My semester long project for Psych 102 - done entirely in R.
https://drive.google.com/file/d/10gyJE_BRbPnxoryNV0D2Fv0TNSS7Z8In/view?usp=sharing
Abstract:
Existing research has demonstrated a strong correlation between stress levels and sleep quality. The current study further investigated this negative correlation between sleep and stress and in addition, examined the interaction of stress caused by the current coronavirus pandemic on sleep. As expected, the first hypothesis was met, in that stress negatively predicts sleep. The data in the study showed a strong, statistically significant relationship between stress and sleep, along with strong statistical power. The second hypothesis attempted to look at the interaction role of the COVID-19 pandemic, whereby the relationship between stress and sleep was stronger (a steeper negative slope) for participants who report that a greater degree of their stress was related to the pandemic. The data collected in the study seemed to suggest that there was a reverse main effect of the pandemic; that is to say, sleep increased as a result of the pandemic, but this was not statistically signficant. In addition, the actual interaction of the pandemic stress on sleep in the current study was also not statistically significant. Recent studies such as that by Katuna and Lau (2020), have however suggested that the relaxed schedule of the pandemic lockdown have allowed for longer sleep times though there were limits to actual sleep quality due to a lack of sleep buildup from decreased physical activity.
# Introduction
Research on stress and its impact on sleep has been ongoing for a while now and the strong negative relationship between them is well established. For instance, it has been found that insomniacs had overactive cortisol levels, which is a measure of hormonal stress; cortisol levels were especially elevated at the time of sleep onset (Vgontzas et al., 2001). To underscore the importance of sleep, lack of REM sleep which helps delete the emotional component of memory is thought to contribute to PTSD (van der Helm et al., 2011). REM sleep also serves a number of functions such as housecleaning extraneous memories accumulated during the day (Crick & Mitchenson, 1983). Given the crucial role of sleep for both forgetting and learning (Walker, 2009), the role of stress as a predictor is an important factor. The current study attempted to further confirm this negative correlation between sleep and stress.
In addition, this study also investigated the impact of stress caused by the current COVID-19 pandemic on sleep. Existing research, though recent due to the newness of the pandemic itself, is steadily growing. Unlike natural disasters like earthquakes and hurricanes, which tend to be localized events, the pandemic has been an unprecedented prolonged and ongoing stress event at a global level with the associated stresses over finances, employment, income, food insecurity, access to medical care, social isolation, as well as challenges of juggling family and work obligations (Morin & Carrier, 2020). Such intense and prolonged stress is bound to have an impact on sleep, especially, “at a time when healthy sleep is particularly important to cope adaptively with this crisis and uncertainty about the future” (Morrin & Carrier, 2020).
A September 2020 study by Katuna and Lau looked at the impact of COVID-19 from a sleep health perspective. They point to data such as that from the National Alliance on Mental Illness of New York City, which showed a 60% increase in calls relating to stress since the pandemic related lockdown was initiated.
At the same time, Katuna and Lau (2020) point to the more relaxed schedule of the lockdown which actually allows for, “longer sleep opportunities and total sleep time.” While it may not make much of an impact on those who enjoyed good sleep earlier, it may benefit those who had been sleep deprived due to the nature of their occupation and time spent in commuting. They do highlight however that there is less actual sleep pressure built up during the day due to the lack of physical activity which can limit improvements to actual sleep quality.
This study thus investigated both the main effect of stress on sleep as well as the interaction effect of stress induced by the pandemic on sleep.
Hypothesis 1: I hypothesize that people sleep less and stress predicts that.
Hypothesis 2: I hypothesize that there is an interaction, whereby the relationship between stress and sleep is stronger (a steeper negative slope), for participants who report that a greater degree of their stress is related to the COVID-19 pandemic.
# Methods
## Participants
Data was downloaded from Qualtrics survey software in the form of a .csv file and read into R to collect information on participants.
