Consistent and robust predictors of Internet Use among older adults over time identified by machine learning

Older adults tend to under-utilize digital technology and online services that can yield substantial benefits to their health and wellbeing. Addressing this problem requires determining robust and consistent predictors of older adults' Internet use over time. Informed by current models of technology use in aging, the present study took a data-driven approach to determine the predictors of Internet use among older adults. Machine learning was applied to a large, nationally representative sample of older Americans with data for hundreds of variables - both before and after the advent of smartphones and tablets. Machine learning models achieved classification accuracy slightly higher than a theory-driven benchmark model, with results largely supporting current models of aging and technology use. Specifically, data from 2002 and 2016 indicated that age, socioeconomics, and cognitive functions that decline with age (immediate memory, delayed memory, and visuospatial skills) were the most robust and consistent predictors of Internet use among older adults. Machine learning also discovered additional factors that should be considered in models of technology use in aging, such as alcohol use. Taken with prior literature, these results suggest that automaticity should be a high design priority to reduce the age-related cognitive challenges that impact technology use.

Automated summary

Older adults tend to under-utilize digital technology and online services that can benefit their health and wellbeing. This study used machine learning to determine the predictors of Internet use among older adults, finding that age, socioeconomics, and cognitive functions were the most robust and consistent predictors. Additional factors, such as alcohol use, were also discovered. The findings suggest that design prioritization should focus on reducing age-related cognitive challenges to improve technology use among older adults.