Psychological impacts from COVID-19 among university studentsRisk factors across seven states in the United States

Matthew H. E. M. Browning, Lincoln R. Larson, Iryna Sharaievska, Alessandro Rigolon, Olivia McAnirlin, Lauren E. Mullenbach, Scott Cloutier, Tue Vu, Jennifer Thomsen, Nathan Reigner, Elizabeth Covelli Metcalf, Ashley D’Antonio, Marco Helbich, Gregory N. Bratman, Hector A. Olvera‐Alvarez•View original

OverviewBalancedalloy voice

Picture a college campus in early 2020. One week, you’re in crowded lecture halls and noisy dining rooms. The next, your dorm is emptied, classes are on Zoom, and every plan you had feels suddenly uncertain. College students already have higher levels of anxiety, mood problems, and stress than the general population. Then COVID-19 accelerated those issues. That's the moment Browning, Larson, Sharaievska, Rigolon, and colleagues stepped into, asking a deceptively simple question: who among students is being hit hardest psychologically, and what makes that more likely? They did something smart. Instead of treating "impact" as a single measure, they built a profile of it. Across seven large U.S. universities, they invited just over fourteen thousand students to an online survey as lockdowns rolled out, from mid-February through mid-May of 2020. Two thousand five hundred thirty-four students completed the survey, with enough data for full modeling on two thousand one hundred forty. The sample skewed female and non-Hispanic White, and about one in five were graduate students. The goal wasn't just to measure distress; it was to map the contours of it and then see which students were more likely to land in the hardest-hit group. The measurement started basic and got rigorous. They began with nine items: feelings like fear and irritability, time spent worrying about the pandemic, and how preoccupied students felt, plus an open-ended prompt about how life had changed. Two researchers coded those open responses and agreed at a strikingly high level, indicating that the themes were clear. On the quantitative side, the team used exploratory factor analysis, a method for discovering the hidden dimensions that organize related survey items. One item, guilt, didn’t behave like the others and was dropped. When the dust settled, two clear dimensions emerged. One captured emotional distress—think afraid, sad, irritable, stressed, and preoccupied. The other captured worry time—how much of your day COVID-19 occupied. The data were strong enough to support that division; sampling adequacy was high, and both dimensions were internally consistent, with reliability in the low to high eighties range. In other words, they weren't measuring noise. Then came the part that makes this more than a checklist. Those two factor scores fed a latent profile analysis, which sorts students into groups based on their pattern across the two dimensions. The model preferred a three-profile solution. One group sat high on both distress and worry, one sat in the middle, and one sat low on both. The split was not what you might hope. Forty-five percent of students landed in the high-impact profile, forty percent were in the moderate group, and fourteen percent were in the low group. If you were in that high group, your days were filled with rumination, and your emotions were running high. The qualitative comments aligned with that picture: lots of worry and stress, a sense of life narrowing to social distancing and screens, and education shifting beneath them. Of course, the next question is the practical one. Who was more likely to be in that high-impact profile? Here the team moved from description to prediction using mixed-effects logistic regression. That’s a mouthful, but the concept is simple: estimate the odds of being in the high-impact group while controlling for which university you attended, because students within the same institution might resemble each other in unseen ways. When they did that, something reassuring emerged—there wasn't much difference between universities. The fraction of psychological impact explained by which school you were at was essentially nil, with an intraclass correlation near zero. Overall, the predictors together accounted for about seven percent of the variance in who was in the high-impact group. That's modest but useful: in mental health, especially amid a once-in-a-century disruption, multiple forces push and pull at the same time. Now for the headline findings. Gender was the clearest predictor. Women were about twice as likely as men to be in the high-impact profile. Age mattered too, but more gently: students aged 18 to 24 had roughly thirty-eight percent higher odds than older peers. General health had a direct connection to impact. Better self-rated health was protective; conversely, students in poorer health had about sixty percent higher odds of landing in the high group. Each of these findings is intuitive, but it's important to have the numbers to support that intuition. Two situational factors also raised risk levels. Knowing someone infected with COVID-19 increased the odds by about forty-five percent. Additionally, spending eight or more hours on screens in the prior day added another modest increase, roughly twenty-two percent. That latter factor may be less about screens themselves and more about confinement: when the world shrinks to rectangles, rumination has space to grow. But note the scale here. These are contributors, not determinants. What about race, ethnicity, and socioeconomic status? In the broad sample, students who identified as non-Hispanic White tended to experience lower impact in simple comparisons, while