Identifying meaningful subpopulation segments among older public assistance recipients: a mixed methods study to develop tailor-made health and welfare interventions.

BACKGROUND

Public assistance recipients have diverse and complex needs for health and social support in addition to financial support. Segmentation, which means dividing the population into subgroups (segments) with similar sociodemographic characteristics, is a useful approach for allocating support resources to the targeted segments. Clustering is a commonly used statistical method of segmentation in a data-driven marketing approach. This explanatory sequential mixed methods study applied a clustering technique, aiming to identify segments among older public assistance recipients quantitatively, and assess the meaningfulness of the identified segments in consultation and support activities for older recipients qualitatively.

METHODS

We identified the segments of older recipients in two municipalities using probabilistic latent semantic analysis, a machine learning-based soft clustering method. Semi-structured interviews were subsequently conducted with caseworkers to ask whether the identified segments could be meaningful for them in practice and to provide a reason if they could not think of any older recipients from the segment.

RESULTS

A total of 3,165 older people on public assistance were included in the analysis. Five distinct segments of older recipients were identified for each sex from 1,483 men and 1,682 women. The qualitative findings suggested most of identified segments reflected older recipients in practice, especially two of them: female Cluster 1 (facility residents aged over 85 years with disability/psychiatric disorder), and female Cluster 2 (workers). Some caseworkers, however, did not recall older recipients in practice when working with certain segments.

CONCLUSIONS

A clustering technique can be useful to identify the meaningful segments among older recipients and can potentially discover previously unrecognized segments that may not emerge through regular consultation practices followed by caseworkers. Future research should investigate whether tailored support interventions for these identified segments are effective.