`r nrow(responsedata)` participants were recruited through Academic Prolific. Participants were limited to those living in the United States, age range from `r min(responsedata$agr)` to `r max(responsedata$agr)`, with a mean age of `r mean(responsedata$agr)` years, and standard deviation of `r sd(responsedata$agr)`. `r sum(responsedata$gdr == 1)` were female, `r sum(responsedata$gdr == 2)` were male and `r sum(responsedata$gdr == 3)` identified as non-binary or other.
## Material
Data was collected via a Qualtrics survey. The survey was intentionally kept very simple and short to encourage maximum participation.
Two existing scales were used. The first was the single item Sleep Quality Index (Snyder et al., 2018). The SQI is a 11-point likert scale and scores 0 as terrible sleep, 1-3 as poor sleep, 4-6 as fair sleep, 7-9 as good sleep and 10 as excellent sleep. In filling out the SQI, participants were asked to consider other components of sleep such as the hours of sleep, frequency of nighttime waking, ease of falling asleep, how refreshing their sleep was and, "how often they woke up earlier than they had to in the morning" (Synder et al., 2018).
Despite being a single item measure, the SQI has been found to show both validity and reliability, relative to the longer sleep questionnaires used in depression and insomnia patients (Synder et al., 2018).
The other established scale that was used was the four item Perceived Stress Scale (PSS-4) which asks for thoughts and feelings over the last month (Cohen et al., 1983). The PSS-4 uses a five-point likert scale with zero being never and four being most often. Individual line items ask, “In the last month, how often have you felt, #1 that you were unable to control the important things in your life, #2 confident about your ability to handle your personal problems, #3 that things were going your way, and #4 difficulties were piling up so high you could not overcome them" (Cohen et al., 1983). As per the scale requirements, items two and three were reverse coded, and individual scores were added up for a single score; the highest possible score of 16 indicating the maximum stress.
The PSS-4 is a well established scale that has been around for a few decades. Initially developed as a 14 item scale (PSS-14), shorter subsequent versions (PSS-10 and PSS-4) were later introduced.
An additional single item measure, "pandemic stress score;" was added to the survey in the form of a five-point likert question; namely - "How much of your stress is related to the pandemic?." The scoring was zero for "none of my stress," to four being, "all of my stress." This measure was used for lack of an existing measure as the pandemic itself is a recent global phenomena. As a result, this measure has not passed any test of validity or reliability.
The resultant survey contained a total of 8 questions, including two questions on demographics of age and gender (male, female, other).
## Procedure
This was an observational study with all data for the study collected over a twenty four hour period through Prolific Academic and with the use of the Qualtrics survey describe above.
The dependent variable in this study was sleep quality which was measured by the single item SQI.
The independent variable in this study was stress, which was measured in two ways. One was via the four item PSS-4. The second was the single item "Pandemic Stress Score," created for this survey as described above.
# Data Cleaning
Some data cleaning was required. The Qualtrics survey software automatically scores data using a 1-5 score, with 1 being the lowest score and 5 being the highest. However, the two established scales used in this study, the SQI and PSS-4, use a 0-4 score, with zero being the lowest score and four the highest. Data for all three variables was therefore adjusted by one to bring them all to a 0-4 scale. A total for the PSS-4 scale was generated after reverse-coding items two and three as per the scale instructions.
# Descriptives
## Means and Standard Deviations
Descriptive statistics for the variables are summarized and reported in the table below.
## Tests for Normality
The Shapiro-Wilk test was run on all variables to check for normality and ggplot was used to create histograms to visually check for skewness. The R output for the Shapiro-Wilk is given below.
## Test for Scale Reliability
Cronbach's Alpha was calculated for the PSS-4 scale and the R-output displayed below. Cronbach's Alpha was not calculated for the other two measures, SQI and Pandemic Stress Score, as both are single item measures.
# Data analysis Plan
R [@R-base] and Papaja [@R-papaja] were used for all the analyses.
To test the first hypothesis, a multiple regression type of analysis was run, which would indicate if stress (as measured by PSS-4) predicts sleep (SQI score).
