Creating Persona Skeletons from Imbalanced Datasets - A Case Study using US Older Adults' Health Data

Incorporating health personas for older adults into design processes can help designers accurately represent older adults by evoking empathy, facilitating consideration of health issues and needs, and reducing stereotype reliance. Toward this goal, we create a two-level quantitative methodology for constructing persona skeletons from imbalanced datasets. We demonstrate our methodology by constructing a set of 4 care-management personas for U.S. older adults via filtering and analyzing demographic, behavior risk factor, and chronic health conditions from 170,704 randomly sampled older adults in a national survey with imbalanced coverage (i.e. between unconditional & conditional questions). We obtain 4 cluster centers for unconditional questions through K-means and iteratively dropping irrelevant features. Within each cluster, we analyze selected respondents for conditional questions. We synthesize results into persona narratives and provide a weighting scheme to quantitatively measure each persona's significance. We contribute a robust persona construction methodology, here applied towards representing older adults.