Study 1: every table and figure

Each analysis from the dissertation, rebuilt from the released data, with a figure beside every table. Hover over any mark for the details.

Deep dive: every table and figure

The rest of this page rebuilds each analysis from the public data. Hover over any mark for the details, and open the guide below if a term is unfamiliar.

Plain-language guide to the terms

Trust: How willing someone is to follow the advice. In these studies, people rated this from 0 (not at all) to 10 (completely).
Quantitative reasoning: Deciding with numbers, data and statistics, like reading a lab result.
Intuition: Deciding with experience and judgment, like a doctor sensing that something is off when the tests look fine.
Effect size (Cohen's d): How big a difference is, on a scale that can be compared across studies. This site uses the Funder and Ozer (2019) labels for psychology: about 0.20 is small, 0.41 is medium, 0.63 is large, and 0.87 or more is very large.
95% interval: The range of values the true average probably falls in. A narrow range means a precise estimate; a range that includes zero means the difference may not be real.
p value: How surprising the result would be if there were no real difference. A p below .05 is the usual bar for calling a result reliable, and "p < .001" means chance is a very unlikely explanation.
Mixed model: A way to analyze data when each person gives many ratings, so that people who rate everything high or low do not distort the results.
Mediation: A test of whether one thing explains part of the link between two others. Here: does seeing a doctor as more competent explain part of why people trust the doctor more?
Preregistered: The plan for the study and its analysis was written and posted publicly before anyone looked at the data.

How to read this page. The seven preregistered tests (H1 to H7) and one exploratory test (E) match the dissertation. Everything else was run in my exploratory analysis notebook and is recomputed here from the public file. The public file groups demographics into broad categories, so the sample table is coarser than the one in the dissertation.

Who took part

The study was preregistered with a target of 186 people recruited through CloudResearch's Prime Panels. After removing preview responses and anyone who did not consent, 188 remain.

Table 1. The people in the released file

What to look for. The share in each group. Ages are spread fairly evenly across the four bands, and women are the larger group.

Group	People	Share
Age: 18-29	55	29.3%
Age: 30-44	51	27.1%
Age: 45-59	42	22.3%
Age: 60+	39	20.7%
Gender: Woman	113	60.1%
Gender: Man	74	39.4%
Race: White	122	64.9%
Race: Black	44	23.4%
Race: Another or multiple	19	10.1%
Education: No college degree	116	61.7%
Education: College degree or more	72	38.3%
Political views: Moderate	74	39.4%
Political views: Liberal	58	30.9%
Political views: Conservative	56	29.8%
Experience with AI: None or a little	119	63.3%
Experience with AI: Moderate or more	69	36.7%

Shares are of all 188 people. Left blank in the file: age (1), gender (1), race (3). Gender had only woman and man responses, so smaller groups do not appear. Race is White or Black only when that is the only group chosen; every other answer, including choosing more than one, is combined because no other group reached 30 people.

Table 1 as a figure

Each bar is the share of the people in that group. Longer means more of the sample. Hover a bar for the head count.

Fifteen medical decisions

For each decision, people slid from 0 (a professional needs only intuition and personal experience) to 100 (needs only quantitative reasoning). Every decision fell above the middle except general wellbeing advice, which sat just below it.

Table 2. Quantitative reasoning needed, by decision (0 to 100)

What to look for. Every decision sits above the middle of the 0 to 100 scale except general wellbeing advice, which sits just below it.

Measure	n	Mean	SD	Median	Max	Skew	Kurtosis	SE
Diagnose diabetes	186	65.04	25.76	69.16	100.00	−0.60	−0.34	1.89
Prescribe insulin and a diet change	186	64.72	27.85	69.68	100.00	−0.63	−0.56	2.04
Recommend knee surgery	186	64.31	26.13	65.84	100.00	−0.50	−0.52	1.92
Diagnose a broken bone	187	63.63	25.64	63.26	100.00	−0.41	−0.37	1.88
Prescribe antibiotics	186	62.35	27.00	62.97	100.00	−0.36	−0.80	1.98
Prescribe pain medication	186	60.72	25.47	58.00	100.00	−0.24	−0.52	1.87
Medical decisions in general	186	60.07	24.13	58.12	100.00	−0.32	−0.32	1.77
Prescribe antidepressants and therapy	186	59.58	26.02	60.48	100.00	−0.27	−0.81	1.91
Diagnose strep throat	186	59.08	26.28	57.30	100.00	−0.20	−0.77	1.93
Diagnose depression	185	57.63	25.47	56.89	100.00	−0.23	−0.62	1.87
Recommend stretching	187	54.49	24.77	53.06	100.00	−0.24	−0.45	1.81
Recommend a strict diet	186	52.27	26.83	51.70	100.00	−0.10	−0.86	1.97
Recommend vitamin C and zinc	186	50.73	27.34	50.75	100.00	−0.03	−1.01	2.00
Recommend meditation	186	50.52	27.79	51.14	100.00	−0.03	−0.92	2.04
Recommend general wellbeing habits	186	49.03	26.74	49.41	100.00	0.10	−0.84	1.96

Sorted from most to least quantitative reasoning needed. Skew and kurtosis are the estimates psych::describe reports.

