- trump harris persuasion study
- 8k AI messages
- 27 strategies (16 generated by LLM + 11 from [Luke's paper](https://www.cambridge.org/core/journals/american-political-science-review/article/how-experiments-help-campaigns-persuade-voters-evidence-from-a-large-archive-of-campaigns-own-experiments/FF5BE6ED1553475F8321F7C4209357F7))
- see [[250131_112806 consolidated strategies v2|this doc for strategies]]
- asked model to rate usage of each strategy
- ratings: 0, 0.5, 1
# overall strategy use frequency across 8k AI messages
![[1738631541.png]]
# strategy use for each condition
## mean ratings
![[1738630974.png]]
# PCA
![[20250203181938.png]]
most features load onto PC1; otherwise, it's messy (see below - after the heatmap for more)
- pc2: negativity?
- pc3: elaboration, deliberation, deep listening?
- pc4: storytelling, anecdotes?
![[1738630510.png]]
top 5 and bottom 5 features for each component
```r
[1] "PC1" everything component
feature component value feature_type
<fctr> <fctr> <num> <char>
1: positive_framing_and_value_alignment PC1 0.7788080 top5
2: emotional_appeal_with_balanced_urgency PC1 0.7737796 top5
3: gradual_persuasion PC1 0.7659196 top5
4: politeness_and_civil_tone PC1 0.7179453 top5
5: transfer_of_association PC1 0.6760987 top5
6: social_proof_and_normative_influence PC1 0.2447723 bottom5
7: stimulate_anger PC1 0.2015709 bottom5
8: name-calling PC1 0.1830510 bottom5
9: use_of_positive_testimonials PC1 0.1198454 bottom5
10: use_of_negative_testimonials PC1 0.0872558 bottom5
[1] "PC2" negativity, trump component
feature component value feature_type
<fctr> <fctr> <num> <char>
1: negative_tone PC2 0.7266798 top5
2: name-calling PC2 0.5960550 top5
3: stimulate_anger PC2 0.5615600 top5
4: contrastive_tone PC2 0.5408932 top5
5: use_of_negative_testimonials PC2 0.4397736 top5
6: localized_focus PC2 -0.2385239 bottom5
7: positive_tone PC2 -0.3409390 bottom5
8: reciprocity_and_mutual_benefit PC2 -0.3453441 bottom5
9: stimulate_enthusiasm PC2 -0.3652947 bottom5
10: build_rapport_and_common_ground PC2 -0.3983873 bottom5
[1] "PC3" understanding, listening, empathy
feature component value feature_type
<fctr> <fctr> <num> <char>
1: cognitive_elaboration PC3 0.6032904 top5
2: active_listening_and_empathy PC3 0.5542455 top5
3: address_objections_and_counterarguments PC3 0.4205270 top5
4: reciprocity_and_mutual_benefit PC3 0.3660916 top5
5: gradual_persuasion PC3 0.2997194 top5
6: positive_tone PC3 -0.3271521 bottom5
7: transfer_of_association PC3 -0.4669102 bottom5
8: encourage_action_with_clear_calls PC3 -0.5314675 bottom5
9: stimulate_enthusiasm PC3 -0.5606148 bottom5
10: aggressive_and_explicit_directives PC3 -0.6410724 bottom5
[1] "PC4" story telling, anecdotes
feature component value feature_type
<fctr> <fctr> <num> <char>
1: use_of_everyday_people_as_messengers PC4 0.6153169 top5
2: localized_focus PC4 0.4594400 top5
3: use_of_positive_testimonials PC4 0.3799987 top5
4: social_proof_and_normative_influence PC4 0.3463055 top5
5: storytelling_and_relatable_anecdotes PC4 0.2788590 top5
