| OLS | LASSO | Ridge |
---|
| Corruption risk | Corruption risk | Corruption risk |
---|
Design-Build | 0.003 | 0.002 | 0.003 |
[0.004] | | |
Urgency | −0.005 | −0.003 | −0.004 |
[0.003] | | |
Negotiated | 0.005 | 0.000 | 0.002 |
[0.005] | | |
Negotiated-No Tender | 0.003 | 0.001 | 0.003 |
[0.004] | | |
Price Only—w. ABA | −0.003 | −0.003 | −0.003 |
[0.004] | | |
Scoring Rule (MEAT) | 0.008∗ | 0.008 | 0.008 |
[0.004] | | |
Open Tender Days | −0.003 | 0.000 | −0.001 |
[0.005] | | |
Open Tender Day V. | 0.000 | 0.001 | 0.001 |
[0.004] | | |
Observations | 12,623 | 12,623 | 12,623 |
Adj R2 | 0.036 | | |
MSE | 0.355 | 0.126 | 0.126 |
False Positive | 3323 | 3352 | 3334 |
False Negative | 2159 | 2158 | 2170 |
Precision | 0.228 | 0.227 | 0.225 |
Recall | 0.312 | 0.313 | 0.309 |
F Measure | 0.264 | 0.263 | 0.261 |
Threshold | 0.192 | 0.189 | 0.189 |
- ∗ p<0.1, ∗∗ p<0.05, ∗∗∗ p<0.01. All specifications include year and region fixed effects. Robust standard errors in parentheses for OLS estimates. Due to the limited number of observations in our sample, LASSO and Ridge regressions are evaluated through a standard k-fold cross-validation method (with k = 10), and not through the more common train-test split. MSE is equal to the root mean squared error for OLS, and to the minimal cross-validation mean squared error for LASSO and Ridge regressions. False Positive indicates the number of cases in which a non-corrupt firm is classified as corrupt by the model. False Negative indicates the number of cases in which a corrupt firm is classified as non-corrupt by the model. Threshold indicates the predicted value of the outcome variable for which a firm is classified a corrupt. See Appendix C for additional results including precision and recall graphs, and ROC curves.