Matthew Leitao

Study 4 · Trust in medical AI

Study 4: 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. Study 4 has two parts. In Part 1 each person read and rated one recommender first, so comparisons are between groups. In Part 2 the same people then rated all three in a random order, so comparisons are within people. Trust and belief run from −3 to 3, warmth and competence from 0 to 4. Every recommender was voiced. Age is in four bands in the public file, so age coefficients differ in scale from the dissertation's.

Who took part

559 adults are in the released file. Not everyone finished Part 2, so the number of people varies by rating.

Table 1. The people in the released file

What to look for. The share in each group. Younger adults and women make up the largest groups.

Group People Share
Age group: 18-4426547.4%
Age group: 45-599817.5%
Age group: 60-69549.7%
Age group: 70+488.6%
Gender: Woman28150.3%
Gender: Man18733.5%
Race: White32858.7%
Race: Another or multiple14225.4%
Education: No college degree31856.9%
Education: College degree or more15227.2%
How religious: Not at all to slightly15828.3%
How religious: Moderately13924.9%
How religious: More than moderately17330.9%

Shares are of all 559 people. Left blank in the file: age group (94), gender (91), race (89), education (89), how religious (89). Race is shown as White or another or multiple because no other group reached 30 people. Age is in bands.

Table 1 as a figure

Each bar is the share of the 559 people in that group. Hover a bar for the head count.

0%20%40%60%80%100% Age group 18-44 (n = 265) 45-59 (n = 98) 60-69 (n = 54) 70+ (n = 48) Gender Woman (n = 281) Man (n = 187) Race White (n = 328) Another or multiple (n = 142) Education No college degree (n = 318) College degree or more (n = 152) How religious Not at all to slightly (n = 158) Moderately (n = 139) More than moderately (n = 173) Share of people (%)

The design and the check

Each person was randomly assigned a recommender to read first: 227 the humanlike ai, 207 the standard ai, 125 the doctor. They then rated all three, in one of six random orders. Letters give the order: H is the humanlike AI, N the standard AI and D the doctor.

Table 2. Who saw what first, and the six orders

What to look for. The groups are not equal in size, so Part 1 compares unequal groups.

Order Seen first People
HDNHumanlike AI110
HNDHumanlike AI117
DHNDoctor64
NHDStandard AI111
DNHDoctor61
NDHStandard AI96
Table 3. How humanlike the AI seemed (0 to 4), first rating

What to look for. Unlike Study 3, the humanlike AI was rated more humanlike. The interval for the difference stays above zero.

Comparison Humanlike AI, Standard AI: mean (SD) Difference [95% CI] t (df) p d [95% CI]
How humanlike the AI seemed: humanlike AI versus standard AI2.11 (1.14), 1.79 (1.17)0.32 [0.10, 0.54]2.91 (425.66).0040.28 [0.09, 0.47]

Welch t test, two-sided. Only people who first saw an AI answered this question.

Table 3 as a figure

The effect of the framing on how humanlike the AI seemed, in standard deviations, with a 95% interval.

smallmediumlarge 0.000.100.200.300.400.500.60 Humanlike versus standard AI Cohen's d

Part 1: each person saw one source first

This part repeats Study 3's design with a voice added. It asks whether the humanlike AI is trusted more than the standard AI when each person judges only one.

Table 4. Ratings by first source

What to look for. Compare means within a measure. The doctor is highest; the humanlike AI sits a little above the standard one.

Measure and source n Mean SD SE 95% CI Median Min Max
Trust (−3 to 3): Humanlike AI2270.941.470.10[0.75, 1.13]1.00−3.003.00
Trust (−3 to 3): Standard AI2070.751.520.11[0.54, 0.96]1.00−3.003.00
Trust (−3 to 3): Doctor1251.221.500.13[0.95, 1.48]1.00−3.003.00
Belief (−3 to 3): Humanlike AI2261.141.460.10[0.95, 1.33]1.00−3.003.00
Belief (−3 to 3): Standard AI2070.911.460.10[0.71, 1.11]1.00−3.003.00
Belief (−3 to 3): Doctor1251.331.450.13[1.07, 1.58]2.00−3.003.00
Warmth (0 to 4): Humanlike AI2272.161.230.08[2.00, 2.32]2.000.004.00
Warmth (0 to 4): Standard AI2071.861.160.08[1.71, 2.02]2.000.004.00
Warmth (0 to 4): Doctor1252.631.120.10[2.43, 2.83]3.000.004.00
Competence (0 to 4): Humanlike AI2272.361.040.07[2.22, 2.50]2.000.004.00
Competence (0 to 4): Standard AI2072.141.090.08[2.00, 2.29]2.000.004.00
Competence (0 to 4): Doctor1252.581.090.10[2.39, 2.78]3.000.004.00