6: cognitive_elaboration PC4 -0.1246677 bottom5
7: gradual_persuasion PC4 -0.2553949 bottom5
8: positive_framing_and_value_alignment PC4 -0.2591820 bottom5
9: positive_tone PC4 -0.2733063 bottom5
10: politeness_and_civil_tone PC4 -0.3451707 bottom5
[1] "PC5" building rapport with negativity
feature component value feature_type
<fctr> <fctr> <num> <char>
1: name-calling PC5 0.4403572 top5
2: build_rapport_and_common_ground PC5 0.4116369 top5
3: use_of_negative_testimonials PC5 0.4103404 top5
4: active_listening_and_empathy PC5 0.3469322 top5
5: negative_tone PC5 0.1860533 top5
6: aggressive_and_explicit_directives PC5 -0.1601957 bottom5
7: address_objections_and_counterarguments PC5 -0.2343283 bottom5
8: localized_focus PC5 -0.2908333 bottom5
9: relatable_hypotheticals PC5 -0.3414939 bottom5
10: evidence-_or_fact-based_arguments PC5 -0.4358370 bottom5
[1] "PC6"
feature component value feature_type
<fctr> <fctr> <num> <char>
1: use_of_positive_testimonials PC6 0.6931575 top5
2: use_of_everyday_people_as_messengers PC6 0.2579346 top5
3: contrastive_tone PC6 0.1850435 top5
4: evidence-_or_fact-based_arguments PC6 0.1733547 top5
5: positive_tone PC6 0.1579372 top5
6: encourage_action_with_clear_calls PC6 -0.1933386 bottom5
7: audience_adaptation PC6 -0.2024014 bottom5
8: reciprocity_and_mutual_benefit PC6 -0.2227717 bottom5
9: localized_focus PC6 -0.2771537 bottom5
10: stimulate_anger PC6 -0.3192187 bottom5
[1] "PC7"
feature component value feature_type
<fctr> <fctr> <num> <char>
1: social_proof_and_normative_influence PC7 0.6523818 top5
2: evidence-_or_fact-based_arguments PC7 0.2052370 top5
3: stimulate_anger PC7 0.1860397 top5
4: encourage_action_with_clear_calls PC7 0.1536522 top5
5: aggressive_and_explicit_directives PC7 0.1203791 top5
6: transfer_of_association PC7 -0.1185417 bottom5
7: use_of_everyday_people_as_messengers PC7 -0.1314816 bottom5
8: name-calling PC7 -0.1319323 bottom5
9: relatable_hypotheticals PC7 -0.2105588 bottom5
10: storytelling_and_relatable_anecdotes PC7 -0.4576658 bottom5
[1] "PC8"
feature component value feature_type
<fctr> <fctr> <num> <char>
1: use_of_negative_testimonials PC8 0.5949265 top5
2: localized_focus PC8 0.3207318 top5
3: evidence-_or_fact-based_arguments PC8 0.2294022 top5
4: positive_tone PC8 0.1725357 top5
5: stimulate_enthusiasm PC8 0.1176925 top5
6: build_rapport_and_common_ground PC8 -0.1371894 bottom5
7: audience_adaptation PC8 -0.1687177 bottom5
8: aggressive_and_explicit_directives PC8 -0.1960477 bottom5
9: active_listening_and_empathy PC8 -0.2113021 bottom5
10: stimulate_anger PC8 -0.2825027 bottom5
[1] "PC9"
feature component value feature_type
<fctr> <fctr> <num> <char>
1: use_of_everyday_people_as_messengers PC9 0.32866571 top5
2: audience_adaptation PC9 0.31429251 top5
3: localized_focus PC9 0.17049784 top5
4: use_of_positive_testimonials PC9 0.12704508 top5
5: evidence-_or_fact-based_arguments PC9 0.11078176 top5
6: emotional_appeal_with_balanced_urgency PC9 -0.06383895 bottom5
7: name-calling PC9 -0.13359499 bottom5
8: relatable_hypotheticals PC9 -0.22299481 bottom5
