```r
users: 514437
condition N
1: c 257133
2: t 257304
period days
1: t0 24
2: t1 7
clustid tweets_in_clust
1: 0 2052740
2: 4 1100140
3: 6 189868
4: 7 19619
5: 8 681887
6: 9 1690166
7: 17 962410
```
# models
quasipoisson count model
```r
# all tweets
GLM estimation, family = quasipoisson, Dep. Var.: t1
Observations: 3,545,815
Fixed-effects: block: 12,287, day: 7
Standard-errors: Clustered (block)
Estimate Std. Error t value Pr(>|t|)
conditionC -0.000864 0.006680 -0.129350 0.89708 # negative estimate but n.s.
t0SC 0.490237 0.000883 555.105816 < 2.2e-16 ***
conditionC:t0SC -0.000951 0.001260 -0.754479 0.45058
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Squared Cor.: 0.570692
# retweets only
GLM estimation, family = quasipoisson, Dep. Var.: t1
Observations: 3,521,623
Fixed-effects: block: 12,001, day: 7
Standard-errors: Clustered (block)
Estimate Std. Error t value Pr(>|t|)
conditionC -0.007949 0.007673 -1.035887 0.30028 # negative estimate but n.s.
t0SC 0.500899 0.000986 508.239614 < 2.2e-16 ***
conditionC:t0SC -0.000278 0.001326 -0.209793 0.83383
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Squared Cor.: 0.593457
# retweets only; without interaction
> feglm(t1 ~ conditionC + t0SC | block + day, d3, family = "quasipoisson", cluster = "block")
NOTE: 1,202/0 fixed-effects (79,436 observations) removed because of only 0 outcomes.
GLM estimation, family = quasipoisson, Dep. Var.: t1
Observations: 3,521,623
Fixed-effects: block: 12,001, day: 7
Standard-errors: Clustered (block)
Estimate Std. Error t value Pr(>|t|)
conditionC -0.00969 0.006400 -1.51408 0.13003
t0SC 0.50090 0.000986 508.26738 < 2.2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Squared Cor.: 0.593459
```
linear probability model
```r
# all tweets
OLS estimation, Dep. Var.: t1bin
Observations: 3,601,059
Fixed-effects: block: 13,203, day: 7
Standard-errors: Clustered (block)
Estimate Std. Error t value Pr(>|t|)
conditionC 0.000726 0.000498 1.45912 0.14456
t0binSC 0.206638 0.000912 226.59257 < 2.2e-16 ***
conditionC:t0binSC 0.001174 0.001017 1.15444 0.24834
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
RMSE: 0.305573 Adj. R2: 0.138594
Within R2: 0.095825
# retweets only
OLS estimation, Dep. Var.: t1bin
Observations: 3,601,059
Fixed-effects: block: 13,203, day: 7
Standard-errors: Clustered (block)
Estimate Std. Error t value Pr(>|t|)
conditionC 0.000299 0.000561 0.532966 0.59407
t0binSC 0.206838 0.000970 213.147799 < 2.2e-16 ***
conditionC:t0binSC 0.000725 0.001138 0.637152 0.52404
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
RMSE: 0.274783 Adj. R2: 0.155288
Within R2: 0.113745
```
logistic regression
```r
# all tweets
GLM estimation, family = binomial, Dep. Var.: t1bin
Observations: 3,545,787
Fixed-effects: block: 12,286, day: 7
Standard-errors: Clustered (block)
Estimate Std. Error t value Pr(>|t|)
conditionC -0.001327 0.009011 -0.147305 0.88289 # negative estimate but n.s.
t0binSC 3.233058 0.010569 305.892519 < 2.2e-16 ***
conditionC:t0binSC 0.018268 0.018180 1.004813 0.31499
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Log-Likelihood: -1,050,815.9 Adj. Pseudo R2: 0.208951
BIC: 2,287,056.1 Squared Cor.: 0.160118
# retweets only
GLM estimation, family = binomial, Dep. Var.: t1bin
Observations: 3,521,595
Fixed-effects: block: 12,000, day: 7
Standard-errors: Clustered (block)
Estimate Std. Error t value Pr(>|t|)
conditionC -0.011487 0.009646 -1.19089 0.23370 # negative estimate but n.s.
t0binSC 3.531437 0.011169 316.18687 < 2.2e-16 ***
conditionC:t0binSC 0.031746 0.019512 1.62699 0.10374
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Log-Likelihood: -853,721.1 Adj. Pseudo R2: 0.253249
BIC: 1,888,471.0 Squared Cor.: 0.182951
```
# descriptives
```r
# no. of users (in both conditions) who shared bad tweets per day
d1[, .N, keyby = .(ymd, period)]
ymd period N
1: 2023-03-24 0 66216 # t0
2: 2023-03-25 0 65812
3: 2023-03-26 0 65193
4: 2023-03-27 0 64198
5: 2023-03-28 0 70359
6: 2023-03-29 0 67636
7: 2023-03-30 0 72490
8: 2023-03-31 0 71532
9: 2023-04-01 0 67544
10: 2023-04-02 0 67302
11: 2023-04-03 0 63957
12: 2023-04-04 0 69728
13: 2023-04-05 0 69189
14: 2023-04-06 0 63429
15: 2023-04-07 0 62827
16: 2023-04-08 0 62128
17: 2023-04-09 0 59859
18: 2023-04-10 0 61157
19: 2023-04-11 0 59834
20: 2023-04-12 0 57851
21: 2023-04-13 0 63182
22: 2023-04-14 0 64408
23: 2023-04-15 0 69808
24: 2023-04-16 0 72291
25: 2023-04-17 1 69726 # t1
26: 2023-04-18 1 66242
27: 2023-04-19 1 60174
28: 2023-04-20 1 65045
29: 2023-04-21 1 63559
30: 2023-04-22 1 62203
31: 2023-04-23 1 60413
ymd period N
```
# models without interaction
```r
> feols(t1bin ~ conditionC + t0binSC| block + day, d3, cluster = "block")
OLS estimation, Dep. Var.: t1bin
Observations: 3,601,059
Fixed-effects: block: 13,203, day: 7
Standard-errors: Clustered (block)
Estimate Std. Error t value Pr(>|t|)
conditionC 0.000241 0.000475 0.507174 0.61204
t0binSC 0.206838 0.000970 213.148396 < 2.2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
RMSE: 0.274783 Adj. R2: 0.155288
Within R2: 0.113745
> feglm(t1bin ~ conditionC + t0binSC | block + day, family = "binomial", d3, cluster = "block")
NOTE: 1,203/0 fixed-effects (79,464 observations) removed because of only 0 (or only 1) outcomes.
GLM estimation, family = binomial, Dep. Var.: t1bin
Observations: 3,521,595
Fixed-effects: block: 12,000, day: 7
Standard-errors: Clustered (block)
Estimate Std. Error t value Pr(>|t|)
conditionC 0.002361 0.006512 0.362538 0.71695
t0binSC 3.531350 0.011169 316.172953 < 2.2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Log-Likelihood: -853,723.0 Adj. Pseudo R2: 0.253248
BIC: 1,888,459.8 Squared Cor.: 0.18295
```