Figure 2. Trust, every person

Each dot is one person's first rating. The curve shows where ratings pile up, and the bar marks the mean with its bootstrapped 95% interval.

Doctor 1.2 Humanlike AI 0.9 Standard AI 0.8 -3-2-10123 Trust (−3 to 3)

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

View as a table
MeasurenMean95% interval
Doctor1251.20.9 to 1.5
Humanlike AI2270.90.7 to 1.1
Standard AI2070.80.6 to 0.9

Figure 3. Belief, every person

Each dot is one person's first rating. The curve shows where ratings pile up, and the bar marks the mean with its bootstrapped 95% interval.

Doctor 1.3 Humanlike AI 1.1 Standard AI 0.9 -3-2-10123 Belief (−3 to 3)

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

View as a table
MeasurenMean95% interval
Doctor1251.31.1 to 1.6
Humanlike AI2261.10.9 to 1.3
Standard AI2070.90.7 to 1.1

Figure 4. Warmth, every person

Each dot is one person's first rating. The curve shows where ratings pile up, and the bar marks the mean with its bootstrapped 95% interval.

Doctor 2.6 Humanlike AI 2.2 Standard AI 1.9 01234 Warmth (0 to 4)

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

View as a table
MeasurenMean95% interval
Doctor1252.62.4 to 2.8
Humanlike AI2272.22.0 to 2.3
Standard AI2071.91.7 to 2.0

Figure 5. Competence, every person

Each dot is one person's first rating. The curve shows where ratings pile up, and the bar marks the mean with its bootstrapped 95% interval.

Doctor 2.6 Humanlike AI 2.4 Standard AI 2.1 01234 Competence (0 to 4)

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

View as a table
MeasurenMean95% interval
Doctor1252.62.4 to 2.8
Humanlike AI2272.42.2 to 2.5
Standard AI2072.12.0 to 2.3

Figure 6. Means and intervals

Table 4 as a figure. Each dot is a mean with its 95% interval, one panel per measure.

Trust 00.30.50.811.31.51.8 Mean (−3 to 3) Belief 00.30.50.811.31.51.8 Mean (−3 to 3) Warmth 1.31.51.822.32.52.83 Mean (0 to 4) Competence 1.822.32.52.83 Mean (0 to 4) Doctor Standard AI Humanlike AI
View as a table
Source Trust: mean [95% interval]Belief: mean [95% interval]Warmth: mean [95% interval]Competence: mean [95% interval]
Doctor 1.2 [0.9, 1.5]1.3 [1.1, 1.6]2.6 [2.4, 2.8]2.6 [2.4, 2.8]
Standard AI 0.8 [0.5, 1.0]0.9 [0.7, 1.1]1.9 [1.7, 2.0]2.1 [2.0, 2.3]
Humanlike AI 0.9 [0.8, 1.1]1.1 [0.9, 1.3]2.2 [2.0, 2.3]2.4 [2.2, 2.5]
Table 5. One-way ANOVAs across the three first sources

What to look for. The F and p columns say whether the sources differ at all; the Tukey columns say which pairs do.

Measure n F (df) p η² Tukey p: doctor, standard AI Tukey p: doctor, humanlike AI Tukey p: humanlike, standard AI
Trust559F(2, 556) = 3.81.023.014.017.219.385
Belief558F(2, 555) = 3.31.037.012.033.468.246
Warmth559F(2, 556) = 16.45< .001.056< .001.001.027
Competence559F(2, 556) = 6.67.001.023.001.150.091
Table 6. Planned and additional comparisons, Part 1

What to look for. The d column shows the size of each difference. The humanlike-versus-standard rows are the test of the framing.