Figure 1. Every decision, every person

Each row is one of the fifteen decisions. Every dot is one person's answer on a 0 to 100 slider, from needs only intuition (0) to needs only quantitative reasoning (100). The curve shows where answers pile up.

Outline: how answers are spread (smoothed). Dots: one person each. Diamond and bar: the average and its bootstrapped 95% interval.

View as a table

Measure	n	Mean	95% interval
Diagnose diabetes	186	65.0	61.3 to 68.7
Prescribe insulin and a diet change	186	64.7	60.8 to 68.6
Recommend knee surgery	186	64.3	60.6 to 68.1
Diagnose a broken bone	187	63.6	59.8 to 67.2
Prescribe antibiotics	186	62.4	58.6 to 66.1
Prescribe pain medication	186	60.7	57.0 to 64.4
Medical decisions in general	186	60.1	56.5 to 63.5
Prescribe antidepressants and therapy	186	59.6	55.7 to 63.4
Diagnose strep throat	186	59.1	55.3 to 62.6
Diagnose depression	185	57.6	54.2 to 61.1
Recommend stretching	187	54.5	51.0 to 58.1
Recommend a strict diet	186	52.3	48.5 to 56.2
Recommend vitamin C and zinc	186	50.7	46.8 to 54.6
Recommend meditation	186	50.5	46.6 to 54.4
Recommend general wellbeing habits	186	49.0	45.1 to 52.9

Table 3. The decision groups used in the tests (0 to 100)

What to look for. The same ratings averaged into the groups used later. Compare the means across rows; preventative decisions need the least quantitative reasoning.

Measure	n	Mean	SD	Median	Min	Max	Skew	Kurtosis	SE
Diagnostic (4 decisions)	185	61.41	18.91	62.57	0.00	100.00	−0.45	0.49	1.39
Prescriptive (4 decisions)	186	61.84	20.21	63.06	0.00	100.00	−0.55	0.45	1.48
Preventative (4 decisions)	186	51.98	20.44	52.17	2.15	100.00	−0.12	−0.36	1.50
Physical health (9 decisions)	186	59.25	16.97	60.70	4.63	100.00	−0.47	0.80	1.24
Mental health (3 decisions)	185	55.84	19.78	55.50	0.00	100.00	−0.06	0.15	1.45
Bodily injury (3 decisions)	186	59.69	17.65	59.08	1.71	100.00	−0.05	0.42	1.29
Acute sickness (3 decisions)	186	57.39	20.46	58.76	0.00	100.00	−0.44	0.18	1.50
Chronic sickness (3 decisions)	186	60.68	20.77	62.54	0.00	100.00	−0.61	0.26	1.52

Each group is the average of its decisions. A person missing any one decision in a group is missing for that group, so n runs from 185 to 186.

Figure 2. By kind of health problem

The same slider ratings, grouped by kind of health problem. Each dot is one person; the curve shows where answers cluster.

Outline: how answers are spread (smoothed). Dots: one person each. Diamond and bar: the average and its bootstrapped 95% interval.

View as a table

Measure	n	Mean	95% interval
Chronic sickness	186	60.7	57.6 to 63.5
Bodily injury	186	59.7	57.3 to 62.3
Physical health	186	59.3	56.8 to 61.6
Acute sickness	186	57.4	54.4 to 60.3
Mental health	185	55.8	53.2 to 58.6

Figure 3. By stage of care

The same slider ratings, grouped by stage of care: preventing, diagnosing or prescribing.

Outline: how answers are spread (smoothed). Dots: one person each. Diamond and bar: the average and its bootstrapped 95% interval.

View as a table

Measure	n	Mean	95% interval
Prescriptive	186	61.8	58.8 to 64.6
Diagnostic	185	61.4	58.8 to 64.0
Preventative	186	52.0	49.0 to 54.7

Do the decision ratings hang together?

Before averaging decisions into groups, I checked how closely the items in each group move together. Alpha runs from 0 to 1; higher means people answered the items in a group alike. The bodily injury group is the weakest (alpha .47), so results that rest on it deserve more caution.

