Table 4:

Evaluation of the zero-truncated modelling approach using synthetic data obtained from the HCP project, using voxelwise (top) and clusterwise (bottom) inference. The true number of missing contrasts (n₀) for each one of the 12 datasets (6 for voxelwise thesholding and 6 for clusterwise thresholding) is shown in the second column. For each of the Negative Binomial and Delaporte methods, the estimated missing contrast count (${\widehat{n}}_0$), 95% bootstrap confidence interval for n₀, AIC score and BIC score are shown (smaller AIC and BIC are better).

Voxelwise
Negative Binomial					Delaporte
Split	n0	${\widehat{\mathbf{n}}}_0$	AIC	BIC	${\widehat{\mathbf{n}}}_0$	AIC	BIC
1	7	20 [14,27]	6576.4	6585.4	1 [1,4]	6562.8	6576.4
2	5	22 [15,29]	6583.3	6592.4	3 [1,8]	6576.2	6589.8
3	5	22 [16,30]	6575.9	6585.0	1 [1,21]	6566.1	6579.7
4	4	25 [18,33]	6603.8	6612.9	4 [1,25]	6601.7	6615.3
5	10	18 [13,24]	6539.4	6548.5	1 [0,1]	6504.1	6517.6
6	10	21 [15,29]	6557.0	6566.0	2 [1,10]	6550.0	6563.5
Clusterwise
Negative Binomial					Delaporte
Split	n0	${\widehat{\mathbf{n}}}_0$	AIC	BIC	${\widehat{\mathbf{n}}}_0$	AIC	BIC
1	148	167 [115,248]	3167.8	3176.4	71 [50,108]	3166.9	3179.8
2	144	151 [104,217]	3209.3	3217.9	58 [44,79]	3204.5	3217.4
3	150	161 [109,246]	3163.2	3171.8	95 [62,184]	3164.4	3177.2
4	151	154 [107,231]	3156.3	3164.8	106 [57,159]	3158.0	3170.9
5	153	148 [101,291]	3175.0	3183.6	98 [60,174]	3176.5	3189.4
6	152	151 [103,223]	3198.2	3206.8	89 [54,157]	3199.7	3212.5