Comparison First, Second: mean (SD) Difference [95% CI] t (df) p d [95% CI] Kind Tukey p
Trust: Doctor versus the two AI conditions combined1.22 (1.50), 0.85 (1.50)0.37 [0.07, 0.67]2.42 (200.37).0160.25 [0.05, 0.45]Planned
Trust: Humanlike AI versus standard AI0.94 (1.47), 0.75 (1.52)0.19 [−0.09, 0.47]1.32 (425.02).1880.13 [−0.06, 0.32]Planned.385
Trust: Doctor versus standard AI1.22 (1.50), 0.75 (1.52)0.47 [0.13, 0.80]2.73 (264.44).0070.31 [0.09, 0.53]Additional.017
Trust: Doctor versus humanlike AI1.22 (1.50), 0.94 (1.47)0.28 [−0.05, 0.60]1.67 (250.94).0950.19 [−0.03, 0.41]Additional.219
Belief: Doctor versus the two AI conditions combined1.33 (1.45), 1.03 (1.46)0.30 [0.01, 0.59]2.02 (202.76).0440.20 [0.00, 0.40]Planned
Belief: Humanlike AI versus standard AI1.14 (1.46), 0.91 (1.46)0.22 [−0.05, 0.50]1.60 (427.99).1110.15 [−0.04, 0.34]Planned.246
Belief: Doctor versus standard AI1.33 (1.45), 0.91 (1.46)0.41 [0.09, 0.74]2.53 (262.81).0120.29 [0.06, 0.51]Additional.033
Belief: Doctor versus humanlike AI1.33 (1.45), 1.14 (1.46)0.19 [−0.13, 0.51]1.18 (258.19).2390.13 [−0.09, 0.35]Additional.468
Warmth: Doctor versus the two AI conditions combined2.63 (1.12), 2.02 (1.21)0.61 [0.39, 0.84]5.31 (214.12)< .0010.52 [0.32, 0.72]Planned
Warmth: Humanlike AI versus standard AI2.16 (1.23), 1.86 (1.16)0.29 [0.07, 0.52]2.56 (431.65).0110.25 [0.06, 0.43]Planned.027
Warmth: Doctor versus standard AI2.63 (1.12), 1.86 (1.16)0.77 [0.51, 1.02]5.98 (268.74)< .0010.67 [0.44, 0.90]Additional< .001
Warmth: Doctor versus humanlike AI2.63 (1.12), 2.16 (1.23)0.47 [0.22, 0.73]3.66 (277.61)< .0010.40 [0.18, 0.62]Additional.001
Competence: Doctor versus the two AI conditions combined2.58 (1.09), 2.26 (1.07)0.33 [0.11, 0.54]2.95 (197.68).0040.30 [0.10, 0.50]Planned
Competence: Humanlike AI versus standard AI2.36 (1.04), 2.14 (1.09)0.22 [0.01, 0.42]2.10 (424.02).0360.20 [0.01, 0.39]Planned.091
Competence: Doctor versus standard AI2.58 (1.09), 2.14 (1.09)0.44 [0.20, 0.68]3.55 (261.22)< .0010.40 [0.18, 0.63]Additional.001
Competence: Doctor versus humanlike AI2.58 (1.09), 2.36 (1.04)0.22 [−0.01, 0.46]1.86 (245.7).0640.21 [−0.01, 0.43]Additional.150

Welch t tests, two-sided. d uses the pooled standard deviation and is positive when the first group is higher.

Figure 7. Effect sizes, Part 1

Table 6 as a figure. Each dot is Cohen's d with a 95% interval. The dashed lines mark a small, medium and large effect.

smallmediumlarge −0.30−0.20−0.100.000.100.200.300.400.500.600.700.80 Trust Doctor versus both AIs Humanlike versus standard AI Doctor versus standard AI Doctor versus humanlike AI Belief Doctor versus both AIs Humanlike versus standard AI Doctor versus standard AI Doctor versus humanlike AI Warmth Doctor versus both AIs Humanlike versus standard AI Doctor versus standard AI Doctor versus humanlike AI Competence Doctor versus both AIs Humanlike versus standard AI Doctor versus standard AI Doctor versus humanlike AI Cohen's d, first group minus second
Table 7. What Part 1 could detect

What to look for. The d column is the smallest true difference this part would find 80% of the time.

