Who Made That Decision and Why? Users’ Perceptions of Human Versus AI Decision-Making and the Power of Explainable-AI

Figures & data

Table 1. Participants’ demographic distribution (gender and age group).

# participants (%)	18–34	35–50	50+
Female	451 (14.70%)	572 (18.64%)	334 (10.89%)
Male	737 (24.02%)	665 (21.68%)	309 (10.07%)

Table 2. Distribution of the various manipulations used in the experiment.

Manipulation			HDM			ADM
Input	Output	Explanation	N	Fairness	Comprehension	N	Fairness	Comprehension
Above average candidate	Desirable	NON	40	2.125 (0.871)	2.225 (0.880)	45	1.933 (1.104)	1.778 (1.030)
		CAS	44	1.955 (0.928)	2.182 (0.833)	46	1.935 (0.791)	1.630 (1.240)
		CER	45	2.133 (1.108)	2.222 (0.813)	39	1.923 (1.095)	2.026 (1.000)
		DEM	44	1.250 (1.639)	1.500 (1.390)	45	1.311 (1.411)	1.244 (1.537)
		INP	39	2.026 (0.620)	1.974 (1.050)	41	1.805 (0.943)	1.634 (0.957)
		SEN	45	2.044 (0.759)	1.756 (1.057)	44	1.909 (0.793)	1.864 (0.967)
	Neutral	NON	43	1.605 (1.164)	1.767 (1.309)	42	0.810 (1.722)	1.429 (1.400)
		CAS	44	1.318 (1.592)	2.045 (0.824)	44	1.250 (1.680)	1.864 (1.307)
		CER	40	1.425 (1.430)	1.625 (1.317)	43	1.581 (1.105)	1.860 (0.930)
		DEM	36	0.917 (2.019)	1.500 (1.555)	41	0.610 (1.898)	1.732 (0.856)
		INP	45	1.422 (1.390)	1.622 (1.371)	41	1.171 (1.430)	1.512 (1.500)
		SEN	45	1.422 (1.358)	2.000 (0.789)	46	1.283 (1.455)	1.913 (0.952)
	Undesirable	NON	42	0.167 (2.034)	0.643 (2.068)	42	−0.238 (2.056)	0.524 (2.026)
		CAS	41	0.951 (1.667)	1.634 (1.478)	43	0.209 (2.030)	1.233 (1.378)
		CER	43	0.256 (2.092)	0.698 (1.824)	44	0.409 (1.992)	1.250 (1.611)
		DEM	44	0.886 (1.682)	1.205 (1.455)	40	−0.200 (2.040)	1.050 (1.731)
		INP	44	0.545 (1.712)	1.727 (1.420)	47	0.149 (2.000)	1.553 (1.555)
		SEN	39	0.667 (1.858)	1.538 (1.482)	43	0.977 (1.705)	1.907 (1.030)
Below average candidate	Desirable	NON	41	0.610 (1.765)	0.951 (1.987)	44	0.136 (1.866)	0.773 (2.098)
		CAS	40	1.025 (1.557)	1.775 (1.557)	44	1.091 (1.635)	1.409 (1.497)
		CER	44	1.432 (1.498)	1.500 (1.406)	43	0.814 (1.980)	1.116 (1.768)
		DEM	39	0.923 (1.685)	1.744 (1.126)	44	0.591 (1.981)	1.318 (1.634)
		INP	37	1.432 (1.733)	1.649 (1.340)	41	1.073 (1.702)	1.537 (1.579)
		SEN	43	1.279 (1.436)	1.744 (1.296)	39	1.308 (1.651)	1.872 (0.882)
	Neutral	NON	42	1.690 (1.123)	2.214 (0.674)	47	1.660 (0.832)	1.957 (0.849)
		CAS	46	1.457 (1.280)	1.804 (1.191)	40	1.375 (1.155)	1.675 (1.034)
		CER	45	1.956 (0.918)	2.022 (1.022)	41	2.146 (0.926)	2.195 (0.917)
		DEM	43	1.721 (1.148)	1.721 (1.127)	46	1.065 (1.607)	1.565 (1.313)
		INP	46	1.674 (1.490)	1.978 (1.277)	38	1.658 (1.382)	1.842 (1.288)
		SEN	39	1.795 (0.992)	2.077 (0.888)	39	1.282 (1.467)	1.667 (1.268)
	Undesirable	NON	46	1.913 (1.039)	2.239 (0.785)	43	1.884 (0.993)	2.279 (0.726)
		CAS	42	1.929 (1.055)	1.881 (1.074)	43	1.674 (1.215)	1.907 (0.910)
		CER	44	1.909 (1.018)	2.091 (0.949)	45	1.533 (1.343)	1.756 (1.319)
		DEM	41	1.561 (1.531)	1.829 (1.323)	44	1.545 (1.339)	1.955 (0.952)
		INP	43	1.744 (1.123)	2.070 (0.899)	45	1.311 (1.488)	1.511 (1.204)
		SEN	41	1.780 (1.179)	2.073 (0.777)	41	1.854 (1.159)	2.268 (0.699)

The number of cases (N), average score, and STD (in parentheses) of the fairness perception and comprehension perception for each combination are presented.

