top of page
Life_CMU the study.jpeg


We followed cohorts of first and second year undergraduates at Carnegie Mellon University over the entire Spring semester. We are currently measuring the student experience during COVID-19. We have published some initial papers (1) (2) on this work and have more scientific papers on the way. We are interested in finding ways to communicate this work to broader audiences (please contact David Creswell).


In both student samples we observed a significant increase from the beginning to the end of the semester, such that there was significantly more depressive symptomatology during final exams (in above figure, blue bar is beginning of term survey, and gold bar is end of term survey). We see similar levels of depressive symptomatology when we compare our samples to other college student samples (e.g., Acharya et al., 2018). Overall, our students ( like other college students) are showing high rates of depressive symptomatology. The measure of depressive symptomatology we collected is self-report and not to be used for formal diagnosis of major depression, but suggests that about 50% of students by the end of semester are at high risk for major depression (in above figure, A, the dotted line is a cutoff score that some researchers use to indicate high risk).

  • Graph B: In our two studies we have a longitudinal subsample of participants who both participated in the first wave of data collection as first year students and then also participated as second year students (in the second wave of data collection) during Spring semester. While this is a small sample of 80 students, it is interesting to note that while depressive symptomatology goes up from the beginning to end of each term in each year, it does fully recover back to beginning of term levels when students return for Spring semester in their second year.

Using machine learning, our team was able to predict end of semester depressive symptomatology with 85.7% accuracy. The features included mobile sensing data and Fitbit actigraphy data. We are currently working on several publications focusing on predicting end of semester depressive symptomatology.

Center for Epidemiologic Studies Depression Scale (CES-D) [Radloff, L. S., 1977] Scale ranges from 0-60; higher scores indicates higher levels of depressive symptomatology. scores above 16 indicate elevated risk for major depression. sample item: “I thought my life had been a failure”


We tracked nightly sleep using both sleep diaries (over 21 days) and using fitbits (over the entire semester). Above is a figure from our second wave of data describing the average weekly sleep duration each week of the semester (fitbit is blue line, and self reported sleep diaries is orange line). On average students are sleeping about 7 hours per night. Students are accumulating some sleep debt in that we see spikes in sleep recovery during the Spring Break week. Below we plot the range of when students are going to bed (from fitbit data) during the week (left figure) and when they are waking up (right figure). Average bedtime and waketime are later in our samples relative to other college student samples, the average weekly bedtime is 2:23am and the average weekly waketime is 9:30am. When we look at these measures on the weekend they shift about 30-60 minutes later (so now closer to a 3am bedtime). We have new research showing how fitbit sleep variables predict end of semester academic success and mental health, keep an eye out for forthcoming publications.


Like depressive symptomatology, stress levels go up over the course of the semester. The stress levels in our samples look comparable to other college student samples (e.g., Orucu & Demir, 2008). Our collaborators have been collecting parallel measures in University of Washington first year students and when we compare the same stress measures in our samples to theirs (see below), we see similar results (note that University of Washington has a quarter system so they measured stress levels at the beginning and end of both winter and spring quarters (hence three bars).

  • In contrast to depressive symptomatology, we do not see the same recovery of stress levels back to pre-term first year baseline as they come back for Spring semester in their second year (notice mean stress score in the Second-Year Baseline).

Comparison Study: Örücü & Demir (2008) 508 participants; mean age of 18.57 and PSS of 18.89


Students in our samples show high levels of happiness and life satisfaction, and these levels look similar to other published college student samples (e.g., Pavot & Diener, 2009). We see the same pattern when we look at other measures such as purpose in life or gratitude. It is interesting to note that college students in our samples are showing both high stress and depression but also high levels of life satisfaction relative to other age groups.

Comparison Study: Pavot & Diener (2009), College student sample with scores ranging 23-25

Life_CMU Happiness Life Satisfaction.jpeg


Our student samples have high levels of loneliness that stay pretty constant at the beginning and end of semester. There is not much population prevalence research on loneliness right now to make strong inferences about whether our students have high or risky levels of loneliness. From initial reports, it seems that our samples have comparable levels of loneliness relative to other college student samples (e.g., Wang et al., 2014), but that college students as a group have higher loneliness relative to all other older age group demographics. More work is needed to better understand the dynamics of young adult loneliness, and it is particularly concerning that there’s such high loneliness among young adults who are living in socially dense dorms on college campuses. We know that loneliness is a major risk factor for mental and physical health problems, for example, we have found that beginning of semester loneliness predicts a greater likelihood of increases in depressive symptomatology in our students (see figure below).


One striking set of findings is that our student samples show lower rates of substance abuse relative to other college students samples, and this is true whether we are looking at binge drinking or marijuana use. This is perhaps one of the most significant differences we see between CMU and other college student samples.

There are much lower rates of binge drinking (drinking 5 or more drinks in a row) across the semester in our student samples relative to nationally representative samples of college students (20% at CMU relative to 33% nationally). The figure above refers to the percentage of students how had at least one binge drinking episode in weeks 1, 7 or 15 of the semester.

Marijuana use (at least once per week) was also low in our CMU student samples relative to nationally  representative samples.

Binge Drinking Comparison Study: Schulenberg et al (2017), 33% of college students report binge drinking over the past two weeks.

Marijuana Use Comparison Study: Schulenberg et al (2017), 21% of college students report marijuana use over the past thirty days.

Life_CMU Substance Abuse.jpeg
bottom of page