```r
condition topic p_completed n_started n_completed
<char> <char> <num> <int> <num>
1: persuadeHarris personality 0.9128368 631 576
2: persuadeHarris policy 0.9001536 651 586
3: persuadeTrump personality 0.9251248 601 556
4: persuadeTrump policy 0.9209677 620 571
```
predict attrit status (1: attrit, 0: complete) with different logistic regression specifications
```r
# attrit ~ condition
> summ(feglm(attrit ~ condition, d0, family = binomial))
term result
<char> <char>
1: (Intercept) b = -2.27 (0.10) [-2.46, -2.08] p < .001
2: conditionpersuadeTrump b = -0.21 (0.14) [-0.50, 0.07] p = .138
# initial lean/preference
> summ(feglm(attrit ~ lean_bidentrump_1, d0, family = binomial))
term result
<char> <char>
1: (Intercept) b = -2.40 (0.10) [-2.61, -2.20] p < .001
2: lean_bidentrump_1 b = 0.00 (0.00) [0.00, 0.00] p = .967
# topic: personality or policy
> summ(feglm(attrit ~ topic, d0, family = binomial))
term result
<char> <char>
1: (Intercept) b = -2.43 (0.10) [-2.63, -2.22] p < .001
2: topicpolicy b = 0.11 (0.14) [-0.17, 0.39] p = .446
# interactions
> summ(feglm(attrit ~ condition * lean_bidentrump_1Z * topicZ, d0, family = binomial))
term result
<char> <char>
1: (Intercept) b = -2.29 (0.10) [-2.48, -2.10] p < .001
2: conditionpersuadeTrump b = -0.24 (0.15) [-0.53, 0.05] p = .103
3: lean_bidentrump_1Z b = -0.05 (0.10) [-0.24, 0.14] p = .637
4: topicZ b = 0.08 (0.10) [-0.11, 0.27] p = .432
5: conditionpersuadeTrump × lean_bidentrump_1Z b = 0.11 (0.15) [-0.18, 0.39] p = .473
6: conditionpersuadeTrump × topicZ b = -0.04 (0.15) [-0.33, 0.24] p = .760
7: lean_bidentrump_1Z × topicZ b = 0.02 (0.10) [-0.17, 0.21] p = .842
8: conditionpersuadeTrump × lean_bidentrump_1Z × topicZ b = -0.02 (0.15) [-0.31, 0.27] p = .892
```
among those who completed, treatment did not predict initial preference
```r
> summ(feols(lean_bidentrump_1 ~ condition, d0[completed == 1]))
term result
<char> <char>
1: (Intercept) b = 45.26 (1.25) [42.81, 47.71] p < .001
2: conditionpersuadeTrump b = -2.10 (1.78) [-5.59, 1.39] p = .237
```