non-Hispanic Asian students leaned the other way. When you put variables into the model together, those patterns softened. Non-Hispanic Asian status hovered at the edge of statistical significance, and socioeconomic status became clearer upon closer examination. In a sensitivity analysis limited to one university with a representative sample—North Carolina State—students who reported above-average social class were about twenty-three percent less likely to be in the high-impact group. Interestingly, screen time lost its significance in that subsample. This tells you two things at once: some risk signals are consistent across settings, and some are sensitive to how the sample is drawn. Let's pause and translate the modeling into plain language. Mixed-effects regression adds a "random effect" for each university, which acts like a local baseline. If one school is uniquely anxious or uniquely buffered, the model gives it a nudge without assuming we know exactly why. In this case, those nudges were small. The fixed effects—the student-level predictors like gender, age, health, and experience with COVID in your social circle—did the real work. Even then, they explained a small slice of the total picture. That isn't a flaw; it's a reminder that a pandemic’s psychological footprint is vast. Measures that weren't captured here—like substance use or detailed financial strain—may carry additional weight. The qualitative material offers that reminder a human shape. Students described a sudden contraction of social life and movement. They talked about classes that felt different not just in format but in meaning, and about mental energy consumed by worry. Those narratives reflect the two-factor structure: how you feel and how much your day is consumed by those feelings. Put differently, impact is both intensity and duration. It's the weather and the forecast. Method enthusiasts might appreciate a few details under the hood. The team handled about five percent missingness in the quantitative items using a machine-learning approach called bag imputation, which borrows strength from patterns in the data to fill in gaps. They checked for measurement adequacy before running the factor analysis, and the two dimensions—emotional distress and worry time—showed solid statistical foundations. They also sanity-checked collinearity among predictors; none of the variables interfered with each other enough to distort the model. Plenty of studies stop at averages. This one took the extra step of profiling, which is where the policy action lives. Now for the caveats, because they matter. This was a cross-sectional snapshot taken early in the pandemic. It can tell you who was suffering more, but not why in a causal sense. The sample overrepresented non-Hispanic White students and drew on convenience methods at most sites, which limits how far we can generalize. The measures were tailored to this moment rather than standardized mental health instruments, so you should read "emotional distress" and "worry time" as constructs built for this context. The model also left most variance unexplained, which is typical in mental health but easy to overlook when you're staring at tidy bar charts. Even with those limits, the practical message is clear. If nearly half your student body is in a high-impact profile during a crisis, you need two layers of response. One is targeted triage. Women, younger students, and those in poorer health or with infections touching their social circles are at elevated risk; you can prioritize outreach, screening, and services for those groups. The other layer is universal. When distress and worry time rise campus-wide, broad supports—tele-mental health slots, low-friction counseling access, peer support networks, and course accommodations that reduce cognitive load—help everyone. In that North Carolina State subsample, higher social class correlated with lower impact, hinting at the buffering role of resources. Meeting basic needs is not a soft add-on; it's core infrastructure. You can hear the educator's shorthand for this: Maslow before Bloom. If safety, stability, and a little predictability aren't in place, higher-level learning becomes difficult. During a pandemic, that's not just philosophy; it's logistics—emergency grants, food and housing security, and predictable communication. There's also a small but important signal about screens. In the full sample, long screen time days were associated with higher impact; in the representative subsample, that association weakened. The takeaway isn't to demonize screens; it’s to incorporate breaks that encourage students to unplug and get outside when possible. Interestingly, in simple comparisons, more outdoor time appeared helpful, but that didn’t hold when all variables were considered together. So treat nature as a likely ally, not a magical fix. What I appreciate about Browning and colleagues' approach is that it respects complexity without losing sight of it. Two dimensions, three profiles, a handful of predictors that help you spot who needs extra attention, and a clear sense of the limits. It's enough structure to act on, with enough humility to refine as circumstances change. If you're a university leader, the move is to establish a profiling mindset, not just conduct a one-off survey. Keep an eye on those two dials—distress and worry time—over time, and direct resources where they'll decrease both. If you’re a student listening to this, the numbers reflect what you already felt: this was hard, and it hit some people harder. The aim of measuring it wasn't to label you; it was to ensure the help reaches those who need it.

