What Predicts Psychological Resilience Among Older Adults with Sensory Impairment? a Predictive Modeling Approach

Psychological resilience is increasingly conceptualized as a dynamic adaptive process shaped by interactions between individual and contextual factors rather than a fixed trait. Social ecological frameworks emphasize that resilience emerges from the interplay of modifiable psychosocial resources and environmental conditions. Older adults with sensory impairment, including vision, hearing, or dual impairments, face elevated risks of social isolation and functional decline but remain understudied. This study examines which demographic, psychosocial, and environmental factors best predict psychological resilience among adults aged 50 and older with sensory impairment using a prediction focused analytic framework.

Data were drawn from the 2016 and 2018 waves of the Health and Retirement Study (HRS). The analytic sample was restricted to respondents aged 50 and above with self reported sensory impairment and complete data across all study variables (N = 3,459). Psychological resilience was measured using the HRS Simplified Resilience Score. Predictors were organized into three domains: demographic characteristics including age, gender, education, marital status, and employment; modifiable psychosocial factors including loneliness, depressive symptoms, purpose in life, and life satisfaction; and environmental factors including neighborhood cohesion, physical disorder, and safety.

A predictive modeling approach using multiple linear regression was implemented with an emphasis on generalizability rather than model fit within a single sample. The dataset was randomly partitioned into a training sample of 80 percent and a hold out sample of 20 percent. Within the training sample, ten fold cross validation was conducted. In each fold, stepwise variable selection was performed independently to identify predictors that optimized out of sample prediction. Predicted values were generated for each validation fold, and predictive performance was quantified using correlations between observed and predicted resilience scores. Mean cross validated correlations were used to estimate model performance.

Three nested models were evaluated to assess incremental predictive contribution. Model 1 included demographic variables only. Model 2 added psychosocial factors. Model 3 further incorporated environmental variables. After cross validation, a final model was estimated using the full training sample with stepwise selection applied to determine the final set of predictors. This model was then evaluated in the hold out sample to assess out of sample performance. Diagnostic checks indicated that model assumptions were largely satisfied.

Results showed that Model 1 demonstrated weak predictive performance with cross validated r equal to .22 and hold out r equal to .26. Model performance improved substantially with the inclusion of psychosocial variables, with Model 2 achieving cross validated r equal to .74 and hold out r equal to .78. The addition of environmental variables did not improve predictive performance. In the final model, purpose in life and life satisfaction were strong positive predictors, while loneliness and depressive symptoms were strong negative predictors.

These findings demonstrate that prediction of psychological resilience in later life is driven primarily by modifiable psychosocial factors. Methodologically, this study shows the value of cross validated predictive modeling for identifying robust predictors and avoiding overfitting. Future research should extend this framework to multilevel modeling and longitudinal prediction.