Table 4. Reliability of each group of decisions

What to look for. Alpha near or above .80 means the items in a group can safely be averaged. The bodily injury group is the weakest, so lean on it less.

Group	Items	Cronbach’s alpha	Average item correlation
Diagnostic decisions	4	.71	.38
Prescriptive decisions	4	.76	.44
Preventative decisions	4	.76	.45
Physical health decisions	9	.82	.34
Mental health decisions	3	.61	.34
Bodily injury decisions	3	.47	.23
Acute sickness decisions	3	.64	.37
Chronic sickness decisions	3	.67	.40

Table 4 as a figure

Each dot is the alpha for one group of decisions: how alike people answered the items inside it. The dashed lines mark .70, .80 and .90; above .70 is usually taken as acceptable and above .80 as good.

Item correlations behind each alpha

Diagnostic decisions

Item	1	2	3	4
1. Diagnose a broken bone		.265	.521	.520
2. Diagnose depression	.265		.170	.281
3. Diagnose strep throat	.521	.170		.548
4. Diagnose diabetes	.520	.281	.548

Prescriptive decisions

Item	1	2	3	4
1. Prescribe pain medication		.475	.429	.401
2. Prescribe antidepressants and therapy	.475		.423	.365
3. Prescribe antibiotics	.429	.423		.529
4. Prescribe insulin and a diet change	.401	.365	.529

Preventative decisions

Item	1	2	3	4
1. Recommend stretching		.466	.408	.429
2. Recommend meditation	.466		.473	.489
3. Recommend vitamin C and zinc	.408	.473		.418
4. Recommend a strict diet	.429	.489	.418

Physical health decisions

Item	1	2	3	4	5	6	7	8	9
1. Diagnose a broken bone		.521	.520	.125	.240	.250	.416	.580	.477
2. Diagnose strep throat	.521		.548	.184	.157	.271	.331	.706	.460
3. Diagnose diabetes	.520	.548		.019	.103	.226	.455	.542	.650
4. Recommend stretching	.125	.184	.019		.408	.429	.151	.084	.107
5. Recommend vitamin C and zinc	.240	.157	.103	.408		.418	.227	.253	.226
6. Recommend a strict diet	.250	.271	.226	.429	.418		.265	.274	.321
7. Prescribe pain medication	.416	.331	.455	.151	.227	.265		.429	.401
8. Prescribe antibiotics	.580	.706	.542	.084	.253	.274	.429		.529
9. Prescribe insulin and a diet change	.477	.460	.650	.107	.226	.321	.401	.529

Mental health decisions

Item	1	2	3
1. Diagnose depression		.219	.588
2. Recommend meditation	.219		.229
3. Prescribe antidepressants and therapy	.588	.229

Bodily injury decisions

Item	1	2	3
1. Diagnose a broken bone		.125	.416
2. Recommend stretching	.125		.151
3. Prescribe pain medication	.416	.151

Acute sickness decisions

Item	1	2	3
1. Diagnose strep throat		.157	.706
2. Recommend vitamin C and zinc	.157		.253
3. Prescribe antibiotics	.706	.253

Chronic sickness decisions

Item	1	2	3
1. Diagnose diabetes		.226	.650
2. Recommend a strict diet	.226		.321
3. Prescribe insulin and a diet change	.650	.321

Figure 4. How many factors?

A scree plot lists the strength of each possible factor in the twelve decisions that make up the groups (185 people answered all twelve). Three factors stand above 1, the usual cut-off, and a fourth falls short.

View as a table

Factor	Eigenvalue
1	4.55
2	1.99
3	1.19
4	0.68
5	0.64
6	0.60
7	0.59
8	0.48
9	0.46
10	0.31
11	0.28
12	0.24

Table 5. A three-factor solution

Three factors, allowed to correlate, sort the twelve decisions by what they ask of a clinician. The names are my reading of the pattern, not part of the analysis. Pain medication for a knee injury is the one decision that splits across factors.

Decision	Factor 1Diagnosing and treating physical illness	Factor 2Preventing illness and promoting health	Factor 3Treating mental health	Shared
Prescribe antibiotics	0.82	0.01	−0.02	0.67
Diagnose strep throat	0.81	0.05	−0.12	0.67
Diagnose diabetes	0.69	−0.10	0.19	0.53
Prescribe insulin and a diet change	0.67	0.02	0.06	0.45
Diagnose a broken bone	0.62	0.03	0.13	0.41
Prescribe pain medication	0.35	0.11	0.32	0.23
Recommend meditation	−0.13	0.73	0.12	0.56
Recommend stretching	0.03	0.70	−0.13	0.51
Recommend a strict diet	0.21	0.61	−0.02	0.42
Recommend vitamin C and zinc	0.05	0.59	0.12	0.37
Prescribe antidepressants and therapy	0.06	−0.03	0.92	0.85
Diagnose depression	−0.07	0.19	0.61	0.41
Variance explained	24%	15%	12%	51%

Factors correlate 0.24 (1 and 2), 0.47 (1 and 3) and 0.23 (2 and 3). Bold loadings are 0.30 or larger.

