Correlation Plots

UPDATED - using high detail side by side MTurk results

I begin with simple correlation plots between vote-share and the respective MTurk label value. This is to see whether we can replicate (or even begin to see a similar relationship) figure 1(b) in the Todorov et.al (2005) paper (https://science.sciencemag.org/content/308/5728/1623). For competence I also plot the difference in vote-share (for a more exact comparison).

NOTE Each plot contains both a simple linear regression model and the correlation in red

NOTE The correlation found in Todorov et.al (2005) ranges between 0.37 and 0.44 for the Competence Feature and the difference in vote-share. Their set up is comparable to my second figure.

Single-variable Regressions

I run single and combined variable regression to support the trend findings above. Notice that comptence and trustworthiness are highly significant.

Single Variable Side-By-Side MTurk Regressions
Dependent variable:
Election Outcome (did-win)
(1) (2) (3) (4) (5)
competence 0.703*** 0.509***
(0.505, 0.901) (0.271, 0.746)
trustworthiness 0.592*** 0.324**
(0.405, 0.778) (0.102, 0.547)
dominance 0.267**
(0.054, 0.479)
attractiveness 0.193*
(0.022, 0.365)
Constant 0.149** 0.204*** 0.367*** 0.403*** 0.084
(0.044, 0.253) (0.105, 0.303) (0.255, 0.479) (0.311, 0.496) (-0.030, 0.197)
Observations 548 548 548 548 548
Adjusted R2 0.057 0.046 0.006 0.004 0.065
F Statistic 34.201*** (df = 1; 546) 27.289*** (df = 1; 546) 4.259** (df = 1; 546) 3.427* (df = 1; 546) 20.117*** (df = 2; 545)
Note: p<0.1; p<0.05; p<0.01

Density Plots

Below are conditional density plots between win-loose groups, there is no significant difference. We can see that Competence and Trustworthiness show actual distributional shifts w.r.t the different did-win classes which supports what we saw above.