Predicting participation willingness in ecological momentary assessment of general population health and behavior: Machine learning study

Aja Louise Murray*, Anastasia Ushakova, Xinxin Zhu, Yi Yang, Zhuoni Xiao, Ruth Brown, Lydia Speyer, Denis Ribeaud, Manuel Eisner

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract / Description of output

Ecological momentary assessment (EMA) is widely used in health research to capture individuals’ experiences in the flow of daily life. The majority of EMA studies, however, rely on nonprobability sampling approaches, leaving open the possibility of nonrandom participation concerning the individual characteristics of interest in EMA research. Knowledge of the factors that predict participation in EMA research is required to evaluate this possibility and can also inform optimal recruitment strategies.

This study aimed to examine the extent to which being willing to participate in EMA research is related to respondent characteristics and to identify the most critical predictors of participation.

We leveraged the availability of comprehensive data on a general young adult population pool of potential EMA participants and used and compared logistic regression, classification and regression trees, and random forest approaches to evaluate respondents’ characteristic predictors of willingness to participate in the Decades-to-Minutes EMA study.

In unadjusted logistic regression models, gender, migration background, anxiety, attention deficit hyperactivity disorder symptoms, stress, and prosociality were significant predictors of participation willingness; in logistic regression models, mutually adjusting for all predictors, migration background, tobacco use, and social exclusion were significant predictors. Tree-based approaches also identified migration status, tobacco use, and prosociality as prominent predictors. However, overall, willingness to participate in the Decades-to-Minutes EMA study was only weakly predictable from respondent characteristics. Cross-validation areas under the curve for the best models were only in the range of 0.56 to 0.57.

Results suggest that migration background is the single most promising target for improving EMA participation and sample representativeness; however, more research is needed to improve prediction of participation in EMA studies in health.
Original languageEnglish
Article numbere41412
JournalJournal of medical Internet research
Early online date28 Jul 2023
Publication statusPublished - 2023

Keywords / Materials (for Non-textual outputs)

  • ecological momentary assessment
  • experience sampling
  • sampling
  • machine learning
  • recruitment


Dive into the research topics of 'Predicting participation willingness in ecological momentary assessment of general population health and behavior: Machine learning study'. Together they form a unique fingerprint.

Cite this