Three factors, minimum residual extraction and oblimin rotation, on the 12 decisions that form the composites (everyone with all 12 answers). The notebook asked for minchi extraction, which gives the same answer when every correlation rests on the same number of people. Loadings are the pattern matrix.

Figure 4b. The factor diagram

The same solution as a diagram, as the notebook’s factor diagram drew it. Each decision points to the factor it loads on most (solid line); dashed lines are smaller loadings of .30 or more. The curves give the correlations between factors.

The tests

The preregistered plan was a series of paired t tests, since each person answered every item. H1 to H5 compare kinds of decision; H6 and H7 compare a doctor with an AI. The AI here is the general medical AI. Test E asks the same of quantitative reasoning and was not preregistered.

Table 6. Paired t tests

What to look for. The difference column and p. H1 shows no gap, H2 and H3 show a clear gap, and H6, H7 and E show the doctor rated well above the AI.

Test	Comparison	Means (SD)	Difference [95% CI]	t (df)	p	d [95% CI]
H1	Diagnostic vs prescriptive	61.41 (18.91), 61.79 (20.25)	−0.37 [-2.16, 1.41]	−0.41 (184)	.680	−0.03 [-0.17, 0.11]
H2	Diagnostic vs preventative	61.41 (18.91), 51.90 (20.47)	9.51 [6.15, 12.87]	5.58 (184)	< .001	0.41 [0.26, 0.56]
H3	Prescriptive vs preventative	61.84 (20.21), 51.98 (20.44)	9.86 [6.42, 13.30]	5.66 (185)	< .001	0.42 [0.26, 0.56]
H4	Physical vs mental health	59.21 (17.01), 55.84 (19.78)	3.37 [0.80, 5.94]	2.59 (184)	.010	0.19 [0.04, 0.34]
H5	Bodily injury vs acute sickness	59.69 (17.65), 57.39 (20.46)	2.31 [-0.10, 4.72]	1.89 (185)	.061	0.14 [-0.01, 0.28]
H6	Doctor vs AI: intuition	6.73 (2.93), 4.51 (3.04)	2.22 [1.67, 2.78]	7.92 (187)	< .001	0.58 [0.42, 0.73]
H7	Doctor vs AI: trust	7.50 (2.55), 4.69 (3.06)	2.81 [2.29, 3.33]	10.73 (184)	< .001	0.79 [0.62, 0.95]
E	Doctor vs AI: quantitative reasoning (exploratory)	7.22 (2.95), 5.20 (3.18)	2.02 [1.50, 2.54]	7.63 (187)	< .001	0.56 [0.40, 0.71]

H1 to H5 compare how much quantitative reasoning decisions need (0 to 100); H6, H7 and E compare ratings of a doctor and an AI (0 to 10). A positive difference means the first item was rated higher. Tests are two-sided at alpha = .05, the criterion in the preregistration.

Figure 5. Person as effect size

An average can hide how many people actually differ. For each comparison, this counts people who rated the first item higher, the same, or lower. Even for the largest gap, trust in a doctor over an AI, 14% of people trusted the AI more and 19% rated the two the same. The notebook counted this for trust (doctor versus AI, and layperson versus AI); I ran the same count for every test.

First item rated higher
Rated the same
Second item rated higher
Doctor is blue, AI is coral, wherever they appear

Table 7. Follow-up comparisons from the notebook

What to look for. Which comparisons show a clear gap and which do not: look at whether the interval for the difference stays away from zero.

Test	Comparison	Means (SD)	Difference [95% CI]	t (df)	p	d [95% CI]
X1	Trust: layperson versus AI	4.34 (3.10), 4.67 (3.06)	−0.33 [-0.75, 0.10]	−1.51 (183)	.132	−0.11 [-0.26, 0.03]
X2	Intuition: layperson versus AI	4.92 (3.16), 4.51 (3.04)	0.41 [-0.01, 0.84]	1.93 (187)	.055	0.14 [0.00, 0.28]
X3	AI: quantitative reasoning versus intuition	5.20 (3.18), 4.51 (3.04)	0.69 [0.24, 1.14]	3.05 (187)	.003	0.22 [0.08, 0.37]
X4	Trust: doctor versus doctor using AI	7.49 (2.55), 5.65 (3.07)	1.84 [1.34, 2.34]	7.20 (186)	< .001	0.53 [0.37, 0.68]

Exploratory comparisons the notebook ran after the preregistered tests: a layperson against the AI, the AI's quantitative reasoning against its intuition, and a doctor against a doctor using AI.