Comparison n, first n, second Smallest detectable d
Doctor versus the two AI conditions combined1254340.28
Humanlike AI versus standard AI2272070.27
Doctor versus standard AI1252070.32
Doctor versus humanlike AI1252270.31

Two-sided Welch test, α = .05, power = .80.

Part 2: everyone rated all three

After the first rating, each person rated the humanlike AI, the standard AI and the doctor. Seeing all three lets people compare them directly.

Table 8. Ratings of each recommender, side by side

What to look for. The standard AI is lowest. The humanlike AI and the doctor sit close together.

Measure and recommender n Mean SD SE 95% CI Median Min Max
Trust (−3 to 3): Humanlike AI5310.991.520.07[0.86, 1.12]1.00−3.003.00
Trust (−3 to 3): Standard AI5250.511.640.07[0.37, 0.65]0.00−3.003.00
Trust (−3 to 3): Doctor5251.131.510.07[1.00, 1.26]1.00−3.003.00
Belief (−3 to 3): Humanlike AI5301.131.510.07[1.00, 1.26]1.00−3.003.00
Belief (−3 to 3): Standard AI5250.631.640.07[0.49, 0.77]1.00−3.003.00
Belief (−3 to 3): Doctor5241.231.480.06[1.10, 1.35]1.00−3.003.00
Warmth (0 to 4): Humanlike AI5312.301.220.05[2.19, 2.40]2.000.004.00
Warmth (0 to 4): Standard AI5251.711.310.06[1.60, 1.83]2.000.004.00
Warmth (0 to 4): Doctor5232.511.130.05[2.42, 2.61]3.000.004.00
Competence (0 to 4): Humanlike AI5312.421.100.05[2.32, 2.51]2.000.004.00
Competence (0 to 4): Standard AI5251.941.210.05[1.84, 2.04]2.000.004.00
Competence (0 to 4): Doctor5252.551.100.05[2.46, 2.65]3.000.004.00

Figure 8. Trust, every rating

Each dot is one person's rating of that recommender in Part 2.

Doctor 1.1 Humanlike AI 1.0 Standard AI 0.5 -3-2-10123 Trust (−3 to 3)

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

View as a table
MeasurenMean95% interval
Doctor5251.11.0 to 1.3
Humanlike AI5311.00.9 to 1.1
Standard AI5250.50.4 to 0.7

Figure 9. Belief, every rating

Each dot is one person's rating of that recommender in Part 2.

Doctor 1.2 Humanlike AI 1.1 Standard AI 0.6 -3-2-10123 Belief (−3 to 3)

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

View as a table
MeasurenMean95% interval
Doctor5241.21.1 to 1.4
Humanlike AI5301.11.0 to 1.3
Standard AI5250.60.5 to 0.8

Figure 10. Warmth, every rating

Each dot is one person's rating of that recommender in Part 2.

Doctor 2.5 Humanlike AI 2.3 Standard AI 1.7 01234 Warmth (0 to 4)

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

View as a table
MeasurenMean95% interval
Doctor5232.52.4 to 2.6
Humanlike AI5312.32.2 to 2.4
Standard AI5251.71.6 to 1.8

Figure 11. Competence, every rating

Each dot is one person's rating of that recommender in Part 2.

Doctor 2.5 Humanlike AI 2.4 Standard AI 1.9 01234 Competence (0 to 4)

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

View as a table
MeasurenMean95% interval
Doctor5252.52.5 to 2.6
Humanlike AI5312.42.3 to 2.5
Standard AI5251.91.8 to 2.0

Figure 12. Means and intervals, side by side

Table 8 as a figure. Each dot is a mean with its 95% interval, one panel per measure.

Trust 00.30.50.811.31.51.8 Mean (−3 to 3) Belief 00.30.50.811.31.51.8 Mean (−3 to 3) Warmth 1.31.51.822.32.52.83 Mean (0 to 4) Competence 1.822.32.52.83 Mean (0 to 4) Doctor Standard AI Humanlike AI
View as a table
Recommender Trust: mean [95% interval]Belief: mean [95% interval]Warmth: mean [95% interval]Competence: mean [95% interval]
Doctor 1.1 [1.0, 1.3]1.2 [1.1, 1.4]2.5 [2.4, 2.6]2.5 [2.5, 2.6]
Standard AI 0.5 [0.4, 0.7]0.6 [0.5, 0.8]1.7 [1.6, 1.8]1.9 [1.8, 2.0]
Humanlike AI 1.0 [0.9, 1.1]1.1 [1.0, 1.3]2.3 [2.2, 2.4]2.4 [2.3, 2.5]
Table 9. Repeated-measures ANOVAs across the three recommenders

What to look for. The corrected p column says whether the three recommenders differ. The last column is what the dissertation reported for the same test.

