One Sentence Summary

I now include a control for the quality of the election images, which reduces us to an 800 image subset. We see that MTurk signal increases from 0.012 to 0.015.

Outline

This markdown contains table 01s for elections with an image quality control. These are based on the lower detail side-by-side MTurk labels which have on average 4 voters per image. This is because including the image quality control decreases our data-set significantly and if I used high detail MTurk labels we would not have enough observations. I present a Table 01 Configuration 4 (using own-party-vote-share-jacknife in the election_lm).

NOTE I am now including a control for image quality in the less MTurk voters columns. This restricts our data to 800 observations including some with no image quality labels. With the control for image quality, taking those with above average quality, we see an increase in the MTurk label signal from 0.012 to 0.015.

Definitions

Before jumping into regression output, I will put all definitions of important terms here:

  1. Vote-Share : In all tables below, when vote-share is a LHS variable this means the vote-share of a specific candidate election. I.e. how much of the total vote did a specific candidate get in their election.

  2. Own-Party-Vote-Share-Jacknife : This is a feature and computed as the “average vote share of my party in this state over all years excluding the current”

Table 1 - Configuration 4 - Comparison between High/Low detail

Replacing own_party_win_rate with own_party_vote_share_jacknife in the election lm. This now includes:

  1. Party (limited to Democrat vs. Republican)
  2. Vote-share

Own Party Vote-share-Jacknife is computed as: “average vote share of my party in this state over all years excluding the current”

Less Voters are using regression output from the previous survey in which we have between 3-4 voters per image, but over 1,300 observations. This now includes an image quality control an includes 800 above average quality pairs.

Table 01 - Election Regressions - Comparing High/Low Detail MTurk
Fit measured in adjusted R squared and AUC
Model Configuration Election Outcome Vote Share
R-Sqrt Less Voters AUC Less Voters R-Sqrt Less Voters
Single Variable Model
Election LM 0.0929 0.6746 0.0809
Lower 95% C.I. 0.0632 0.6373 0.0531
Upper 95% C.I. 0.1272 0.7119 0.1136
Sex 0.0002 0.5082 0.0041
−0.0012 0.4802 −0.0011
0.0069 0.5363 0.0153
Skine-Tone −0.0100 0.5479 −0.0114
−0.0040 0.5087 −0.0042
0.0277 0.5871 0.0295
MTurk Features 0.0150 0.5752 0.0033
0.0044 0.5357 −0.0015
0.0410 0.6147 0.0212
P_hat_cnn 0.1111 0.6964 0.0960
0.0794 0.6603 0.0648
0.1476 0.7325 0.1309
Combined Variable Model
P_hat_cnn + MTurk Features 0.1163 0.7020 0.0948
0.0885 0.6661 0.0664
0.1611 0.7379 0.1340
Election LM + P_hat_cnn 0.1899 0.7570 0.1680
0.1523 0.7239 0.1309
0.2317 0.7901 0.2102
Election LM + Sex 0.0926 0.6768 0.0845
0.0638 0.6397 0.0571
0.1270 0.7139 0.1197
Election LM + Sex + P_hat_cnn 0.1884 0.7574 0.1696
0.1547 0.7244 0.1343
0.2315 0.7905 0.2146
Election LM + Sex + Skin-Tone 0.0843 0.6878 0.0741
0.0748 0.6512 0.0650
0.1448 0.7244 0.1329
Election LM + Sex + Skin-Tone + P_hat_cnn 0.1803 0.7638 0.1596
0.1580 0.7310 0.1383
0.2444 0.7965 0.2282
Election LM + Sex + Skin-Tone + MTurk 0.1005 0.7026 0.0778
0.0915 0.6669 0.0728
0.1677 0.7384 0.1417
Election LM + Sex + Skin-Tone + MTurk + P_hat_cnn 0.1864 0.7693 0.1599
0.1680 0.7370 0.1445
0.2506 0.8016 0.2295