9: social_proof_and_normative_influence PC9 -0.43092697 bottom5
10: storytelling_and_relatable_anecdotes PC9 -0.43336313 bottom5
[1] "PC10"
feature component value feature_type
<fctr> <fctr> <num> <char>
1: use_of_negative_testimonials PC10 0.3913890 top5
2: use_of_positive_testimonials PC10 0.3541201 top5
3: aggressive_and_explicit_directives PC10 0.2760644 top5
4: encourage_action_with_clear_calls PC10 0.1805196 top5
5: storytelling_and_relatable_anecdotes PC10 0.1671059 top5
6: localized_focus PC10 -0.1874561 bottom5
7: negative_tone PC10 -0.1879528 bottom5
8: name-calling PC10 -0.2087545 bottom5
9: social_proof_and_normative_influence PC10 -0.2115479 bottom5
10: use_of_everyday_people_as_messengers PC10 -0.2260449 bottom5
[1] "PC11"
feature component value feature_type
<fctr> <fctr> <num> <char>
1: use_of_everyday_people_as_messengers PC11 0.4156124 top5
2: aggressive_and_explicit_directives PC11 0.1905905 top5
3: use_of_negative_testimonials PC11 0.1851716 top5
4: social_proof_and_normative_influence PC11 0.1657042 top5
5: contrastive_tone PC11 0.0985911 top5
6: name-calling PC11 -0.1059363 bottom5
7: emotional_appeal_with_balanced_urgency PC11 -0.1221705 bottom5
8: stimulate_anger PC11 -0.2662746 bottom5
9: localized_focus PC11 -0.2869204 bottom5
10: use_of_positive_testimonials PC11 -0.3751052 bottom5
[1] "PC12"
feature component value feature_type
<fctr> <fctr> <num> <char>
1: stimulate_anger PC12 0.3695700 top5
2: use_of_everyday_people_as_messengers PC12 0.2742338 top5
3: politeness_and_civil_tone PC12 0.1704799 top5
4: emotional_appeal_with_balanced_urgency PC12 0.1449118 top5
5: positive_framing_and_value_alignment PC12 0.1351281 top5
6: name-calling PC12 -0.1518101 bottom5
7: reciprocity_and_mutual_benefit PC12 -0.1594460 bottom5
8: build_rapport_and_common_ground PC12 -0.1638914 bottom5
9: contrastive_tone PC12 -0.1798421 bottom5
10: audience_adaptation PC12 -0.3571513 bottom5
[1] "PC13"
feature component value feature_type
<fctr> <fctr> <num> <char>
1: reciprocity_and_mutual_benefit PC13 0.4408247 top5
2: aggressive_and_explicit_directives PC13 0.2106336 top5
3: name-calling PC13 0.2070898 top5
4: encourage_action_with_clear_calls PC13 0.1995851 top5
5: use_of_everyday_people_as_messengers PC13 0.1143203 top5
6: social_proof_and_normative_influence PC13 -0.1253331 bottom5
7: transfer_of_association PC13 -0.1340864 bottom5
8: use_of_negative_testimonials PC13 -0.1444289 bottom5
9: stimulate_anger PC13 -0.1695354 bottom5
10: audience_adaptation PC13 -0.2907530 bottom5
[1] "PC14"
feature component value feature_type
<fctr> <fctr> <num> <char>
1: reciprocity_and_mutual_benefit PC14 0.33447281 top5
2: contrastive_tone PC14 0.25491711 top5
3: stimulate_anger PC14 0.20707721 top5
4: positive_tone PC14 0.15449566 top5
5: use_of_negative_testimonials PC14 0.09828821 top5
6: audience_adaptation PC14 -0.11482166 bottom5
7: politeness_and_civil_tone PC14 -0.12296126 bottom5
8: localized_focus PC14 -0.14321321 bottom5
9: gradual_persuasion PC14 -0.15375755 bottom5
10: name-calling PC14 -0.32077062 bottom5
```