Table 3. Aggregated results according to the decision-maker type.

Decision-maker type	N	Fairness	Comprehension
HDM	1525	1.422 (1.511)	1.760 (1.312)
ADM	1543	1.188 (1.648)	1.626 (1.354)

The number of cases (N) for each decision-maker type, the average score, and STD (in parentheses) of the fairness and comprehension perceptions are presented.

Table 4. Multivariate ordinal regression results.

Factors	Comparisons	FAIRNESS		COMPREHENSION
		HDM	ADM	HDM	ADM
Input (Candidate)	Above average vs. Below average	−0.235 (0.094)*	−0.285 (0.093)**	−0.207 (0.095)*	−0.223 (0.094)*
Output	Desirable vs. Neutral	0.016 (0.115)	0.028 (0.114)	−0.085 (0.116)	−0.278 (0.115)*
	Desirable vs. Undesirable	0.326 (0.115)**	0.369 (0.113)***	0.152 (0.116)	−0.125 (0.114)
	Neutral vs. Undesirable	0.310 (0.114)**	0.341 (0.114)**	0.236 (0.115)*	0.153 (0.115)
Explanation	CAS vs. NON	0.033 (0.162)	0.283 (0.159)	0.136 (0.164)	0.121 (0.161)
	CER vs. NON	0.218 (0.162)	0.451 (0.16)**	−0.118 (0.163)	0.289 (0.162)
	DEM vs. NON	−0.160 (0.163)	−0.164 (0.158)	−0.374 (0.165)*	−0.027 (0.160)
	INP vs. NON	0.110 (0.163)	0.224 (0.161)	0.101 (0.165)	0.129 (0.162)
	SEN vs. NON	0.068 (0.163)	0.456 (0.161)**	0.018 (0.164)	0.458 (0.163)**
	CAS vs. SEN	−0.035 (0.162)	−0.173 (0.161)	0.118 (0.164)	−0.337 (0.163)*
	CER vs. SEN	0.150 (0.162)	−0.005 (0.163)	−0.136 (0.163)	−0.169 (0.165)
	DEM vs. SEN	−0.228 (0.164)	−0.620 (0.161)***	−0.392 (0.165)*	−0.485 (0.163)**
	INP vs. SEN	0.042 (0.163)	−0.232 (0.163)	0.083 (0.165)	−0.329 (0.165)*
	CAS vs. INP	−0.076 (0.162)	0.059 (0.161)	0.035 (0.164)	−0.008 (0.162)
	CER vs. INP	0.108 (0.162)	0.227 (0.163)	−0.218 (0.163)	0.160 (0.164)
	DEM vs. INP	−0.270 (0.164)	−0.270 (0.161)*	−0.475 (0.165)**	−0.156 (0.162)
	CAS vs. DEM	0.193 (0.163)	0.615 (0.161)**	0.510 (0.165)**	0.149 (0.161)
	CER vs. DEM	0.378 (0.163)*	0.388 (0.161)***	0.257 (0.163)	0.316 (0.162)
	CAS vs. CER	−0.185 (0.161)	−0.168 (0.161)	0.253 (0.163)	−0.167 (0.163)
Gender	Female vs. Male	0.001 (0.096)	0.013 (0.095)	0.062 (0.096)	0.068 (0.096)
Age Group	21–34 vs. 50+	0.358 (0.129)**	0.299 (0.132)*	−0.191 (0.13)	0.036 (0.134)
	35–50 vs. 50+	0.255 (0.125)*	0.126 (0.129)	−0.202 (0.127)	−0.207 (0.131)
	21–34 vs. 35–50	0.103 (0.108)	0.173 (0.106)	0.011 (0.108)	0.243 (0.107)*
Education Level	Bachelor’s degree vs. Master’s degree or higher	−0.069 (0.123)	−0.177 (0.123)	−0.018 (0.124)	−0.084 (0.124)
	High school or lower vs. Master’s degree or higher	−0.301 (0.163)	−0.721 (0.16)***	0.002 (0.165)	−0.272 (0.162)
	Bachelor’s degree vs. High school or lower	0.233 (0.136)	0.544 (0.133)***	−0.020 (0.137)	0.188 (0.135)
Computer Skills	Excellent vs. Average	0.293 (0.155)	0.800 (0.149)***	0.781 (0.157)***	0.746 (0.151)
	Good vs. Average	−0.097 (0.150)	0.404 (0.142)**	0.031 (0.151)	−0.007 (0.144)
	Excellent vs. Good	0.390 (0.102)***	0.396 (0.103)***	0.750 (0.104)***	0.753 (0.105)***
Agreeableness	Low vs. High	0.061 (0.120)	0.119 (0.120)	−0.076 (0.121)	−0.212 (0.120)
Conscientiousness	Low vs. High	−0.077 (0.144)	−0.378 (0.150)*	−0.459 (0.145)**	−0.035 (0.149)
Emotional Stability	Low vs. High	−0.275 (0.124)*	−0.059 (0.119)	−0.436 (0.124)***	−0.285 (0.120)*
Extraversion	Low vs. High	−0.395 (0.101)***	−0.256 (0.101)*	−0.261 (0.102)**	−0.343 (0.102)***
Openness	Low vs. High	−0.142 (0.120)	0.193 (0.126)	−0.284 (0.120)*	−0.340 (0.126)**

