for each PVT measure is a challenging task which can be investigated through various algorithms and associated sensitivity analyses. We are currently working on how improve the regression models by developing appropriate algorithms that can find optimal weight values.
In addition, we had assumed that the initial fatigue levels of all the participants should be somewhat similar since the sleep and experiment time were controlled to the best of our abilities. In our future research, baseline measurements of initial fatigue should be obtained before the experiment is conducted.
The reason for the significant differences in term of the PVT measures seems that the experts might have developed more effective visual scanning strategies to reduce fatigue, and more indepth analysis on the visual scanning strategies will be needed as future research. In more detail, the visual scanpaths were analyzed using the concept of visual entropy in this research; however, the visual scanpaths can be also characterized and classified based on the concept of visual groupings [
Finally, this research can be used as a foundation to further develop near realtime fatigue detection models that can be used to alert the stakeholders and provide scaffolding options to the pilots, but we currently do not know what the threshold should be to trigger such alerts or the scaffolding options. If we could identify the possible thresholds, then the alerting and scaffolding options can be used in conjunction with the Boeing Alertness model [
Conceptualization of the research topic and the methodology were developed by S.N. and Z.K. Experiment scenarios were designed by S.N. Data were collected by S.N. Data analysis approaches were devised by S.N. and Z.K. Data analysis was performed by S.N. and S.M. Data analysis results were validated by Z.K. and K.K. Original draft was prepared by S.N. and S.M. Final draft was prepared by Z.K. and K.K. All authors have read and agreed to the published version of the manuscript.
This material was based upon work supported by the National Science Foundation under Grant No. 1943526. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board of the University of Oklahoma. Approved protocol code is 7325.
Informed consent was obtained from all subjects involved in the study. Written informed consent has been obtained from the participants to publish this paper.
Data are available by contacting the corresponding author, Ziho Kang.
We sincerely thank the aircraft pilots at the University of Oklahoma who participated in this simulated flight experiment.
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
Measures used to investigate fatigue for a multiphase flight task: FN is eye fixation numbers, FD is eye fixation durations, PS is pupil size, Ht is transition visual entropy, Hs is stationary visual entropy, RT is reaction times, FS is number of false starts, and L is number of lapses.
Four consecutive tasks (without any rest) labeled as tasks 1 through 4. Each task lasted approximately 1 h. The total duration was approximately 4 h.
Contextspecific AOIs that were defined based on the instrument fight rules (IFRs). The AOI names were as follows: engine oil pressure: EOP; engine indicators: EIs; enhanced visual screen: EVS; attitude indicator: ATT; horizontal situation indicator: HS; flight command indicator: FC; altimeter: ALT; airspeed indicator: AS; true airspeed indicator: TS; heading indicator: HI; vertical velocity indicator: VV; radar altimeter: RA; Mach indicator: MI; standby horizon indicator: SHS. Most of the eye fixations occurred on these AOIs during an IFR flight when we observed the recorded data after the experiments. The response variables related to the eye movements were eye fixation number on the AOIs, eye fixation duration on the AOIs, pupil size, and visual entropy (both transition and stationary entropy).
Means and standard errors of the PVT measures.
Examples of visual scanpaths of an expert and a novice pilot: The yellow circles represent the eye fixations where the numbers represent its index. The yellow lines represent the saccades. The size of the eye fixation circles have been kept at a fixed size for visual clarity. In addition, only 40 s of data are provided for each sample. FN is the eye fixation number, and FD is the eye fixation duration.
Means and standard errors of the eye movement measures.
Classification of existing research in pilot fatigue: Classifications are mostly based on the fatigue evaluation method. The last three listed in the table are studies not related to fatigue but worth mentioning.
Research Related to Pilots’ Fatigue  Research Topic  Fatigue Evaluation Method  Expertise  Single or Multiple Takeoff Landings  Short (1~3 h) vs. Long Duration (3+ h) Flight  Statistical Method 

[ 
Fatigue  Subjective  Experts  Single  Short and long  Multiple regression and ANOVA 
[ 
Fatigue  Subjective  Experts  Single  Long and Short  linear mixedmodel 
[ 
Workload and fatigue  PVTand Subjective  Experts  Single  Short  Stepwise Regression and correlation 
[ 
Reaction time  EEG  Novice  None  Short  Robust linear model 
[ 
Mental workload and fatigue  EEG        Literature review 
[ 
Fatigue  EEG  Novices  Single  Short  Classification model 
[ 
Fatigue  Eye tracking  Experts  Single  Short and Long  ANOVA and linear regression 
[ 
Fatigue  Eye tracking  Experts  Single  Short  Pre/PostTest design 
[ 
Fatigue  Eye tracking and PVT  Novices  Single  Short  ANOVA and regression 
[ 
Fatigue  Eye tracking and PVT  Novices and experts  Single  Short  MannWhitneyWilcoxon tests 
[ 
Fatigue  Eye tracking and PVT  Novices  Multiple  Long  KruskalWallis test 
[ 
Fatigue and sustained attention  PVT and Subjective  Experts  Single  Long  Linear mixed model regression 
[ 
Fatigue and Performance  PVT and Subjective  Experts  Single  Long and ultralong  Mixedmodel ANOVA 
[ 
Fatigue and performance  PVT and Subjective  Experts  Single  Long and ultralong  ANOVA 
[ 
Fatigue  PVT and Subjective  Experts  Single  Short and Long  Statistical/Machine learning model 
[ 
Fatigue  EEG  Novice  Single  Long  ANOVA 
[ 
Fatigue  Eye tracking and PVT  Experts  Multiple  Long  ANOVA 
[ 
Cognitive load  Eye tracking  Experts  Single  Short  Paired ttests 
[ 
Workload  EEG, Eye tracking, and Subjective  Experts  Single  Short  ANOVA and correlation 
[ 
Performance  Eye tracking  Novices and experts  Single  Short  ANOVA 
Example of two different eye fixation transition probability matrices.




