Table 3.
Average reconstruction error for the different methods over 100 simulated datasets with standard deviations in brackets
| L1 | L2 | |||||||
|---|---|---|---|---|---|---|---|---|
| Method | Sim 1 | Sim 2 | Sim 3 | Sim 4 | Sim 1 | Sim 2 | Sim 3 | Sim 4 |
| SMT | 2.76 (0.94) | 9.62 (1.16) | 41.95 (2.33) | 193.14 (2.64) | 1.20 (0.27) | 2.22 (0.29) | 4.95 (0.29) | 11.80 (0.18) |
| GMT | 3.67 (0.64) | 13.60 (0.58) | 45.68 (1.35) | 174.58 (3.04) | 1.52 (0.11) | 2.67 (0.18) | 5.01 (0.21) | 10.69 (0.21) |
| SGMT | 3.66 (2.05) | 10.05 (1.57) | 39.63 (2.30) | 176.60 (3.64) | 1.21 (0.31) | 2.09 (0.25) | 4.52 (0.27) | 10.85 (0.24) |
| ASSET | 2.88 (0.55) | 10.25 (1.48) | 47.36 (1.99) | 202.60 (1.87) | 1.47 (0.23) | 2.79 (0.26) | 6.04 (0.19) | 12.69 (0.1) |
| GPA | 6.34 (7.45) | 12.33 (3.67) | 49.04 (1.27) | 203.32 (1.13) | 1.87 (0.64) | 3.06 (0.19) | 6.21 (0.11) | 12.73 (0.06) |
The estimated coefficients for the penalised methods correspond to the estimate with tuning parameters chosen from cross validation. The estimated coefficients for ASSET and GPA are set using the summary statistics of the active variables. An active variable for ASSET and GPA was one with a FDR corrected p-value less than 0.05