Table 3 Top 15 VIM values for 314 variables used to predict change to Annalena Baerbock (sorted for 1 = change present).

From: How to convince in a televised debate: the application of machine learning to analyze why viewers changed their winner perception during the 2021 German chancellor discussion

Variable name

all

0

1

Variable explanation

pre.victor

0.1831233

0.4437236

0.6504343

Pre-victor expectation

pre.chancellor.ab

0.0383334

0.0898314

0.1460049

Pre-debate chancellor preference: Annalena Baerbock

pre.ab.comp

0.0075872

0.0107126

0.0517047

Pre-debate competence evaluation: Annalena Baerbock

pre.ab.cred

0.0055469

0.0091875

0.0334266

Pre-debate credibility evaluation: Annalena Baerbock

pre.chancellor.al

0.0047994

0.0072426

0.0312141

Pre-debate chancellor preference: Armin Laschet

V204

0.0039255

0.0079897

0.0188561

Debate moment concerning lack of climate protection (Speaker: AB)

V261

0.0011112

−0.0004507

0.0140922

Debate moment concerning immigration laws (Speaker: AL)

pre.al.comp

0.0014602

0.0016733

0.0112020

Pre-debate competence evaluation: Armin Laschet

pre.ab.symp

0.0021647

0.0043406

0.0106040

Pre-debate competence evaluation: Annalena Baerbock

V137

0.0018666

0.0040751

0.0080726

Debate moment concerning digitalization and central coordination of future topics (Speaker: AB)

pre.os.comp

0.0038237

0.0111571

0.0074747

Pre-debate competence evaluation: Olaf Scholz

V288

0.0017317

0.0040566

0.0065976

Debate moment Laschet addressing the voters directly about trust and making Germany strong (Speaker: AL)

V202

0.0016074

0.0037602

0.0061392

Debate moment Baerbock says the next government decisive to limit the damages from climate change (AB)

pre.age

0.0000877

−0.0012658

0.0050828

Pre-debate data: Participants age

V260

0.0004259

0.0000803

0.0045845

Debate moment concerning the stabilization of the pension system (Speaker: AB)

  1. Variable importance corresponds with significance in terms of determining the overall impact of a variable on the prediction model. VIMs are determined by excluding and computing models once without each individual feature (Out-of-bag or OOB) and measuring the difference in predictive quality compared to the full model. – precise text of each speech moment is available in the Annex (in German and English).
Back to article page