The Beta coefficients, STD (in parentheses), and significance are presented for each comparison of each factor. Significant results are in bold, *p < 0.05; **p < 0.01; ***p < 0.001.

Table 5. Aggregated results according to the decision-maker type and explanation style.

Explanation style	HDM			ADM
	N	Fairness	Comprehension	N	Fairness	Comprehension
NON	254	1.358 (1.563)	1.681 (1.536)	263	1.049 (1.718)	1.468 (1.579)
CAS	257	1.447 (1.425)	1.891 (1.198)	256	1.250 (1.579)	1.613 (1.276)
CER	261	1.529 (1.528)	1.701 (1.363)	252	1.377 (1.605)	1.683 (1.372)
DEM	247	1.219 (1.659)	1.579 (1.353)	258	0.829 (1.816)	1.473 (1.412)
INP	254	1.465 (1.476)	1.839 (1.256)	250	1.160 (1.641)	1.588 (1.378)
SEN	252	1.508 (1.376)	1.865 (1.094)	252	1.437 (1.442)	1.917 (0.995)
All Explanations	1271	1.435 (1.499)	1.776 (1.262)	1280	1.216 (1.631)	1.659 (1.301)

The average score and STD (in parentheses) of the fairness and comprehension perceptions of the output for each combination are presented. The fairness comprehension perceptions are colored in orange and blue, respectively – the darker the color, the higher the level of perception.

Table 6. Kruskal-Wallis Results of HDM compared to ADM for a specific explanation style. Significant results appear in bold font.

Comparison		Fairness (H3.3)	Comprehension (H3.4)
HDM- CAS	ADM-CAS	H(1)=1.378	H(1)=8.576**, ${\mathit{\eta }}^2$=0.017
HDM-CER	ADM-CER	H(1)=1.079	H(1)=0.019
HDM-DEM	ADM-DEM	H(1)=7.114**, ${\mathit{\eta }}^2$=0.012	H(1)=0.777
HDM-INP	ADM-INP	H(1)=4.286*, ${\mathit{\eta }}^2$=0.008	H(1)=4.819*, ${\mathit{\eta }}^2$=0.009
HDM-SEN	ADM-SEN	H(1)=0.203	H(1)=0.753

*p < 0.05; **p < 0.01; ***p < 0.001.

Table 7. Kruskal-Wallis Results of HDM with no explanation vs. ADM with specific explanation style.

Comparison		Fairness (H3.7)	Comprehension (H3.8)
HDM-NON	ADM-CAS	H(1)=1.041	H(1)=15.381*, ${\mathit{\eta }}^2$=0.028
HDM-NON	ADM-CER	H(1)=0.106	H(1)=0.651
HDM-NON	ADM-DEM	H(1)=13.260***, ${\mathit{\eta }}^2$=0.025	H(1)=8.289**, ${\mathit{\eta }}^2$=0.016
HDM-NON	ADM-INP	H(1)=2.219	H(1)=3.268
HDM-NON	ADM-SEN	H(1)=2.219	H(1)=0.008

Significant results appear in bold font.

*p < 0.05; **p < 0.01; ***p < 0.001.

Figure 1. Distribution of the fairness (in orange) and compression (in blue) perception scores for each decision-maker type within the explanation style. The average score of the fairness and comprehension perceptions appear as the grey line intersecting each box. The box limits indicate the range of the Central 50% of the data (for example: dark blue represents the second quarter, light blue represents the third quarter), with a Central line marking the median value. Comparison (hypotheses) that were significant are marked with a connecting line.

Data availability statement

The dataset that was collected can be found at: https://zenodo.org/records/10374874.