Table 12.

Estimated percentages of those who changed their responses on importance of values between 2 scenarios and, of those, the percentage that ranked the value to be more important in the first question than in the second question (C01 vs C02 or D01 vs D02), with associated 95% CIs and the P value for the test of equal cell proportions^a.

Domain and values			Percentage of those who changed		Percentage ranking the value as more important in C01 (vs C02) or D01 (vs D02) (95% CI)		P valueb		Design effect
Health—C01c vs C02d
	Explanation	34.3		47.6 (42.8-52.4)		.33		1.68
	Speed	34.9		39.5 (35.2-44.1)		<.001		1.52
	Accuracy	25.1		49.5 (43.8-55.2)		.86		1.68
	Human contact	29.9		50.3 (45-55.5)		.92		1.70
	Responsibility	28.3		47.7 (42.5-53)		.40		1.69
	Reducing costs	33		59.2 (54.3-63.9)		<.001		1.66
	Fairness	29.3		53.7 (48.5-58.8)		.16		1.66
Welfare—D01e vs D02f
	Explanation	39.6		63.7 (59.4-67.7)		<.001		1.55
	Speed	32.7		41.8 (37-46.6)		.001		1.66
	Accuracy	26.4		48.4 (43.2-53.7)		.56		1.57
	Human contact	30.7		43.9 (39.1-48.8)		.02		1.64
	Personal tailoring	33.1		43.9 (39.1-48.8)		.01		1.69
	Reducing costs	35.1		58.8 (54.3-63.1)		<.001		1.58
	Fairness	27.1		51.7 (46.3-57.1)		.53		1.70

^aPercentages and CIs adjusted for weighting.

^bAdjusted Pearson F test for equal proportions.

^cC01: machine reads medical test, diagnoses, and recommends treatment.

^dC02: machine triages when you are unwell.

^eD01: machine processes application for unemployment benefits (data sharing required).

^fD02: chatbot advises about carer payments.