Measure People F (df) p Partial η² ε p, corrected As printed in the dissertation
Trust493F(2, 984) = 55.54< .001.101.94< .001F(2, 1573) = 1.65, p = .192
Belief493F(2, 984) = 47.29< .001.088.92< .001F(2, 1575) = 1.41, p = .245

Computed on the people who rated all three recommenders. The values printed in the dissertation could not be reproduced from the data: their degrees of freedom are larger than the number of people allows, which points to a test that counted ratings rather than people. The pairwise tests in Table 10 agree with the repeated-measures result.

Table 10. Paired comparisons, Part 2

What to look for. The dz column shows how large each difference is within a person. Humanlike versus standard is clearly positive; doctor versus humanlike is close to zero.

Comparison n First, second: mean (SD) Difference [95% CI] t (df) p Bonferroni p dz [95% CI] Higher / same / lower
Trust: Humanlike AI versus standard AI5041.02 (1.50), 0.51 (1.65)0.50 [0.38, 0.63]7.98 (503)< .001< .0010.36 [0.27, 0.45]192 / 251 / 61
Trust: Doctor versus standard AI5041.14 (1.48), 0.52 (1.65)0.61 [0.48, 0.74]9.28 (503)< .001< .0010.41 [0.32, 0.50]200 / 244 / 60
Trust: Doctor versus humanlike AI5071.13 (1.51), 1.02 (1.51)0.11 [0.01, 0.21]2.11 (506).036.1070.09 [0.01, 0.18]119 / 303 / 85
Belief: Humanlike AI versus standard AI5041.13 (1.50), 0.63 (1.65)0.50 [0.36, 0.63]7.20 (503)< .001< .0010.32 [0.23, 0.41]193 / 238 / 73
Belief: Doctor versus standard AI5041.22 (1.46), 0.65 (1.64)0.57 [0.44, 0.70]8.65 (503)< .001< .0010.39 [0.29, 0.48]199 / 242 / 63
Belief: Doctor versus humanlike AI5061.23 (1.48), 1.15 (1.51)0.08 [−0.03, 0.18]1.48 (505).139.4170.07 [−0.02, 0.15]113 / 306 / 87
Warmth: Humanlike AI versus standard AI5042.32 (1.22), 1.70 (1.33)0.61 [0.51, 0.72]11.25 (503)< .001< .0010.50 [0.41, 0.59]227 / 220 / 57
Warmth: Doctor versus standard AI5032.51 (1.13), 1.70 (1.33)0.81 [0.69, 0.92]13.91 (502)< .001< .0010.62 [0.52, 0.72]252 / 202 / 49
Warmth: Doctor versus humanlike AI5052.51 (1.14), 2.31 (1.22)0.20 [0.12, 0.28]4.76 (504)< .001< .0010.21 [0.12, 0.30]130 / 300 / 75
Competence: Humanlike AI versus standard AI5042.44 (1.10), 1.93 (1.23)0.50 [0.41, 0.60]10.24 (503)< .001< .0010.46 [0.36, 0.55]205 / 241 / 58
Competence: Doctor versus standard AI5042.54 (1.10), 1.93 (1.22)0.61 [0.51, 0.71]12.00 (503)< .001< .0010.53 [0.44, 0.63]226 / 226 / 52
Competence: Doctor versus humanlike AI5072.55 (1.10), 2.43 (1.11)0.12 [0.04, 0.19]2.95 (506).003.0100.13 [0.04, 0.22]127 / 295 / 85

Paired t tests, two-sided, with a Bonferroni correction across the three comparisons of each measure. dz is positive when the first recommender is higher.