College students already have higher levels of anxiety, mood problems, and stress than the general population. Then COVID-19 accelerated those issues. That's the moment Browning, Larson, Sharaievska, Rigolon, and colleagues stepped into, asking a deceptively simple question: who among students is being hit hardest psychologically, and what makes that more likely?

They did something smart. Instead of treating "impact" as a single measure, they built a profile of it. Across seven large U.S. universities, they invited just over fourteen thousand students to an online survey as lockdowns rolled out, from mid-February through mid-May of 2020.

Two thousand five hundred thirty-four students completed the survey, with enough data for full modeling on two thousand one hundred forty. The sample skewed female and non-Hispanic White, and about one in five were graduate students. The goal wasn't just to measure distress; it was to map the contours of it and then see which students were more likely to land in the hardest-hit group.

The measurement started basic and got rigorous. They began with nine items: feelings like fear and irritability, time spent worrying about the pandemic, and how preoccupied students felt, plus an open-ended prompt about how life had changed. Two researchers coded those open responses and agreed at a strikingly high level, indicating that the themes were clear.

On the quantitative side, the team used exploratory factor analysis, a method for discovering the hidden dimensions that organize related survey items. One item, guilt, didn’t behave like the others and was dropped. When the dust settled, two clear dimensions emerged.

One captured emotional distress—think afraid, sad, irritable, stressed, and preoccupied. The other captured worry time—how much of your day COVID-19 occupied. The data were strong enough to support that division; sampling adequacy was high, and both dimensions were internally consistent, with reliability in the low to high eighties range. In other words, they weren't measuring noise.

Then came the part that makes this more than a checklist. Those two factor scores fed a latent profile analysis, which sorts students into groups based on their pattern across the two dimensions. The model preferred a three-profile solution.

One group sat high on both distress and worry, one sat in the middle, and one sat low on both. The split was not what you might hope. Forty-five percent of students landed in the high-impact profile, forty percent were in the moderate group, and fourteen percent were in the low group.

If you were in that high group, your days were filled with rumination, and your emotions were running high. The qualitative comments aligned with that picture: lots of worry and stress, a sense of life narrowing to social distancing and screens, and education shifting beneath them.

Of course, the next question is the practical one. Who was more likely to be in that high-impact profile? Here the team moved from description to prediction using mixed-effects logistic regression.

That’s a mouthful, but the concept is simple: estimate the odds of being in the high-impact group while controlling for which university you attended, because students within the same institution might resemble each other in unseen ways. When they did that, something reassuring emerged—there wasn't much difference between universities. The fraction of psychological impact explained by which school you were at was essentially nil, with an intraclass correlation near zero.

Overall, the predictors together accounted for about seven percent of the variance in who was in the high-impact group. That's modest but useful: in mental health, especially amid a once-in-a-century disruption, multiple forces push and pull at the same time.

Now for the headline findings. Gender was the clearest predictor. Women were about twice as likely as men to be in the high-impact profile.

Age mattered too, but more gently: students aged 18 to 24 had roughly thirty-eight percent higher odds than older peers. General health had a direct connection to impact. Better self-rated health was protective; conversely, students in poorer health had about sixty percent higher odds of landing in the high group.

Each of these findings is intuitive, but it's important to have the numbers to support that intuition.

Two situational factors also raised risk levels. Knowing someone infected with COVID-19 increased the odds by about forty-five percent. Additionally, spending eight or more hours on screens in the prior day added another modest increase, roughly twenty-two percent.

That latter factor may be less about screens themselves and more about confinement: when the world shrinks to rectangles, rumination has space to grow. But note the scale here. These are contributors, not determinants.

What about race, ethnicity, and socioeconomic status? In the broad sample, students who identified as non-Hispanic White tended to experience lower impact in simple comparisons, while non-Hispanic Asian students leaned the other way. When you put variables into the model together, those patterns softened.

