Background

Over the course of this project, we collected quite a few responses to prospective-taking prompts. We did it in different ways and with different checks for response quality. This is the breakdown:
Study 1: Prolific, Neutral PT task, paste enabled, no off-task mouse-tracker (so no flagged responses)
Study 2: Connect, Neutral PT task, paste enabled, no off-task mouse-tracker (so no flagged responses)
Study 3: Connect, PT-0 and PT-10, paste enabled, no off-task mouse-tracker (so no flagged responses)
Study 4: Connect, PT-10, paste disabled, off-task mouse-tracker
Study 5: Connect, PT-10, paste disabled, off-task mouse-tracker, competitive/cooperative framing manipulation

My approach was to clean the open responses by getting rid of all the filler words, stop words, and punctuation. And then, i got vectors for each word (based on co-occurrence on wikipedia), took the cosine similarity of each vector with the vectors of “self” and “other,” and took the mean score of those cosine similarity scores per every response.

Here, I’ll see if those similarity scores are predictive of amount forwarded.

Distributions

First, what are the distributions of those similarity scores for each type of PT task?

hmm, ok. Let’s see some correlations

Analysis

Correlation plot

uh oh. looks like we’re gonna get a big nothing-burger. Especially because self and other are so highly correlated (they should be negatively correlated). I’m guessing part of is just the nature of those words. Like, in this context, they might mean different things, but compared to all of the words in the dictionary, they’re actually pretty similar in meaning. That might be a problem.

I will note the only thing here that might be interesting: sender compassion is negatively associated with cosine_self. Pretty weak, but pretty cool.

linear models predictin amount forwarded

cosine_self as predictor

Neutral PT

(#tab:unnamed-chunk-4)
**
Predictor \(b\) 95% CI \(t\) \(\mathit{df}\) \(p\)
Intercept 6.05 [4.13, 7.98] 6.20 206 < .001
Cosine self -0.32 [-11.22, 10.59] -0.06 206 .954

PT-10

(#tab:unnamed-chunk-5)
**
Predictor \(b\) 95% CI \(t\) \(\mathit{df}\) \(p\)
Intercept 5.50 [3.88, 7.11] 6.70 312 < .001
Cosine self 3.76 [-5.88, 13.39] 0.77 312 .444

PT-0

(#tab:unnamed-chunk-6)
**
Predictor \(b\) 95% CI \(t\) \(\mathit{df}\) \(p\)
Intercept 4.78 [2.15, 7.40] 3.62 91 < .001
Cosine self 2.53 [-14.50, 19.57] 0.30 91 .769

cosine_other as predictor

Neutral PT

(#tab:unnamed-chunk-7)
**
Predictor \(b\) 95% CI \(t\) \(\mathit{df}\) \(p\)
Intercept 4.95 [3.31, 6.60] 5.94 206 < .001
Cosine other 4.74 [-2.40, 11.87] 1.31 206 .192

PT-10

(#tab:unnamed-chunk-8)
**
Predictor \(b\) 95% CI \(t\) \(\mathit{df}\) \(p\)
Intercept 5.90 [4.63, 7.18] 9.12 312 < .001
Cosine other 1.03 [-5.16, 7.22] 0.33 312 .743

PT-0

(#tab:unnamed-chunk-9)
**
Predictor \(b\) 95% CI \(t\) \(\mathit{df}\) \(p\)
Intercept 4.72 [2.53, 6.91] 4.29 91 < .001
Cosine other 2.27 [-8.58, 13.11] 0.41 91 .679