Eleven recommenders

Each person rated all eleven recommenders on three things, each from 0 to 10: how much they trust it to make health recommendations, how capable it is of quantitative reasoning, and how capable it is of intuition. A doctor using AI is treated as AI in the models later on, as in the notebook.

Table 8. Ratings of each recommender, mean (SD)

What to look for. Read down the trust column. Doctors and nurses sit near the top, the AI recommenders sit below the human experts, and a doctor using AI sits in between.

Recommender	Trust	Quantitative reasoning	Intuition	n (trust)
Doctor	7.45 (2.60)	7.22 (2.95)	6.73 (2.93)	188
Nurse	6.85 (2.62)	6.44 (2.76)	6.54 (2.70)	188
Doctor, remote visit	6.14 (2.80)	5.95 (2.84)	6.05 (2.90)	187
Therapist	6.03 (2.85)	5.90 (2.78)	6.23 (2.81)	188
Doctor using AI	5.65 (3.07)	5.78 (2.96)	5.16 (3.15)	187
Personal trainer	5.32 (2.71)	5.05 (2.72)	5.40 (2.95)	187
AI, general	4.69 (3.06)	5.20 (3.18)	4.51 (3.04)	185
AI diagnostic tool	4.42 (3.01)	5.19 (3.14)	4.32 (3.01)	184
Layperson	4.34 (3.11)	4.37 (2.97)	4.92 (3.16)	186
AI fitness coach	4.30 (3.07)	4.81 (3.06)	4.04 (2.99)	187
AI chatbot	4.25 (3.12)	4.85 (3.09)	4.16 (3.09)	186

Sorted from most to least trusted.

Full descriptives for each rating

Trust (0 to 10)

Measure	n	Mean	SD	Median	Max	Skew	Kurtosis	SE
Doctor	188	7.45	2.60	8.00	10.00	−0.78	−0.52	0.19
Nurse	188	6.85	2.62	8.00	10.00	−0.55	−0.73	0.19
Doctor, remote visit	187	6.14	2.80	6.00	10.00	−0.41	−0.63	0.21
Therapist	188	6.03	2.85	6.00	10.00	−0.46	−0.53	0.21
Doctor using AI	187	5.65	3.07	6.00	10.00	−0.33	−0.94	0.22
Personal trainer	187	5.32	2.71	5.00	10.00	−0.04	−0.72	0.20
AI, general	185	4.69	3.06	4.00	10.00	0.19	−0.98	0.23
AI diagnostic tool	184	4.42	3.01	4.50	10.00	0.23	−1.00	0.22
Layperson	186	4.34	3.11	4.00	10.00	0.24	−1.02	0.23
AI fitness coach	187	4.30	3.07	4.00	10.00	0.24	−1.02	0.22
AI chatbot	186	4.25	3.12	4.00	10.00	0.25	−1.14	0.23

Quantitative reasoning (0 to 10)

Measure	n	Mean	SD	Median	Max	Skew	Kurtosis	SE
Doctor	188	7.22	2.95	8.00	10.00	−0.78	−0.58	0.22
Nurse	188	6.44	2.76	7.00	10.00	−0.33	−0.90	0.20
Doctor, remote visit	188	5.95	2.84	6.00	10.00	−0.27	−0.92	0.21
Therapist	188	5.90	2.78	6.00	10.00	−0.18	−0.84	0.20
Doctor using AI	188	5.78	2.96	6.00	10.00	−0.24	−0.90	0.22
Personal trainer	188	5.05	2.72	5.00	10.00	0.11	−0.81	0.20
AI, general	188	5.20	3.18	5.00	10.00	0.02	−1.22	0.23
AI diagnostic tool	188	5.19	3.14	5.00	10.00	0.03	−1.06	0.23
Layperson	188	4.37	2.97	4.00	10.00	0.19	−0.95	0.22
AI fitness coach	187	4.81	3.06	5.00	10.00	0.06	−1.07	0.22
AI chatbot	188	4.85	3.09	5.00	10.00	0.03	−1.05	0.23

Intuition (0 to 10)

