Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
Students’ Sleep and Academic Performance
Project Lead: Ângela M. Katsuyama, UW Biology
Advisor: Horacio O. de la Iglesia, UW Biology
UW e-Science Institute Liaison: Bill Howe, Associate Director and Senior Data Science Fellow, Computer Science & Engineering, and Daniel Halperin, Director of Research for Scalable Data Analytics
Introduction and scientific questions:
This project investigates the impact of sleep in college academic performance. We hypothesize that poor academic performance in college students correlates with poor sleep behaviors. To address this hypothesis, we collected data from 72 senior students enrolled in the Spring 2014 Biological Clocks and Rhythms course. Some of the considered sleep parameters for analysis were: chronotype (preference to be a morning vs. evening type, as mismatches between the chronotype and the work schedule can lead to poor performance), social jetlag (difference in sleep timing between school days and weekend due to usual compensation of sleep debt on the weekends), variability of sleep onset, offset, duration, etc. Sleep parameters were recorded via sleep diaries, as well as with wrist data loggers (see below); performance was measured through grades.
The datasets consisted of: 1-) Activity and light exposure to three wavelengths, as well as white light, collected over 6 days (including one weekend) using a wrist actimeter (ActiWatch®); 2-) Sleep diary containing information about bed time, wake time, rise time, sleep duration and number of disturbances throughout the night (and of what kind), all self-reported throughout the days the student was wearing the watch; 3-) Chronotype score based on validated questionnaires from the literature; 4-) Grades (midterms, quizzes and final grade).
Our main goal was to determine potential correlations between sleep parameters and grades. We specifically assessed how light exposure is associated with to sleep patterns, and whether day-to-day variability or weekday vs. weekend variability have impact in academic performance. The challenge was to automatize data analysis across the entire population of students so this project could be scaled-up.
To achieve our goals, the following steps were taken:
1-) Parsing the activity and light data (by Bill Howe), which generated two datasets: one dataset containing the watch statistics used for comparisons with sleep diary information, and another dataset containing the raw data used for waveform analysis. Waveforms portray activity counts or light exposure across time.
2-) Generation of SQL Share queries to obtain the following information:
- Waveforms: the original 15 seconds of activity/light data was binned into 10-min intervals to obtain average/SEM across time, for weekday and weekend, for each student. This dataset was then used to calculate the average/SEM for the entire class. Student codes were used as identifier at all times for confidentiality purposes. These automatically-generated waveforms were compared to manually-generated waveforms.
- Bed time, wake time and rise time comparisons between actiwatch data and sleep diary information to assess accuracy of self-reported data.
- Correlations between the sleep parameters (averages and variability of bedtime, waketime, risetime, sleep duration, chronotype score, social jetlag), and grades (final midterm grade, final quiz grade, final weekly quiz grade, final quarter grade), as well as between sleep parameters and gender. The variability in sleep times was based on standard deviation in minutes.
- Chronotype analysis to compare scores from two different chronotype questionnaires used in this study and to correlate chronotype to social jetlag.
- The waveforms obtained from automatic analysis using SQL share are identical to waveforms obtained from manual analysis. However, automatic analysis allows data processing and plotting within minutes/hours as oppose to days/weeks.
- During weekends, students go to bed later, wake up later and sleep more than during weekdays. On weekends, the light exposure in the morning is also delayed.
- Self-reported diary information correlated with actiwatch data when looking at bed time and wake time.
- Surprisingly, we did not find a correlation between sleep onset mean or sleep onset variability with grades. The only significant results were when sleep diary wake time variability was compared with some grades and when sleep diary rise time variability (regarding the time when the students really got out of bed) was compared with the final midterm grade. However, higher grades were correlated with higher standard deviations, which was the opposite of what we predicted. Correlations were also not so strong. Therefore, we need a bigger sample size to make conclusive remarks.
- Our collected data agrees with the literature: 1-) late chronotypes tend to have bigger social jetlags, 2-) chronotype questionnaires scores correlate with each other, 3-) female tend to be earlier chronotypes than males of the same age group.
Some of the future directions are:
1-) Use iPhyton for graphing.
2-) Automatically extract individual waveform information.
3-) Obtain a bigger sample size to improve quality of results.
4-) Publish findings.