Baseline Table 01 - Version 1

NOTE In this table, the election_lm is glm using:

  1. Election year
  2. Party affiliation
  3. State

The variable MTurk contains our 4 MTurk labels, skin_tone consists of 18 skin-tones, P_hat_cnn is the output from the election_cnn.

In this version of table 1 I report both adjusted-\(R^2\) and ROC-AUC

Table 01 - Election Regressions
Fit measured in adjusted R squared
Model Configuration Adjusted R Squared ROC AUC
Single Variable Model
Election LM 0.0160 0.5650278
Lower 95% C.I. 0.0076 0.5348018
Upper 95% C.I. 0.0282 0.5952538
Sex 0.0133 0.5568305
0.0066 0.5317242
0.0271 0.5819367
Skine-Tone −0.0020 0.5505362
0.0124 0.5205064
0.0343 0.5805660
MTurk Features 0.0013 0.5297182
0.0018 0.4987869
0.0145 0.5606496
P_hat_cnn 0.1017 0.6882254
0.0782 0.6601295
0.1289 0.7163214
Combined Variable Model
Election LM + Sex 0.0236 0.5895029
0.0145 0.5593947
0.0416 0.6196112
Election LM + Sex + P_hat_cnn 0.1047 0.6916630
0.0841 0.6637147
0.1354 0.7196112
Election LM + Sex + Skin-Tone 0.0204 0.6071566
0.0323 0.5773785
0.0656 0.6369348
Election LM + Sex + Skin-Tone + P_hat_cnn 0.1026 0.7009157
0.1024 0.6731509
0.1567 0.7286806
Election LM + Sex + Skin-Tone + MTurk 0.0207 0.6083600
0.0375 0.5785359
0.0727 0.6381840
Election LM + Sex + Skin-Tone + MTurk + P_hat_cnn 0.1025 0.7035807
0.1087 0.6759005
0.1624 0.7312608

Visualization

Below I plot the different model configurations and their respective r_squared and auc values together with their confidence intervals. I’m having trouble adding a legend, the green series in AUC and the red series is R-sqrt.

Baseline Table 01 - Version 2

This table summarizes the results from a new parameterization of the election_lm as well as the inclusion of two new features.

  1. Election LM now includes a state*party interaction
  2. Win_Rate_Prior captures a parties win record in that state over the years prior to the current (i.e only the past)
  3. Win_Rate_Total captures a parties win record over all years excluding the current (i.e. past and future)

I include these two new variables as controls and use the new election_lm as a benchmark below:

Table 01 - Version 02 - Election Regressions
Fit measured in adjusted R squared and AUC
Model Configuration Adjusted R Squared ROC AUC
Single Variable Model
Election LM 0.0692 0.6608260
Lower 95% C.I. 0.0494 0.6320236
Upper 95% C.I. 0.0946 0.6896285
Win_Rate_Prior −0.0004 0.5112823
0.0000 0.4810200
0.0031 0.5415446
Win_Rate_Total −0.0007 0.5084073
0.0000 0.4779351
0.0029 0.5388796
Sex 0.0133 0.5568305
0.0061 0.5317242
0.0266 0.5819367
Skine-Tone −0.0020 0.5505362
0.0126 0.5205064
0.0344 0.5805660
MTurk Features 0.0013 0.5297182
0.0018 0.4987869
0.0146 0.5606496
P_hat_cnn 0.1017 0.6882254
0.0786 0.6601295
0.1296 0.7163214
Combined Variable Model
Election LM + Win_Rate_Prior 0.0705 0.6588415
0.0526 0.6298820
0.0967 0.6878010
Election LM + Win_Rate_Total 0.0743 0.6616854
0.0552 0.6328183
0.1004 0.6905525
Election LM + Win_Rate_Prior + Win_Rate_Total 0.0739 0.6614711
0.0571 0.6326084
0.1014 0.6903338
+ Sex 0.0781 0.6679679
0.0623 0.6393015
0.1078 0.6966343
Election LM + Win Rates + Sex + P_hat_cnn 0.1445 0.7270625
0.1224 0.7003961
0.1799 0.7537289
Election LM + Win Rates + Sex + Skin-Tone 0.0734 0.6729988
0.0794 0.6445690
0.1277 0.7014286
Election LM + Win Rates + Sex + Skin-Tone + P_hat_cnn 0.1416 0.7312276
0.1417 0.7047674
0.1998 0.7576878
Election LM + Win Rates + Sex + Skin-Tone + MTurk 0.0735 0.6757151
0.0837 0.6473659
0.1332 0.7040644
Election LM + Win Rates + Sex + Skin-Tone + MTurk + P_hat_cnn 0.1414 0.7325317
0.1454 0.7061260
0.2048 0.7589373