Measure	n	Mean	SD	Median	Max	Skew	Kurtosis	SE
Doctor	188	6.73	2.93	7.00	10.00	−0.54	−0.82	0.21
Nurse	188	6.54	2.70	6.50	10.00	−0.40	−0.59	0.20
Doctor, remote visit	187	6.05	2.90	6.00	10.00	−0.30	−0.87	0.21
Therapist	187	6.23	2.81	6.00	10.00	−0.33	−0.67	0.21
Doctor using AI	188	5.16	3.15	5.00	10.00	−0.04	−1.12	0.23
Personal trainer	188	5.40	2.95	5.00	10.00	−0.03	−0.94	0.22
AI, general	188	4.51	3.04	4.00	10.00	0.15	−0.97	0.22
AI diagnostic tool	187	4.32	3.01	4.00	10.00	0.26	−0.89	0.22
Layperson	188	4.92	3.16	5.00	10.00	−0.04	−1.18	0.23
AI fitness coach	188	4.04	2.99	4.00	10.00	0.32	−0.87	0.22
AI chatbot	188	4.16	3.09	4.00	10.00	0.27	−0.98	0.23

Figure 6. Trust in each recommender, every person

Each row is a recommender. Every dot is one person's trust rating on a 0 to 10 scale, and the curve shows where ratings cluster. Higher means more trust.

Outline: how answers are spread (smoothed). Dots: one person each. Diamond and bar: the average and its bootstrapped 95% interval.

View as a table

Measure	n	Mean	95% interval
Doctor	188	7.5	7.1 to 7.8
Nurse	188	6.8	6.5 to 7.2
Doctor, remote visit	187	6.1	5.8 to 6.5
Therapist	188	6.0	5.6 to 6.4
Doctor using AI	187	5.7	5.2 to 6.1
Personal trainer	187	5.3	4.9 to 5.7
AI, general	185	4.7	4.3 to 5.1
AI diagnostic tool	184	4.4	4.0 to 4.8
Layperson	186	4.3	3.9 to 4.8
AI fitness coach	187	4.3	3.9 to 4.8
AI chatbot	186	4.3	3.8 to 4.7

Figure 7. Where each recommender sits

Doctors sit highest on both kinds of capability. Every AI is seen as less capable of intuition than of quantitative reasoning, and sits below the doctor on both. A doctor using AI lands between the doctor and the AIs.

Human
Doctor using AI
AI
Bars are 95% confidence intervals

View as a table

Recommender	Type	Intuition	Quantitative reasoning	n
Doctor	Human	6.73 ± 0.42	7.22 ± 0.42	188
Nurse	Human	6.54 ± 0.39	6.44 ± 0.39	188
Doctor, remote visit	Human	6.05 ± 0.42	5.95 ± 0.41	188
Therapist	Human	6.23 ± 0.40	5.90 ± 0.40	188
Personal trainer	Human	5.40 ± 0.42	5.05 ± 0.39	188
Layperson	Human	4.92 ± 0.45	4.37 ± 0.43	188
Doctor using AI	Doctor using AI	5.16 ± 0.45	5.78 ± 0.42	188
AI, general	AI	4.51 ± 0.44	5.20 ± 0.45	188
AI diagnostic tool	AI	4.32 ± 0.43	5.19 ± 0.45	188
AI chatbot	AI	4.16 ± 0.44	4.85 ± 0.44	188
AI fitness coach	AI	4.04 ± 0.43	4.81 ± 0.44	187

What goes with trust

Trust tracks both kinds of perceived capability. These analyses are exploratory, and they show what moves together, not what causes what. First, one recommender at a time.

Table 9. Trust predicted by intuition and quantitative reasoning, for three recommenders

What to look for. The β column. Both intuition and quantitative reasoning go with trust for all three recommenders; for the doctor quantitative reasoning is stronger, and for the AI intuition is.

Recommender and predictor	b	SE	t	p	β
Doctor: Intercept	2.80	0.39	7.21	< .001
Doctor: Intuition	0.26	0.06	4.12	< .001	.30
Doctor: Quantitative reasoning	0.40	0.06	6.32	< .001	.45
AI, general: Intercept	0.85	0.33	2.60	.010
AI, general: Intuition	0.49	0.06	8.20	< .001	.49
AI, general: Quantitative reasoning	0.31	0.06	5.36	< .001	.32
Layperson: Intercept	0.58	0.32	1.83	.069
Layperson: Intuition	0.45	0.07	6.83	< .001	.45
Layperson: Quantitative reasoning	0.36	0.07	5.17	< .001	.34

Table 9 as a figure

Each dot is how strongly a rating goes with trust, in standardized units (β), with a 95% interval. Dashed lines mark a small, medium and large effect on the Funder and Ozer scale. Intervals that stay right of zero mean the rating reliably goes with trust.

Table 10. How well each model fits

What to look for. R² is the share of the variation in trust the two ratings explain. About half, so the ratings matter but are not the whole story.

