Predicting weight loss in online weight management programs

Shah, Miti
Journal Title
Journal ISSN
Volume Title

The rates of overweight and obesity have steadily increased in the last two decades and a variety of programs have become available to address the growing need for weight management. Shifts in lifestyle have created a growing digital user base forcing even storefront operations to optimize their online offerings. Yet it is not well known what social and psychological factors predict weight loss in online programs. Additionally, there is limited research with large, real-world data to provide an understanding of general trends related to weight loss. In collaboration with a well-known weight management company, the current project aimed to explore the cognitive, emotional, and social aspects of weight loss in a large online weight loss app. Using computerized text analysis methods (e.g., LIWC, topic modeling, and language style matching), the study analyzed the language of ~64,000 users over 12 weeks (relying on over 16 million posts and comments). Users’ social engagement behaviors on the app like making posts, the number of posts, comments and reactions made, average days of food logging per week were also assessed. The findings revealed that narrative thinking styles and the use of language indicative of actively working through problems were associated with more weight loss. Discussing specific topics related to food, weight and fitness and seeking information were also associated with weight loss. Specific types of engagement such as a higher degree of language style matching and averagely weekly food logging were also associated with more weight loss. The relationship between emotions and weight loss was mixed. Feelings of affiliation with the community and themes related to one’s social life and encouragement were not predictive of weight loss success. Engagement in the form of making posts or the number of posts and reactions was also not correlated with weight loss. A hierarchical forced-entry multiple regression revealed high predictive value of engagement behaviors and modest predictive value of language and topic modeling predictors on users’ weight loss. Possible explanations of the findings and applications are discussed.