TO  
From  A  B  C  D 
A  0  0.33  0.33  0.34 
B  0.33  0  0.33  0.34 
C  0.33  0.34  0  0.33 
D  0.33  0.33  0.34  0 




TO  
From  A  B  C  D 
A  0  0.99  0.01  0 
B  0  0  0.99  0.01 
C  0.99  0  0  0.01 
D  0  0.99  0.01  0 
Results of the mixed model analysis of variance on PVT measures: Exp is expertise factor (experts vs. novices) related to the betweensubjects design, and task is the task factor (tasks 1, 2, 3, and 4) related to the withinsubjects design.
BetweenSubjects  WithinSubjects  

F (1,18) 


F (3,54) 



Reaction time (RT)  
Exp  82.45  <0.001  0.8  
Task  177  <0.001  0.91  
Exp × Task  5.26  <0.003  0.23  
Lapse (L)  
Exp  104.7  <0.001  5.26  
Task  35.11  <0.001  0.66  
Exp × Task  4.67  <0.001  0.21  
False start (FS)  
Exp  90.72  <0.001  0.83  
Task  39.87  <0.001  0.69  
Exp × Task  6.99  <0.001  0.28 
Mixedmodel analysis on eye movement measures.
BetweenSubjects  WithinSubjects  

F (1,18) 


F (3,54) 



Eye fixation number (FN)  
Exp  72.41  <0.001  0.80  
Task #  157.12  <0.001  0.89  
Exp 
5.25  <0.003  0.22  
Eye fixation duration (FD)  
Exp  459.9  <0.001 


Task #  168.75  <0.001  0.90  
Exp 
7.51  <0.001  0.29  
Pupil size (PS)  
Exp  101.89  <0.001 


Task #  408.79  <0.001  0.96  
Exp 
21.24  <0.001  0.54  
Transition entropy ( 

Exp  210.88  <0.001 


Task #  200.75  <0.001  0.92  
Exp 
9.15  <0.001  0.34  
Stationary entropy ( 

Exp  119.11  <0.001 


Task # 

<0.001  0.81  
Exp 
3.85  <0.014  0.18 
Results of the oneway repeated measures analysis of variance on eye movements measures in which the task number (tasks 1, 2, 3, and 4) is the factor.
DV  Experts  Novices  

F (3,27) 

F (3,27) 


Eye fixation number (FN)  64.66  <0.001  107.72  <0.001 
Eye fixation duration (FD)  151.37  <0.001  71.98  <0.001 
Pupil size (PS)  160.11  <0.001  264.57  <0.001 
Transition entropy ( 
125.02  <0.001  98.08  <0.001 
Stationary entropy ( 
42.39  <0.001  38.59  <0.001 
Correlations among PVT measures and eye movement measures.
Expert  Novice  

RT  L  FS  RT  L  FS  
FN  −0.69  −0.61  −0.49  −0.84  −0.73  −0.76 
FD  0.76  0.68  0.61  0.88  0.78  0.74 
PS  −0.81  −0.56  −0.53  −0.86  −0.78  −0.70 

0.75  0.63  0.63  0.79  0.68  0.70 

0.65  0.56  0.59  0.73  0.64  0.62 
Stepwise regression (backward) results with unified PVT measure as response and eye movement measures as predictors for both expert and novice pilots.
Variables  Expert  Novice  

Step I  Step II  Step III  Step IV  Step V  Step I  Step II  
(Constant)  0.31  0.31  0.29  0.20  0.23  0.46  0.43 
FD  0.33  0.33  0.35  0.40  0.65  0.42  0.45 
FN  −0.05  −0.05  −0.27  −0.31  
PS  −0.07  −0.09  −0.08  0.42  
HT  0.25  −0.07  0.25  0.29  0.42  −0.38  
HS  −0.007  0.25  −0.40  
Adjusted R  0.61  0.62  0.63  0.64  0.64  0.83  0.84 
F  13.26 *  17.06 *  23.32 *  35.61 *  68.72 *  39.17 *  50.0 * 
AIC  −145.12  −147.1  −149.02  −150.78  −151.15  −177.03  −178.76 
*