Here I’m comparing the predictions from the JUST FACE and the FACE + STRUCTURED DATA models. These are called ML Face CNN and ML Face Fusion respectively. The fusion model is a first pass that only includes data on:
The main point here is that the adjusted r-squared for the fusion model is significantly higher than just the face (as seen in table 01). This goes from 0.0332 to 0.1030. It’s also very promising that in table 2, where I’m including the charge data, the gun crime and property crime flags are insignificant with ML Face Fusion. I’m thinking this indicates the model is using that data to make its prediction !
| Dependent variable: | ||
| Release-Final-Outcome | ||
| (1) | (2) | |
| ML-Face Fusion | 1.1126*** | |
| (1.0517, 1.1735) | ||
| ML-Face CNN | 0.6960*** | |
| (0.6263, 0.7656) | ||
| Constant | -0.1120*** | 0.2437*** |
| (-0.1604, -0.0636) | (0.1914, 0.2959) | |
| Observations | 7,851 | 7,851 |
| Adjusted R2 | 0.1030 | 0.0332 |
| F Statistic (df = 1; 7849) | 902.3631*** | 270.3141*** |
| Note: | p<0.1; p<0.05; p<0.01 | |
| † We encode skin-tone as a continuous variable increasing in brightness | ||
| Dependent variable: | ||
| Release-Final-Outcome | ||
| (1) | (2) | |
| SexM | -0.0279** | 0.0025 |
| (-0.0477, -0.0080) | (-0.0185, 0.0235) | |
| RaceB | 0.0360 | 0.0504 |
| (-0.0501, 0.1221) | (-0.0355, 0.1363) | |
| Age | 0.0010 | 0.00004 |
| (-0.00002, 0.0020) | (-0.0010, 0.0010) | |
| † skin-tone | 0.0542*** | 0.0469** |
| (0.0215, 0.0870) | (0.0142, 0.0796) | |
| Attractiveness | -0.0159 | -0.0139 |
| (-0.0488, 0.0170) | (-0.0467, 0.0189) | |
| Competence | 0.0106 | 0.0101 |
| (-0.0223, 0.0436) | (-0.0228, 0.0430) | |
| Dominance | -0.0026 | -0.0002 |
| (-0.0295, 0.0243) | (-0.0271, 0.0266) | |
| Trustworthiness | 0.0312 | 0.0280 |
| (-0.0013, 0.0637) | (-0.0044, 0.0604) | |
| Felony-Dummie | -0.0653*** | -0.1991*** |
| (-0.0924, -0.0382) | (-0.2158, -0.1824) | |
| Gun-Crime-Dummie | 0.0186 | 0.0382** |
| (-0.0112, 0.0484) | (0.0087, 0.0677) | |
| Drug-Crime-Dummie | 0.0437*** | 0.0415*** |
| (0.0234, 0.0640) | (0.0213, 0.0617) | |
| Violent-Crime-Dummie | -0.0685*** | -0.1649*** |
| (-0.1005, -0.0365) | (-0.1920, -0.1377) | |
| Property-Crime-Dummie | 0.0099 | -0.0248** |
| (-0.0090, 0.0288) | (-0.0425, -0.0071) | |
| Risk Prediction | -0.7867*** | -0.7759*** |
| (-0.8970, -0.6764) | (-0.8859, -0.6658) | |
| Well-groomed | 0.0089* | 0.0081 |
| (0.0006, 0.0171) | (-0.0001, 0.0163) | |
| Messy | 0.0032 | 0.0051 |
| (-0.0042, 0.0105) | (-0.0022, 0.0125) | |
| ML-Face Fusion | 0.7738*** | |
| (0.6494, 0.8982) | ||
| ML-Face CNN | 0.5470*** | |
| (0.4710, 0.6229) | ||
| Constant | 0.2921*** | 0.5612*** |
| (0.1180, 0.4662) | (0.4200, 0.7024) | |
| Observations | 7,851 | 7,851 |
| Adjusted R2 | 0.1243 | 0.1282 |
| F Statistic (df = 20; 7830) | 56.6999*** | 58.7155*** |
| Note: | p<0.1; p<0.05; p<0.01 | |
| † We encode skin-tone as a continuous variable increasing in brightness | ||