Recommender	n	R²	Adjusted R²	F (df)	p
Doctor	188	.47	.47	82.57 (2, 185)	< .001
AI, general	185	.51	.50	93.72 (2, 182)	< .001
Layperson	186	.51	.50	94.02 (2, 183)	< .001

Table 10 as a figure

Each bar is the share of the variation in trust that the two ratings explain (R²). A bar at 50% means the ratings account for half of why people trusted one recommender more than another.

Table 11. Correlations among the three ratings

What to look for. Every correlation is positive: people who rate a recommender high on one thing tend to rate it high on the others.

Recommender and pair	r	p	n
Doctor: Trust and Quantitative reasoning	.650	< .001	188
Doctor: Trust and Intuition	.598	< .001	188
Doctor: Quantitative reasoning and Intuition	.667	< .001	188
AI, general: Trust and Quantitative reasoning	.571	< .001	185
AI, general: Trust and Intuition	.655	< .001	185
AI, general: Quantitative reasoning and Intuition	.503	< .001	185
Layperson: Trust and Quantitative reasoning	.617	< .001	186
Layperson: Trust and Intuition	.659	< .001	186
Layperson: Quantitative reasoning and Intuition	.615	< .001	186

Table 11 as a figure

Each square is the correlation between two ratings for one recommender. Darker blue means the two move together more strongly.

Mixed models across all eleven recommenders

Each person rated eleven recommenders, so their ratings are not independent. These models give every person their own baseline level of trust. The AI recommenders are the four AI tools plus a doctor using AI, as in the notebook.

Table 12. The models side by side

What to look for. Marginal R² jumps from .05 to .52 once the two ratings enter. A lower AIC means a better balance of fit and simplicity.

Model	Marginal R²	Conditional R²	Person share	AIC
1. AI versus human only	.05	.44	.41	9691
2. Quantitative reasoning and intuition	.52	.64	.25	8640
3. Add AI versus human	.51	.64	.26	8610
4. Let the slopes differ for AI	.51	.64	.26	8594
5. Between and within people	.52	.64	.25	8642
6. Between and within people, plus AI	.53	.65	.26	8609
7. AI recommenders only	.43	.68	.44	3855
8. Human recommenders only	.51	.64	.25	4708

Random intercept for each person, restricted maximum likelihood, 188 people with 11 ratings each. Tests are Wald z tests. Standardized betas are b times the SD of the predictor over the SD of trust. Model comparisons refit by maximum likelihood. The notebook's last model (3D) uses grand-mean variables the notebook never creates, so it is not reproduced here. A doctor using AI counts as AI, as in the notebook.

Table 12 as a figure. Each bar is the share of the variation in trust a model explains: first by its predictors alone, then with each person's own baseline added. Longer is a better fit.

Coefficients for each model

Model 1. AI versus human only

Predictor	b	SE	95% CI	z	p	β
Intercept	6.02	0.16	5.71 to 6.33	38.20	< .001
AI recommender (vs human)	−1.37	0.10	−1.58 to -1.17	−13.31	< .001	−.22

2,051 ratings from 188 people. Person share .41.

Model 2. Quantitative reasoning and intuition

Predictor	b	SE	95% CI	z	p	β
Intercept	1.03	0.14	0.76 to 1.31	7.48	< .001
Quantitative reasoning	0.37	0.02	0.34 to 0.41	19.63	< .001	.37
Intuition	0.44	0.02	0.40 to 0.47	23.92	< .001	.44

2,051 ratings from 188 people. Person share .25.

Model 3. Add AI versus human

Predictor	b	SE	95% CI	z	p	β
Intercept	1.42	0.15	1.12 to 1.72	9.23	< .001
Quantitative reasoning	0.37	0.02	0.34 to 0.41	19.79	< .001	.37
Intuition	0.41	0.02	0.37 to 0.44	21.48	< .001	.41
AI recommender (vs human)	−0.49	0.09	−0.66 to -0.32	−5.71	< .001	−.08

2,051 ratings from 188 people. Person share .26.

Model 4. Let the slopes differ for AI

Predictor	b	SE	95% CI	z	p	β
Intercept	1.45	0.17	1.11 to 1.79	8.38	< .001
Quantitative reasoning	0.44	0.03	0.39 to 0.49	17.70	< .001	.43
AI recommender (vs human)	−0.43	0.19	−0.79 to -0.06	−2.26	.024	−.07
Quantitative reasoning × AI	−0.14	0.04	−0.21 to -0.07	−4.10	< .001
Intuition	0.33	0.03	0.28 to 0.39	13.00	< .001	.34
Intuition × AI	0.14	0.04	0.07 to 0.20	3.89	< .001

2,051 ratings from 188 people. Person share .26.