Non-Hispanic Asian status hovered at the edge of statistical significance, and socioeconomic status became clearer upon closer examination. In a sensitivity analysis limited to one university with a representative sample—North Carolina State—students who reported above-average social class were about twenty-three percent less likely to be in the high-impact group. Interestingly, screen time lost its significance in that subsample.

This tells you two things at once: some risk signals are consistent across settings, and some are sensitive to how the sample is drawn.

Let's pause and translate the modeling into plain language. Mixed-effects regression adds a "random effect" for each university, which acts like a local baseline. If one school is uniquely anxious or uniquely buffered, the model gives it a nudge without assuming we know exactly why.

In this case, those nudges were small. The fixed effects—the student-level predictors like gender, age, health, and experience with COVID in your social circle—did the real work. Even then, they explained a small slice of the total picture.

That isn't a flaw; it's a reminder that a pandemic’s psychological footprint is vast. Measures that weren't captured here—like substance use or detailed financial strain—may carry additional weight.

The qualitative material offers that reminder a human shape. Students described a sudden contraction of social life and movement. They talked about classes that felt different not just in format but in meaning, and about mental energy consumed by worry.

Those narratives reflect the two-factor structure: how you feel and how much your day is consumed by those feelings. Put differently, impact is both intensity and duration. It's the weather and the forecast.

Method enthusiasts might appreciate a few details under the hood. The team handled about five percent missingness in the quantitative items using a machine-learning approach called bag imputation, which borrows strength from patterns in the data to fill in gaps. They checked for measurement adequacy before running the factor analysis, and the two dimensions—emotional distress and worry time—showed solid statistical foundations.

They also sanity-checked collinearity among predictors; none of the variables interfered with each other enough to distort the model. Plenty of studies stop at averages. This one took the extra step of profiling, which is where the policy action lives.

Now for the caveats, because they matter. This was a cross-sectional snapshot taken early in the pandemic. It can tell you who was suffering more, but not why in a causal sense.

The sample overrepresented non-Hispanic White students and drew on convenience methods at most sites, which limits how far we can generalize. The measures were tailored to this moment rather than standardized mental health instruments, so you should read "emotional distress" and "worry time" as constructs built for this context. The model also left most variance unexplained, which is typical in mental health but easy to overlook when you're staring at tidy bar charts.

Even with those limits, the practical message is clear. If nearly half your student body is in a high-impact profile during a crisis, you need two layers of response. One is targeted triage.

Women, younger students, and those in poorer health or with infections touching their social circles are at elevated risk; you can prioritize outreach, screening, and services for those groups. The other layer is universal. When distress and worry time rise campus-wide, broad supports—tele-mental health slots, low-friction counseling access, peer support networks, and course accommodations that reduce cognitive load—help everyone.

In that North Carolina State subsample, higher social class correlated with lower impact, hinting at the buffering role of resources. Meeting basic needs is not a soft add-on; it's core infrastructure. You can hear the educator's shorthand for this: Maslow before Bloom.

If safety, stability, and a little predictability aren't in place, higher-level learning becomes difficult. During a pandemic, that's not just philosophy; it's logistics—emergency grants, food and housing security, and predictable communication.

There's also a small but important signal about screens. In the full sample, long screen time days were associated with higher impact; in the representative subsample, that association weakened. The takeaway isn't to demonize screens; it’s to incorporate breaks that encourage students to unplug and get outside when possible.

Interestingly, in simple comparisons, more outdoor time appeared helpful, but that didn’t hold when all variables were considered together. So treat nature as a likely ally, not a magical fix.

What I appreciate about Browning and colleagues' approach is that it respects complexity without losing sight of it. Two dimensions, three profiles, a handful of predictors that help you spot who needs extra attention, and a clear sense of the limits. It's enough structure to act on, with enough humility to refine as circumstances change.

If you're a university leader, the move is to establish a profiling mindset, not just conduct a one-off survey. Keep an eye on those two dials—distress and worry time—over time, and direct resources where they'll decrease both. If you’re a student listening to this, the numbers reflect what you already felt: this was hard, and it hit some people harder.

The aim of measuring it wasn't to label you; it was to ensure the help reaches those who need it.

Related lectures