A recent article in Biological Conservation suggested that bird singing and chirping patterns were negatively affected by poor air quality and smoke exposure during the 2023 Canadian wildfires. 

For researchers at UM-Dearborn, this finding begged the question — if birds are impacted by poor air quality, what about humans?

Faculty in the social sciences and in computer science set out last fall to explore answers to this question, creating a study that examined air quality, human well-being and the effect poor air quality has on happiness.

“We want people to feel happy. We want people to be productive so they can accomplish their goals and contribute in a way they find to be meaningful,” said Natalia Czap, professor of economics and Department of Social Sciences chair. “Air quality is extremely important in that.” 

Czap is the principal investigator on the UM-Dearborn study. So far, “Air Quality and Human Wellbeing: Assessing Emotional Impact of Lower Air Quality Using Autonomous Artificial Intelligence-Based Distributed Sensing Systems” has resulted in two outcomes: strengthening the research connection between air quality and mood and developing an algorithm that resulted in high participant engagement with self-reporting studies.

The research was funded by a U-M Bold Challenges Boost grant.

More than 120 participants took part in the study during the fall semester. They used portable Atmotube Pro air quality sensors and tracked their levels of happiness four times a day for three weeks.

To get optimal research insight, Czap and her long-term collaborator and husband, Hans Czap, associate professor of economics, connected with Zheng Song, assistant professor of computer and information science, and Qiang Zhu, William E. Stirton Professor and professor of computer and information science, to develop a data collection system.

“All researchers want to have high-quality data. But it can be challenging to motivate participants because they have other things to focus on in daily life,” Natalia Czap said.

“While trying to establish a connection between the air quality and the way participants were feeling using ecological momentary assessment, we expected a response rate of about 75% based on our research of the literature. We far exceeded that.” 

Ecological momentary assessment is a research method that captures real-time data on behaviors and moods within a participant’s natural environment using smartphones or wearable devices.

The average response rate for most participants during the study was above 90%. 

Previously published studies measured the impact on an aggregate level, but the research team wanted to look closer and answer questions, including: Are two people in the same city, but in different neighborhoods, affected differently? Does it matter how close you are to a factory? Are you in a better situation if your neighborhood has many trees?

Song and Zhu worked with computer science graduate students to customize one of their newly developed explore and exploit, or E&E, algorithms for the project. Natalia Czap clarified that “exploit” is meant in a scientific machine-learning way.

“Think of it like choosing a place to eat. ‘Explore’ is like trying new restaurants to see which one is best. ‘Exploit’ means you are going to your favorite restaurant because you see it as the best one,” said graduate student Shashank Chauhan.

“Based on participant behavior, the E&E algorithm helped us learn the best time to prompt people to enter their data.”

In the study, each research volunteer received a text message four times a day — morning, early afternoon, late afternoon and evening — through a UM-Dearborn-created smart survey system. The survey prompted them to answer questions related to their well-being and how sunny it is outside. 

Accounting for sunshine is important because it is a proven determinant of a person’s mood. The responses to the questions were combined with objective data collected from the air quality monitors that users carried with them.

During the first week, the SMS messages were sent to everyone at similar time intervals with randomly selected times. After that first week, the E&E algorithm began detecting patterns in individual response times and tailored future text prompts to more opportune times for each participant.

“If a morning SMS was sent and it took a user a very long time to respond during that first week, maybe they are not a morning person. The E&E algorithm detects this. We kept learning to find better times for each person from the previous week’s response times,” Song said. 

The prompts were also tied to air quality numbers on each person’s air quality sensor. For example, if someone was in a low air quality environment, an SMS would be sent. Researchers also used a gamification component — an activity leaderboard — to further encourage participation.

The UM-Dearborn researchers began analyzing data in January. The research team noted higher response rates when there were interventions like the algorithm or gamification versus a control. Initial results also showed a negative correlation between air quality and wellbeing, with lower happiness linked to higher levels of PM2.5, which are small inhalable particles that come from sources like vehicle exhaust, factories and wildfires. 

While the project’s preliminary results are not a complete surprise, they highlight the importance of addressing air quality to help create change. Natalia Czap said adding their findings to the existing literature showing negative effects on humans provides further evidence for interventions that help improve public health and cognitive performance. 

“Let’s say there is a high-stakes test and the air quality near you is poor. If you know how you and people around you are likely to be affected, you might wait and do the test on another day,” she said. “Even when something seems obvious because of anecdotal evidence, that isn’t enough. You need data — and high-quality data — to help create policies.” 

This is an abbreviated version of a story that can be found on the UM-Dearborn website.