Figure 13. Effect sizes, Part 2

Table 10 as a figure. Each dot is dz with a 95% interval. The dashed lines mark a small, medium and large effect.

smallmediumlarge −0.100.000.100.200.300.400.500.600.700.80 Trust Humanlike versus standard AI Doctor versus standard AI Doctor versus humanlike AI Belief Humanlike versus standard AI Doctor versus standard AI Doctor versus humanlike AI Warmth Humanlike versus standard AI Doctor versus standard AI Doctor versus humanlike AI Competence Humanlike versus standard AI Doctor versus standard AI Doctor versus humanlike AI dz, first recommender minus second

Alone versus side by side

The same contrasts, estimated in the two parts. The humanlike effect grows when people can compare. Part 1 uses the pooled-SD d and Part 2 uses dz, so the two are close but not identical measures.

Table 11. The same effects in the two parts

What to look for. Compare the two columns for each row. The humanlike-versus-standard effect is larger side by side.

Comparison Seen alone (Part 1): d [95% CI] Side by side (Part 2): dz [95% CI]
Trust: Humanlike versus standard AI0.13 [−0.06, 0.32]0.36 [0.27, 0.45]
Trust: Doctor versus standard AI0.31 [0.09, 0.53]0.41 [0.32, 0.50]
Trust: Doctor versus humanlike AI0.19 [−0.03, 0.41]0.09 [0.01, 0.18]
Belief: Humanlike versus standard AI0.15 [−0.04, 0.34]0.32 [0.23, 0.41]
Belief: Doctor versus standard AI0.29 [0.06, 0.51]0.39 [0.29, 0.48]
Belief: Doctor versus humanlike AI0.13 [−0.09, 0.35]0.07 [−0.02, 0.15]

Figure 14. The same effects in the two parts

Table 11 as a figure. Each pair of dots is one contrast, first seen alone and then side by side.

smallmediumlarge −0.20−0.100.000.100.200.300.400.500.600.70 Trust Humanlike versus standard AI, seen alone Humanlike versus standard AI, side by side Doctor versus standard AI, seen alone Doctor versus standard AI, side by side Doctor versus humanlike AI, seen alone Doctor versus humanlike AI, side by side Belief Humanlike versus standard AI, seen alone Humanlike versus standard AI, side by side Doctor versus standard AI, seen alone Doctor versus standard AI, side by side Doctor versus humanlike AI, seen alone Doctor versus humanlike AI, side by side Effect size

Models

Part 1 is modelled with regressions and Part 2 with mixed models that give each person their own baseline. The reference is the humanlike AI. When warmth and competence enter, the standard-AI and doctor coefficients drop to about zero.

Table 12a1. Part 1, trust: condition only

What to look for. The Standard AI and Doctor rows are differences from the humanlike AI.

Predictor b 95% CI SE t p
Intercept0.938[0.744, 1.132]0.0999.45< .001
Standard AI (versus humanlike AI)−0.190[−0.472, 0.092]0.144−1.32.188
Doctor (versus humanlike AI)0.278[−0.049, 0.605]0.1671.67.096

559 people. R² .014, adjusted R² .010, 556 residual degrees of freedom. Intervals are b ± 1.96 SE.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.60−0.40−0.200.000.200.400.600.80 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Effect on trust (points)
Table 12a2. Part 1, trust: add gender and age

What to look for. The Standard AI and Doctor rows are differences from the humanlike AI.

Predictor b 95% CI SE t p
Intercept0.919[0.590, 1.248]0.1685.47< .001
Standard AI (versus humanlike AI)−0.176[−0.490, 0.138]0.160−1.11.270
Doctor (versus humanlike AI)0.231[−0.118, 0.580]0.1781.30.195
Man (versus woman)0.134[−0.144, 0.412]0.1420.94.346
Age group (1 to 4)0.019[−0.114, 0.152]0.0680.27.785

463 people. R² .013, adjusted R² .005, 458 residual degrees of freedom. Intervals are b ± 1.96 SE.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.60−0.40−0.200.000.200.400.600.80 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Man (versus woman) Age group (1 to 4) Effect on trust (points)
Table 12a3. Part 1, trust: add warmth and competence

What to look for. The Standard AI and Doctor rows are differences from the humanlike AI.

