Table 4.

Robustness of the reduction in re-identification risk ${\mathit{Risk}}_{\mathit{overall}}$ when shifting from the worst case to the Twitter and the voter registration scenarios.

BN Parameter				Twitter			Voter
$p_m$	$p_c$	$p_{fm}$	$p_{cu}$	Q1	Median	Q3	Q1	Median	Q3
0.2	0.2	0.2	0.2	96.85 (0.17)	100.0 (0.0)	100.0 (0.0)	42.42 (2.03)	91.18 (0.61)	96.84 (0.16)
0.5	0.5	0.5	0.5	97.49 (0.37)	100.0 (0.0)	100.0 (0.0)	37.85 (1.54)	90.38 (0.71)	95.69 (0.35)
0.8	0.8	0.8	0.8	97.51 (0.31)	100.0 (0.0)	100.0 (0.0)	30.2 (0.72)	88.36 (0.92)	94.59 (0.37)
0.5	0.2	0.8	0.5	98.47 (0.03)	100.0 (0.0)	100.0 (0.0)	41.2 (1.81)	95.56 (0.23)	98.51 (0.03)
0.2	0.2	0.5	0.8	98.46 (0.08)	100.0 (0.0)	100.0 (0.0)	65.21 (1.48)	95.47 (0.45)	98.46 (0.08)
…	…	…	…	…	…	…	…	…	…
Average				97.73 (0.22)	100 (0)	100 (0)	38.69 (1.33)	90.89 (0.65)	95.98 (0.26)
SD				0.44 (0.15)	0 (0)	0 (0)	10.22 (0.48)	3.09 (0.25)	1.75 (0.15)

The first several rows report the average (and SD) of the Q1, median, and Q3 across 10 runs of 1000 subjects each for several representative BN parameterizations. The final 2 rows report the average and standard deviation across all BN parameterizations.

BN: Bayesian network; Q: quartile.