Model 5. Between and within people

Predictor	b	SE	95% CI	z	p	β
Intercept	0.92	0.26	0.41 to 1.42	3.56	< .001
Quantitative reasoning, person average	0.31	0.07	0.18 to 0.43	4.66	< .001	.21
Quantitative reasoning, vs own average	0.38	0.02	0.34 to 0.41	18.63	< .001	.27
Intuition, person average	0.53	0.07	0.40 to 0.66	8.06	< .001	.36
Intuition, vs own average	0.43	0.02	0.39 to 0.47	22.29	< .001	.32

2,051 ratings from 188 people. Person share .25.

Model 6. Between and within people, plus AI

Predictor	b	SE	95% CI	z	p	β
Intercept	1.15	0.26	0.64 to 1.66	4.40	< .001
Quantitative reasoning, person average	0.31	0.07	0.18 to 0.43	4.65	< .001	.21
Quantitative reasoning, vs own average	0.37	0.02	0.33 to 0.41	18.67	< .001	.26
Intuition, person average	0.53	0.07	0.40 to 0.66	8.02	< .001	.36
Intuition, vs own average	0.39	0.02	0.35 to 0.43	19.59	< .001	.29
AI recommender (vs human)	−0.51	0.09	−0.68 to -0.34	−5.90	< .001	−.08

2,051 ratings from 188 people. Person share .26.

Model 7. AI recommenders only

Predictor	b	SE	95% CI	z	p	β
Intercept	1.36	0.20	0.97 to 1.75	6.84	< .001
Quantitative reasoning	0.27	0.03	0.21 to 0.33	9.25	< .001	.27
Intuition	0.43	0.03	0.37 to 0.49	14.98	< .001	.43

929 ratings from 187 people. Person share .44.

Model 8. Human recommenders only

Predictor	b	SE	95% CI	z	p	β
Intercept	1.31	0.18	0.96 to 1.66	7.39	< .001
Quantitative reasoning	0.45	0.03	0.40 to 0.49	17.40	< .001	.45
Intuition	0.35	0.03	0.30 to 0.41	13.19	< .001	.35

1,122 ratings from 188 people. Person share .25.

Table 13. Does adding a term improve the fit?

What to look for. Every p is below .001, so each added term improved the model.

Comparison	χ²	df	p
Model 2 to Model 3: add AI versus human	32.32	1	< .001
Model 3 to Model 4: let the slopes differ for AI	20.21	2	< .001
Model 5 to Model 6: add AI versus human	34.55	1	< .001

Likelihood ratio tests, refit by maximum likelihood.

Table 13 as a figure. Each bar is how much a model improves when a term is added (χ²). Longer means a bigger improvement; with one degree of freedom, anything over 3.8 is past the usual cutoff.

Figure 8. What lifts trust differs for AI and for humans

For human recommenders, perceived quantitative reasoning is the stronger partner of trust (b = 0.45 against 0.35 for intuition). For AI, it is intuition (b = 0.43 against 0.27). The interaction model agrees: the two slopes differ by recommender type (Model 4: both interaction terms, p < .001).

Quantitative reasoning
Intuition
The other rating is held at its average. Bands are 95% confidence intervals.

Predicted trust from the AI-only and human-only models, with the other rating held at its average. Bands are 95% confidence intervals for the fixed effects.

Table 14. Correlations between people and within people

What to look for. Compare the two columns. Between-people values reflect who the person is; within-people values reflect which recommender is being rated.

Recommenders and pair	Between people, r (p)	Within people, r (p)
All recommenders: Trust and Intuition	.776 (< .001)	.562 (< .001)
All recommenders: Trust and Quantitative reasoning	.726 (< .001)	.519 (< .001)
All recommenders: Intuition and Quantitative reasoning	.769 (< .001)	.384 (< .001)
AI recommenders: Trust and Intuition	.781 (< .001)	.391 (< .001)
AI recommenders: Trust and Quantitative reasoning	.663 (< .001)	.283 (< .001)
AI recommenders: Intuition and Quantitative reasoning	.614 (< .001)	.224 (< .001)
Human recommenders: Trust and Intuition	.711 (< .001)	.526 (< .001)
Human recommenders: Trust and Quantitative reasoning	.725 (< .001)	.592 (< .001)
Human recommenders: Intuition and Quantitative reasoning	.744 (< .001)	.451 (< .001)

Between-person correlations use each person's average over the recommenders. Within-person correlations use each rating's distance from that person's own average, pooled across people.

Table 14 as a figure, for all recommenders together. The left grid is between people: do people who rate recommenders high on one thing rate them high on the other? The right grid is within people: for one person, do recommenders rated higher on one thing also get rated higher on the other?