Table 5. Logistic Regression Predicting AI Provider Choice Versus Human Physician.
| Model 1: Experimental Conditions | Model 2: + Demographics | Model 3: + Trust | ||||
|---|---|---|---|---|---|---|
| Illness Severity | 0.96 | (0.08) | 1.00 | (0.09) | 0.98 | (0.09) |
| Accuracy | 1.51*** | (0.12) | 1.54*** | (0.14) | 1.48*** | (0.13) |
| Incorporation | 0.92 | (0.07) | 0.91 | (0.08) | 0.88 | (0.08) |
| Established | 1.28** | (0.10) | 1.28** | (0.11) | 1.25* | (0.11) |
| Tailored | 1.06 | (0.09) | 1.07 | (0.09) | 1.11 | (0.10) |
| Listens | 1.20* | (0.10) | 1.23* | (0.11) | 1.27** | (0.12) |
| Race Unbiased | 1.08 | (0.09) | 1.07 | (0.09) | 1.10 | (0.10) |
| Financial Unbiased | 1.13 | (0.09) | 1.12 | (0.10) | 1.10 | (0.10) |
| Order | 1.20* | (0.10) | 1.20* | (0.11) | 1.13 | (0.10) |
| Illness Experience | 1.12 | (0.10) | 1.18+ | (0.11) | 1.17 | (0.12) |
| Black | 0.84 | (0.11) | 0.73* | (0.10) | ||
| Hispanic | 1.00 | (0.14) | 0.88 | (0.12) | ||
| Asian | 0.94 | (0.13) | 0.85 | (0.12) | ||
| Native | 1.26 | (0.19) | 1.37* | (0.21) | ||
| Income | 1.02 | (0.02) | 1.02 | (0.02) | ||
| Age | 0.99* | (0.00) | 0.99* | (0.00) | ||
| Conservative | 0.61*** | (0.07) | 0.65*** | (0.08) | ||
| Religion Important | 0.67*** | (0.06) | 0.64*** | (0.06) | ||
| Sex | 1.00 | (0.09) | 0.98 | (0.09) | ||
| Education | 1.07* | (0.04) | 1.10** | (0.04) | ||
| Married | 0.96 | (0.09) | 0.97 | (0.10) | ||
| Employed Full-Time | 1.03 | (0.10) | 1.05 | (0.11) | ||
| Good Health | 1.09 | (0.12) | 0.99 | (0.11) | ||
| Have Provider | 1.12 | (0.13) | 1.08 | (0.14) | ||
| Trust Provider | 0.96 | (0.03) | ||||
| Trust Hospital | 0.95+ | (0.03) | ||||
| Trust AI Companies | 1.29*** | (0.03) | ||||
| N | 2471 | 2199 | 2198 | |||
Notes: Odds Ratios; Standard errors in parentheses
+ p < .10
* p < .05
** p < .01
*** p < .001. For race, Whites are the reference group. Income (16-point scale from less than $10,000 to more than $500,000), age (years), education (6-point scale from less than high school to graduate school), sex (1 = male, 0 = female). Likert-scales are coded as good health (1 = excellent/very good/good), religion viewed as important (1 = very/somewhat important), and conservative (very conservative/conservative). Analysis excludes respondents that failed an attention check about whether respondent could correctly recall the disease presented in the vignette, and those reporting that they did not understand the vignette (bottom two levels on 10-point scale).