The politics of deceptive borders: ‘biomarkers of deceit’ and the case of iBorderCtrl

Figures & data

Table 1. Experiment interview questions from the ADDS validation from O’Shea et al. (2018).

Question bumber	Question
1	What is your family name?
2	What is in your case?
3	Have you seen any posters of prohibited items?
4	Are there any items from the lists of prohibited items in your case?
5	How many items are in the case?
6	If you open the case and show me what is inside, will it confirm that your answers were true?
7	What is your first name?
8	When were you born?
9	Where were you born?
10	What is your current citizenship?
11	Please tell me the name of a friend or family member who can confirm your identity?
12	What is the relationship of this person to you?
13	Where does this person live?

Table 2. Performance of the ADDS in the validation experiment.

	ADDS performance for unseen persons		ADDS performance including the same person in the train and test datasets
Fold	T	D	T	D
1	100.00	57.00	94.30	93.69
2	50.00	36.00	93.62	94.96
3	50.00	100.00	94.31	95.15
4	90.00	100.00	94.23	95.46
5	100.00	10.00	95.50	95.41
6	72.00	100.00	95.45	95.91
7	100.00	100.00	95.75	96.22
8	38.00	100.00	95.91	96.28
9	80.00	60.00	96.67	96.45
10	-	-	96.55	96.78
Mean Acc.	75.56	73.67	95.23	95.50
STD Acc.	24.37	34.29	1.00	0.86

Source: Table IV in p. 7 in O’Shea et al. (2018).

Table 3. Confusion matrix of conditional probabilities of the ADDS.

		Ground truth
		Lie	No-lie
Test result	Test positive	73.66%	24.45%
	Test negative	26.34%	75.55%

Source Table IV in p.7 in O’Shea et al. (2018).

Table 4. Joint probability matrix of expected performance of the ADDS for a context of 5% of liars (prior probability of 0.05).

		Ground truth
		Lie (0.05)	No-lie (0.95)
Test result	Test positive	(True Pos.) 0.0368	(False Pos.) 0.2323
	Test negative	(False Neg.) 0.0132	(True Neg) 0.7177

Figure 1. Event tree to explain deception detection probability of each outcome. In this case we test the hypothesis of 1% of liars in the population.

Figure 2. Posterior probabilities of a positive and negative test for different scenarios of the frequency of deception. Highlighted values correspond to the scenario of 5% of liars. While the conditional probability is fixed, the probability of someone being a liar after a positive test depends on the general population.

O’Shea, J., Crockett, K., Khan, W., Kindynis, P., Antoniades, A., & Boultadakis, G. (2018). Intelligent deception detection through machine based Interviewing. 2018 International joint conference on neural networks (IJCNN) (pp. 1–8). https://doi.org/https://doi.org/10.1109/IJCNN.2018.8489392

 Google Scholar