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

This study did not test an intervention in the traditional sense. Instead, it investigated the *associations* between various personal characteristics and lifestyle factors (which can be thought of as potential "risk factors") and the levels of psychological impact experienced by university students during the initial phase of the COVID-19 pandemic and associated lockdowns.

The "risk factors" explored included:

**Sociodemographic factors:** Gender (woman vs. man), ethnicity (non-Hispanic Asian, non-Hispanic White, other), age (18-24 years vs. older), general health status (fair/poor vs. good/excellent), and relative family income (below-average vs. above-average social class).

**Lifestyle-related factors:** Daily time spent outdoors (at least two hours vs. less than two hours) and daily time spent on electronic screens (8 or more hours vs. less than 8 hours).

**COVID-19 awareness:** Knowing someone who had been infected with COVID-19 (yes vs. no).

The primary **outcome measures** were the levels of "psychological impact" experienced by students. These impacts were not measured by specific, pre-defined clinical scales (like for anxiety or depression) but were derived from students' close-ended responses to a web-based questionnaire. These responses were analyzed to identify underlying patterns, resulting in two "latent constructs" representing different facets of psychological distress or impact. Based on these constructs, students were categorized into profiles of high, moderate, or low psychological impact.

The study included a total of **2,534 university students** from **seven different universities across the United States**.

The demographic breakdown of the participants was:

**Gender:** 61% were women.

**Ethnicity:** 79% were non-Hispanic White, and a smaller proportion were non-Hispanic Asian (specific percentage not given in abstract, but mentioned as a group). Other ethnicities made up the remainder.

**Student Status:** 20% were graduate students, implying the majority (80%) were undergraduate students.

**Age:** While not explicitly stated for the full sample, age 18-24 years was identified as a risk factor, suggesting a significant portion of the sample fell within this age range.

**Health Status:** Participants self-reported their general health status, with some reporting fair/poor health.

**Income:** Participants reported their relative family income, categorized as below-average or above-average social class.

The data collection occurred between **mid-March to early-May 2020**, a period when most coronavirus-related sheltering-in-place orders were in effect across the U.S. This means the study captured students' experiences during a specific, acute phase of the pandemic.

The researchers collected data using **web-based questionnaires**. These questionnaires contained a series of "close-ended responses" related to students' experiences and feelings during the pandemic.

To measure **psychological impact**, the researchers did not use standard, named psychological scales (e.g., PHQ-9 for depression or GAD-7 for anxiety). Instead, they used a data-driven approach:

**Exploratory Factor Analysis (EFA):** This statistical technique was applied to the close-ended responses to identify underlying patterns or "latent constructs" that represented different dimensions of psychological impact. The EFA resulted in **two distinct latent constructs**, indicating that students' psychological experiences during the pandemic could be understood through at least two major themes. The abstract does not specify what these two constructs represented (e.g., anxiety, stress, loneliness, academic disruption).

**Latent Profile Analysis (LPA):** After identifying the latent constructs, LPA was used to group students into distinct "profiles" based on their scores on these constructs. This allowed the researchers to categorize students into groups with similar patterns of psychological impact. Three such profiles were identified:

* **High psychological impact:** 45% of the sample.

* **Moderate psychological impact:** 40% of the sample.

* **Low psychological impact:** 14% of the sample.

To measure **risk factors**, the questionnaires included items asking about:

**Sociodemographics:** Self-reported gender, ethnicity, age, general health status (e.g., "fair/poor" vs. "good/excellent"), and relative family income (e.g., "below-average" vs. "above-average social class").

**Lifestyle:** Self-reported daily time spent outdoors (e.g., "at least two hours" vs. "less than two hours") and daily time spent on electronic screens (e.g., "8 or more hours" vs. "less than 8 hours").

**COVID-19 awareness:** A simple yes/no question about whether the student knew someone infected with COVID-19.

The reliance on self-report for all measures means that the data reflects participants' perceptions and memories, which can be subject to recall bias or social desirability bias.

This study employed a **cross-sectional, observational design** using web-based questionnaires.

**How they ran the study:**

**Data Collection:** Data was collected via online surveys distributed to students at seven U.S. universities. The survey period was from mid-March to early-May 2020, coinciding with the initial widespread implementation of COVID-19 shelter-in-place orders.

**Sampling:** The researchers used a combination of **representative and convenience sampling**. Representative sampling aims to ensure the sample reflects the broader population, while convenience sampling involves recruiting participants who are readily available. This mixed approach likely means some universities or departments might have used more targeted recruitment (representative) while others relied on broader, less controlled outreach (convenience). The abstract does not specify the exact proportion of each sampling method or how "representative" sampling was achieved in practice.

**Data Analysis:**

* **Exploratory Factor Analysis (EFA):** Used to identify underlying dimensions (latent constructs) within the students' close-ended responses about their psychological experiences. This step helps to reduce a large number of individual survey items into a smaller, more meaningful set of psychological impact themes.

* **Latent Profile Analysis (LPA):** Applied after EFA to group students into distinct profiles based on their scores on the identified latent constructs. This allowed the researchers to classify students into groups experiencing high, moderate, or low overall psychological impact.

* **Bivariate Associations:** Initial analyses examined the relationship between each individual risk factor (e.g., gender, screen time) and the psychological impact profiles, one factor at a time.

* **Multivariate Modeling (Mixed-Effects Logistic Regression):** This advanced statistical technique was used to assess the simultaneous influence of multiple risk factors on the likelihood of experiencing higher levels of psychological impact. "Mixed-effects" models are particularly useful when data comes from different groups (in this case, students from different universities), as they can account for potential clustering or similarities within those groups, which might otherwise inflate statistical significance. Logistic regression is appropriate for predicting a categorical outcome (like high vs. low psychological impact) based on several predictor variables.

**Why this design matters and what it can and cannot prove:**

**Cross-sectional design:** This means data was collected at a single point in time.

* **What it can prove:** It can identify **associations** or **correlations** between risk factors and psychological impact at that specific moment. For example, it can show that students who spent more time on screens *also* tended to report higher psychological impact. It provides a snapshot of the situation during the early pandemic.

* **What it cannot prove:** It **cannot establish causation**. Because all data was collected simultaneously, it's impossible to determine if a risk factor *caused* the psychological impact, or if the psychological impact *led to* changes in the risk factor (e.g., did high screen time cause distress, or did distress lead to more screen time?), or if both were influenced by a third, unmeasured factor. For example, a student might have already been experiencing mental health challenges *before* the pandemic, which then made them more vulnerable to both high screen time and increased psychological impact during lockdown.

**Observational nature:** The researchers observed existing characteristics and behaviors rather than manipulating any variables (like in an experiment).

* **What it can prove:** It's valuable for identifying potential risk factors and generating hypotheses for future, more rigorous studies. It helps understand patterns in real-world populations.

* **What it cannot prove:** It cannot definitively say that changing a risk factor (e.g., reducing screen time) would *cause* a change in psychological impact.

**Sampling methods (representative and convenience):**

* **Why it matters:** While efforts were made for representative sampling, the inclusion of convenience sampling means the sample might not perfectly reflect the entire U.S. university student population. This could limit the generalizability of the findings to all university students. Students who chose to complete an online survey during a pandemic might differ systematically from those who did not.

**Web-based questionnaires and self-report:**

* **Why it matters:** This method is efficient for collecting large amounts of data quickly. However, it relies entirely on participants' honesty, memory, and self-awareness. There's a risk of **self-report bias**, where individuals might underreport negative behaviors