Table 3. Shapley decomposition of survey-weighted linear probability models (LPMs)a.
| Variable group | Video vs audio-only, R2 (%) | Video vs no telehealth, R2 (%) | Audio-only vs no telehealth, R2 (%) | Any telehealth vs no telehealth, R2 (%) | ||||
|---|---|---|---|---|---|---|---|---|
| Demographics | 0.0029 (3.65) | 0.0164 (12.61) | 0.0036 (5.13) | 0.0142 (11.87) | ||||
| Socioeconomic status | 0.0188 (23.49) | 0.0154 (11.83) | 0.0049 (7.03) | 0.0095 (7.91) | ||||
| Health status and needs | 0.0057 (7.12) | 0.0153 (11.79) | 0.0148 (21.08) | 0.0186 (15.53) | ||||
| Health system access | 0.0030 (3.80) | 0.0065 (5.00) | 0.0013 (1.83) | 0.0057 (4.76) | ||||
| Geography | 0.0110 (13.78) | 0.0158 (12.15) | 0.0283 (40.49) | 0.0237 (19.71) | ||||
| Urbanicity | 0.0011 (1.34) | 0.0042 (3.25) | 0.0013 (1.91) | 0.0038 (3.19) | ||||
| Psychosocial factors | 0.0016 (1.95) | 0.0104 (7.97) | 0.0011 (1.53) | 0.0088 (7.34) | ||||
| Digital access | 0.0201 (25.07) | 0.0186 (14.31) | 0.0010 (1.36) | 0.0099 (8.23) | ||||
| Health literacy and eHealth | 0.0122 (15.27) | 0.0248 (19.06) | 0.0062 (8.79) | 0.0209 (17.45) | ||||
| Area context | 0.0036 (4.52) | 0.0026 (2.03) | 0.0076 (10.85) | 0.0048 (4.03) | ||||
| Total | 0.08 (100) | 0.13 (100) | 0.07 (100) | 0.12 (100) | ||||
Shapley decomposition apportions the total R2 of each model into the marginal contributions from distinct groups of predictor variables. All models were survey-weighted. Variable groups defined as (1) demographics: gender, marital status, age group, and race or ethnicity (White vs non-White); (2) socioeconomic status: household income, education level, and employment status; (3) health status and needs: disability status, self-rated health, cancer history, and high blood pressure history; (4) health system access: insurance status; (5) geography: US Census division; (6) urbanicity: urban, suburban, and rural; (7) psychosocial factors: Patient-Reported Outcomes Measurement Information System (PROMIS)—”Social Isolation” and “Meaning and Purpose” scores; (8) digital access: device ownership, satisfaction with home internet for health needs, and county-level percent broadband access; (9) health literacy and eHealth: self-reported health numeracy, use of health or wellness apps, and online health information-seeking behavior; and (10) area context: county-level characteristics including total population, poverty rate, unemployment rate, percent without a high school diploma, and percent of population aged ≥65 years.