# zero-one-inflated beta (zoib) bayesian regression models
for simplicity (faster fitting), dropped conversation focus
model spec: t3 ~ condition * t1Z
## candidate preference
```r
m0_brm <- brm(
bf(
I(lean_bidentrump_3 / 100) ~ condition * scale(lean_bidentrump_1),
phi ~ condition * scale(lean_bidentrump_1),
zoi ~ condition * scale(lean_bidentrump_1),
coi ~ condition * scale(lean_bidentrump_1)
),
data = d0,
family = zero_one_inflated_beta(),
chains = 8,
iter = 2000,
cores = 8
)
m0_brm
summ(m0_brm)
term result p_sig
<char> <glue> <char>
1: b_Intercept b = -0.27 [-0.36, -0.18] *
2: b_phi_Intercept b = 1.80 [1.67, 1.93] *
3: b_zoi_Intercept b = 0.24 [0.12, 0.37] *
4: b_coi_Intercept b = 0.01 [-1.69, 1.79]
# mu of beta dist (logit scale)
5: b_conditionproTrump b = 0.15 [0.02, 0.28] * # significant treatment effect
6: b_scalelean_bidentrump_1 b = 1.56 [1.46, 1.67] *
7: b_conditionproTrump:scalelean_bidentrump_1 b = -0.15 [-0.31, 0.01]
# precision of beta dist (log scale)
8: b_phi_conditionproTrump b = -0.15 [-0.34, 0.04]
9: b_phi_scalelean_bidentrump_1 b = 0.12 [-0.03, 0.27]
10: b_phi_conditionproTrump:scalelean_bidentrump_1 b = -0.04 [-0.26, 0.17]
# whether outcome is extreme (0/1) or not extreme (not [0/1]) (logit scale)
11: b_zoi_conditionproTrump b = 0.10 [-0.08, 0.28]
12: b_zoi_scalelean_bidentrump_1 b = -0.24 [-0.36, -0.11] *
13: b_zoi_conditionproTrump:scalelean_bidentrump_1 b = -0.17 [-0.35, 0.00]
# if it's 0 or 1, is it 1? (standard logistic) (logit scale)
14: b_coi_conditionproTrump b = -0.55 [-2.82, 1.68]
15: b_coi_scalelean_bidentrump_1 b = 8.42 [5.33, 13.89] *
16: b_coi_conditionproTrump:scalelean_bidentrump_1 b = -1.66 [-7.51, 2.69]
```
## vote likelihood
harris supporters
- for people who were closer to 100% vote pretreatment (but not at 100%), the treatment was more likely to make them move to the extreme (100% vote in follow-up)
```r
term result p_sig
<char> <glue> <char>
1: b_Intercept b = 1.42 [1.23, 1.61] *
2: b_phi_Intercept b = 1.58 [1.31, 1.83] *
3: b_zoi_Intercept b = 0.85 [0.64, 1.07] *
4: b_coi_Intercept b = 6.79 [4.23, 11.19] *
# mu of beta dist (logit scale)241120_205234
5: b_conditionHarrisproHarris b = 0.02 [-0.24, 0.28]
6: b_scalevote_chance_1 b = 0.74 [0.61, 0.87] *
7: b_conditionHarrisproHarris:scalevote_chance_1 b = -0.06 [-0.25, 0.12]
# precision of beta dist (log scale)
8: b_phi_conditionHarrisproHarris b = 0.06 [-0.30, 0.43]
9: b_phi_scalevote_chance_1 b = 0.16 [0.00, 0.32] *
10: b_phi_conditionHarrisproHarris:scalevote_chance_1 b = 0.15 [-0.08, 0.38]
# whether outcome is extreme (0/1) or not extreme (not [0/1]) (logit scale)
11: b_zoi_conditionHarrisproHarris b = -0.20 [-0.52, 0.12]
12: b_zoi_scalevote_chance_1 b = 1.22 [0.96, 1.52] *
13: b_zoi_conditionHarrisproHarris:scalevote_chance_1 b = 0.63 [0.15, 1.14] * # effective at moving those who were more likely to vote pre-treatment to the extreme
# if it's 0 or 1, is it 1? (standard logistic) (logit scale)
14: b_coi_conditionHarrisproHarris b = -1.68 [-6.35, 1.58]
15: b_coi_scalevote_chance_1 b = 3.25 [1.99, 5.40] *
16: b_coi_conditionHarrisproHarris:scalevote_chance_1 b = -1.04 [-3.35, 0.63]
```
trump supporters
```r
term result p_sig
<char> <glue> <char>
1: b_Intercept b = 1.31 [1.14, 1.48] *
2: b_phi_Intercept b = 1.59 [1.34, 1.82] *
3: b_zoi_Intercept b = 0.51 [0.30, 0.72] *
4: b_coi_Intercept b = 3.93 [2.95, 5.23] *
5: b_conditionproTrump b = 0.14 [-0.10, 0.38]
6: b_scalevote_chance_1 b = 0.74 [0.58, 0.89] *
7: b_conditionproTrump:scalevote_chance_1 b = 0.05 [-0.17, 0.27]
8: b_phi_conditionproTrump b = 0.17 [-0.18, 0.52]
9: b_phi_scalevote_chance_1 b = 0.26 [0.07, 0.46] *
10: b_phi_conditionproTrump:scalevote_chance_1 b = 0.13 [-0.15, 0.39]
1
1: b_zoi_conditionproTrump b = -0.10 [-0.40, 0.21]
12: b_zoi_scalevote_chance_1 b = 0.90 [0.65, 1.18] *
13: b_zoi_conditionproTrump:scalevote_chance_1 b = 0.14 [-0.25, 0.52]
14: b_coi_conditionproTrump b = 1.04 [-1.13, 3.97]
15: b_coi_scalevote_chance_1 b = 2.01 [1.42, 2.72] *
16: b_coi_conditionproTrump:scalevote_chance_1 b = 1.00 [-0.30, 2.69]
```