To test the second hypothesis, a regression interaction model was run, to see if the relationship between sleep (SQI score) and stress (PSS-4) was especially strong if people reported that their stress is mainly related to COVID-19 (Pandemic Stress Score).
A power analysis was also run to check for effect size and power for this study.
## Libraries
The following libraries were used for the analyses:
## Analysis 1: Multiple Regression Interaction
The R-output of the multiple regression interaction analysis is displayed below.
## Analysis 2: Power Analysis and Effect Size
The R-output for the Power analysis is displayed below.
# Data Visualization
## Plot 1: Normalcy and Skewness of the Data Variables.
Data for all three variables was also visualized in the form of histograms below.
## Plot 2: A look at interaction
The output of the multiple regression interaction is further represented pictorally in Plot 1 below.
# Results
The pictoral visualizations of the data (the SQI sleep score, the PSS-4 score, and Pandemic Stress Score) in the form of the histograms in Plot 1 appeared to approximate normal distributions visually. The Shapiro-Wilk test had however indicated that all three variables differ from the normal. Attempts to transform the data using log or squareroot seemed to only skew the data, so no transformation was done on the data as regressions can typically handle some level of assumption violations.
Cronbach's alpha for the PSS-4 scale for this study was 0.79 which meets the acceptable threshold of 0.70 for scale reliability. This measure indicates the degree to which the set of items in the scale co-vary, relative to their sum score.
The Power analysis revealed a large effect size of 0.38 (as it above the threshold of 0.3) showing a strong relationship between the variables. The study also had strong power at 0.99, indicating that if this study was run multiple times, a statistically significant difference would be obtained 99% of the time.
Descriptive Statistics were summarized in the table in the Descriptives section. Means for the three variables PSS Score, Pandemic Stress Score and SQI score were 7.73, 1.86 and 6.53 respectively and the standard deviations 3.10, 0.91 and 6.00 respectively.
Running multiple regression interaction analysis on the data indicated that if there was no stress (PSS score or Pandemic stress score), then Sleep Quality would score at 8.89 which is ranked as "good" on the SQI scale (a score of 7 to 9 indicates good sleep quality on the SQI scale). This relationship is also evident in the Plot 2 above.
The analysis also revealed a main effect of the PSS stress score on sleep quality. That is, for every one unit increase in PSS score, the sleep quality score will decrease by 0.29. The p-value for this variable shows that there is a statistically significant relationship, such that a higher PSS score predicts lower sleep quality. The second main effect of this study seems to indicate that for every one unit increase in Pandemic stress score, the sleep quality would in fact increase by 0.29; however this relationship is not statistically significant. The model fit indicates that the PSS stress and Pandemic stress together account for 27.5% of the variation in sleep. However the interaction between PSS and Pandemic scores is not statistically significant.
# Discussion
The first hypothesis was met in that stress negatively predicts sleep. The data in the study, as is evident in Plot 2 showed a strong relationship. There is strong statistical power and effect size. The relationship between stress and sleep is a well established relationship and therefore the results were not surprising.
The second hypothesis had attempted to look at the interaction role of the stress due to the COVID-19 pandemic, whereby the relationship between stress and sleep is stronger (a steeper negative slope) for participants who report that a greater degree of their stress was related to the pandemic. While the raw data seems to suggest that there was a reverse main effect of the pandemic; that is to say, sleep increased as a result of the pandemic, this relationship was not statistically significant. Recent studies such as that by Katuna and Lau (2020), have shown that the relaxed schedule of the pandemic lockdown have in fact allowed for longer sleep times though there were limits to actual sleep quality due to a lack of sleep buildup from reduced physical exercise.
The interaction of the pandemic on sleep in the current study however was not statistically significant either. This may be due to the nature of the survey question itself, which unlike the PSS and SQI, is not a scale that has statistical validity or reliability.
A future direction would be to explore this idea further with a scale that passed tests of validity and reliability and truly measures the impact of a global ongoing pandemic specific stress on sleep versus localized stress events such as an earthquake or other natural disasters.
