Table 4 Step 4: Quantitative comparison of causal DAGs. Cells highlighted in bold indicate the best result in each of the metrics. BS and GTT win in those metrics that penalize complexity, but the 2nd Experts’/NO TEARS proposals are the ones that explain best the dataset.
From: Enhancing social science research on cyberbullying through human machine collaboration
Metrics | Naive | 1st experts’ proposal | Bayesian search | PC | GTT | 2nd experts’ and NO TEARS |
|---|---|---|---|---|---|---|
Mean Log likelihood [test sets] | \(-\)1248.54 | \(-\)1234.92 | \(-\)1200.72 | \(-\)1218.82 | \(-\)1200.72 | \(-\)1197.39 |
BIC | 54,121.09 | 54,362.15 | 7274.62 | 15,027.07 | 7274.62 | 7315.01 |
K2 score | \(-\)5521 | \(-\)5546.13 | \(-\)5421.98 | \(-\)5467.67 | \(-\)5421.98 | \(-\)5432.44 |
Correlation score | 0.61 | 0.63 | 0.67 | 0.63 | 0.67 | 0.63 |
