7. CHAPTER 7
7.4. Data Analysis Overview
7.4.1. Variables to be analysed
As justified in the previous section, twelve variables were collected during the experiment as follows:
1. Eye blink durations (average eye blink duration) 2. Eye blink frequency (average eye blink frequency)
3. The absolute value of average deviations from road centre. 4. The average reaction time for colour light changes.
5. Number of out of bounds from road 6. Maximum speed every one minute period 7. Average speed every one minute period 8. EPWORTH sleepiness scale vale
9. Number of hours sleep during the previous 24 hour period. 10. Age of participants
11. Time of day that the test was taken
12. Sleepiness scale before and after the simulation test (SLEPSCAL). The following measures were used for data analysis:
Predictor Variables
Physiological measures
• AVEBLKDU: The average eye blink duration (total blink duration is a length of closing - remaining closed - reopening time), averaged over a one minute period. More details are given in Table 7-3. Standard deviation of average blink duration was calculated for the final linear correlation (SD-AVEBLKDU). • AVEBLKFR: The average blink frequency. A moving average filter was used on eye blink frequency signals to sample over a one minute interval. More details are given in Table 7-3. Standard deviation of average blink frequency was calculated for the final linear correlation (SD-AVEBLKFR).
• MAXSP: The maximum speed. The simulator speed is set for a maximum of 60 km/h. The simulator checks every second during the test if the participants reach this set speed and this is recorded. A moving average filter was used for recorded data to sample over a one minute interval.
• AVESP: The average speed. The simulator records the average speed every second during the 40 minute test. A moving average filter was used to sample the AVESP data over a one minute interval.
Self reported measures
• EPWOSCL: The Epworth sleepiness scale. The Epworth questionnaire gives a numerical value of subjective drowsiness. The maximum score is 24 (from eight questions). If the participant’s total score is 11 or more will indicates that he/she is subjectively sleepy.
• NOHS: The number of hours sleeps during the last 24 hours.
• SLEPSCAL Sleepiness scale; all participants have to log their sleepiness level before and after the simulator test. Participants need to scale their sleepiness level from 1-7 (1=alert and 7=sleepy). • TODD: The time of the day, when the participants starts the
simulation test.
• AGE: Age is categorized according to two groups; younger drivers aged 20-29 years and older drivers aged 30-70 years.
Predicted Variables
• AVEDEV: The average deviation from the centre (calculated as the vehicle deviation from centre line and by the average deviation every 1 second period). These values go through a five minute moving average filter. More details are given in Table 7-3. Standard deviation of average deviations was calculated for the final linear correlation (SD-AVEDEV). • AVEREATM: The average reaction time. Every participant has to respond
to random colour light change runs simultaneously with the simulation test. Total reaction times during the 30-40 minute simulation test go through a moving average filter to sample over a one minute interval. Standard deviation of average reaction time was calculated for the final linear correlation (SD-AVEREATM).
• OUTOFBON: The out of bounds. Each participant is required to drive on the centre line of the road. If they go out of the road limits, this records the time during which the participant has gone out of bounds. OUT-OF-BON indicates the sum of out of bounds in each one minute time interval during the simulation test.
All measures were first computed over one-minute intervals. Data manipulation procedures were then undertaken to prepare data for statistical analysis. Initially, the first five minutes from all measures were deleted as discussed above. This was done
so that the data to be analyzed did not include the time when participants were suspected of “settling in” to the driving task. Even though all participants were given a practice driving session it was thought that, in the first five minutes of driving some participants demonstrated inconsistencies concerning their driving behaviour reactions, and physiological measures.
All measures collected through time were averaged in one minute blocks. Then mean and standard deviations were calculated for minutes 1 to 5, 2 to 6, 3 to 7, etc, giving a five minute moving average filter. The first five minutes of the data had been deleted.
After completion of the moving average procedure, all the data were arranged in five-minute intervals. Five or six-minute averages had been shown previously to have higher correlation values with driver performance measures than one-minute, two-minute, or four-minute averages (Wierwille, et al., 1994). As the results for five and six minute averages were close, the five minute filter was chosen, as being more convenient for data of 35 or 40 minutes duration. (See Table 7-3 for a pictorial overview of the data manipulation procedure.) The studies by Dinges et al. (1985) found that longer intervals were better in the detection and prediction of the danger state of drowsiness.
Table 7-3: Data manipulation procedure
The data analysis for this research composed of two major parts. The first part of the analysis consisted of correlation analyses of all the data. The purpose of the analysis was to determine which of the variables could possibly predict impairment due to drowsiness. The second part of the analysis consisted of linear regression analyses. The purpose of the regression analysis was to find one or more variables that would best predict impairment resulting from drowsiness. Figure 7-3 shows the implicit model: variables change (predicted variables) because of drowsiness and variables exemplify drowsiness (predictor variables).
Figure 7-3: Implicit Model, Predicted and Predictor variables for drowsiness detection