Predictor b 95% CI SE t p
Intercept−1.290[−1.674, −0.906]0.196−6.59< .001
Standard AI (versus humanlike AI)0.061[−0.194, 0.316]0.1300.47.639
Doctor (versus humanlike AI)0.055[−0.229, 0.339]0.1450.38.706
Man (versus woman)0.028[−0.197, 0.253]0.1150.24.810
Age group (1 to 4)0.151[0.041, 0.261]0.0562.71.007
Warmth0.239[0.108, 0.370]0.0673.54< .001
Competence0.614[0.469, 0.759]0.0748.31< .001

463 people. R² .359, adjusted R² .350, 456 residual degrees of freedom. Intervals are b ± 1.96 SE.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.40−0.200.000.200.400.600.801.00 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Man (versus woman) Age group (1 to 4) Warmth Competence Effect on trust (points)
Table 12b1. Part 1, belief: condition only

What to look for. The Standard AI and Doctor rows are differences from the humanlike AI.

Predictor b 95% CI SE t p
Intercept1.137[0.947, 1.327]0.09711.74< .001
Standard AI (versus humanlike AI)−0.224[−0.498, 0.050]0.140−1.60.110
Doctor (versus humanlike AI)0.191[−0.127, 0.509]0.1621.18.240

558 people. R² .012, adjusted R² .008, 555 residual degrees of freedom. Intervals are b ± 1.96 SE.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.60−0.40−0.200.000.200.400.600.80 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Effect on belief (points)
Table 12b2. Part 1, belief: add gender and age

What to look for. The Standard AI and Doctor rows are differences from the humanlike AI.

Predictor b 95% CI SE t p
Intercept1.010[0.689, 1.331]0.1646.16< .001
Standard AI (versus humanlike AI)−0.184[−0.490, 0.122]0.156−1.18.238
Doctor (versus humanlike AI)0.228[−0.111, 0.567]0.1731.32.189
Man (versus woman)0.145[−0.127, 0.417]0.1391.04.297
Age group (1 to 4)0.037[−0.094, 0.168]0.0670.55.581

463 people. R² .015, adjusted R² .007, 458 residual degrees of freedom. Intervals are b ± 1.96 SE.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.60−0.40−0.200.000.200.400.600.80 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Man (versus woman) Age group (1 to 4) Effect on belief (points)
Table 12b3. Part 1, belief: add warmth and competence

What to look for. The Standard AI and Doctor rows are differences from the humanlike AI.

Predictor b 95% CI SE t p
Intercept−1.095[−1.475, −0.715]0.194−5.66< .001
Standard AI (versus humanlike AI)0.042[−0.209, 0.293]0.1280.33.743
Doctor (versus humanlike AI)0.062[−0.220, 0.344]0.1440.43.666
Man (versus woman)0.044[−0.179, 0.267]0.1140.38.701
Age group (1 to 4)0.163[0.055, 0.271]0.0552.96.003
Warmth0.223[0.092, 0.354]0.0673.34< .001
Competence0.588[0.445, 0.731]0.0738.07< .001

463 people. R² .343, adjusted R² .335, 456 residual degrees of freedom. Intervals are b ± 1.96 SE.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.40−0.200.000.200.400.600.801.00 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Man (versus woman) Age group (1 to 4) Warmth Competence Effect on belief (points)
Table 13a1. Part 2, trust: recommender only

What to look for. Read the standardized β column to see which ratings carry the most weight.

Predictor b [95% CI] Standardized β z p
Intercept0.989 [0.858, 1.120]14.78< .001
Standard AI (versus humanlike AI)−0.497 [−0.614, −0.379]−0.15−8.25< .001
Doctor (versus humanlike AI)0.124 [0.006, 0.241]0.042.06.040

1,581 ratings from 559 people. Person variance 1.49, residual 0.94. Share of variance between people (ICC) .61. R² .03 from the predictors alone, .63 with the person effects.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.80−0.60−0.40−0.200.000.200.40 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Effect on trust (points)
Table 13a2. Part 2, trust: add gender and age

What to look for. Read the standardized β column to see which ratings carry the most weight.

Predictor b [95% CI] Standardized β z p
Intercept0.891 [0.627, 1.154]6.62< .001
Standard AI (versus humanlike AI)−0.490 [−0.616, −0.365]−0.15−7.65< .001
Doctor (versus humanlike AI)0.114 [−0.011, 0.240]0.031.79.074
Man (versus woman)0.230 [−0.019, 0.479]0.071.81.070
Age group (1 to 4)0.027 [−0.093, 0.146]0.020.44.661

1,389 ratings from 463 people. Person variance 1.45, residual 0.95. Share of variance between people (ICC) .60. R² .03 from the predictors alone, .62 with the person effects.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.80−0.60−0.40−0.200.000.200.400.60 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Man (versus woman) Age group (1 to 4) Effect on trust (points)
Table 13a3. Part 2, trust: add warmth and competence

What to look for. Read the standardized β column to see which ratings carry the most weight.

Predictor b [95% CI] Standardized β z p
Intercept−1.324 [−1.566, −1.082]−10.71< .001
Standard AI (versus humanlike AI)−0.023 [−0.139, 0.094]−0.01−0.38.704
Doctor (versus humanlike AI)−0.010 [−0.123, 0.102]−0.00−0.18.859
Man (versus woman)0.026 [−0.142, 0.195]0.010.31.760
Age group (1 to 4)0.155 [0.074, 0.235]0.103.75< .001
Warmth0.331 [0.262, 0.400]0.279.43< .001
Competence0.529 [0.452, 0.606]0.4013.50< .001

1,388 ratings from 463 people. Person variance 0.55, residual 0.75. Share of variance between people (ICC) .42. R² .42 from the predictors alone, .67 with the person effects.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.40−0.200.000.200.400.600.80 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Man (versus woman) Age group (1 to 4) Warmth Competence Effect on trust (points)
Table 13b1. Part 2, belief: recommender only

What to look for. Read the standardized β column to see which ratings carry the most weight.

Predictor b [95% CI] Standardized β z p
Intercept1.127 [0.996, 1.257]16.95< .001
Standard AI (versus humanlike AI)−0.494 [−0.616, −0.372]−0.15−7.93< .001
Doctor (versus humanlike AI)0.082 [−0.040, 0.204]0.031.32.188

1,579 ratings from 558 people. Person variance 1.38, residual 1.00. Share of variance between people (ICC) .58. R² .03 from the predictors alone, .59 with the person effects.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.80−0.60−0.40−0.200.000.200.40 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Effect on belief (points)
Table 13b2. Part 2, belief: add gender and age

What to look for. Read the standardized β column to see which ratings carry the most weight.

Predictor b [95% CI] Standardized β z p
Intercept0.963 [0.704, 1.222]7.29< .001
Standard AI (versus humanlike AI)−0.490 [−0.621, −0.360]−0.15−7.38< .001
Doctor (versus humanlike AI)0.093 [−0.037, 0.223]0.031.40.162
Man (versus woman)0.300 [0.056, 0.544]0.092.41.016
Age group (1 to 4)0.024 [−0.093, 0.141]0.020.40.687

1,389 ratings from 463 people. Person variance 1.35, residual 1.02. Share of variance between people (ICC) .57. R² .04 from the predictors alone, .58 with the person effects.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.80−0.60−0.40−0.200.000.200.400.600.80 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Man (versus woman) Age group (1 to 4) Effect on belief (points)
Table 13b3. Part 2, belief: add warmth and competence

What to look for. Read the standardized β column to see which ratings carry the most weight.

Predictor b [95% CI] Standardized β z p
Intercept−1.245 [−1.487, −1.003]−10.06< .001
Standard AI (versus humanlike AI)−0.025 [−0.146, 0.096]−0.01−0.41.683
Doctor (versus humanlike AI)−0.030 [−0.146, 0.087]−0.01−0.50.618
Man (versus woman)0.098 [−0.069, 0.264]0.031.15.252
Age group (1 to 4)0.151 [0.071, 0.231]0.103.70< .001
Warmth0.319 [0.249, 0.390]0.268.84< .001
Competence0.538 [0.459, 0.616]0.4013.37< .001

1,388 ratings from 463 people. Person variance 0.52, residual 0.81. Share of variance between people (ICC) .39. R² .42 from the predictors alone, .64 with the person effects.

As a figure. Each dot is a coefficient (b) with its 95% interval.

−0.40−0.200.000.200.400.600.80 Standard AI (versus humanlike AI) Doctor (versus humanlike AI) Man (versus woman) Age group (1 to 4) Warmth Competence Effect on belief (points)