- see [[230512_123040 most visited domains|most visited domains]]
- Sample size: 233 control, 261 treatment
- Duration of study for users ranged from 1 to 104 days, but total study duration is 155 days (2022-11-07 to 2023-04-10)
- 1 prompt/day
- Sites/domains visited per user range from 64 to 157071
```r
# users per arm/condition
> d0[, n_distinct(anon_user), keyby = arm]
arm V1
1: CONTROL 233
2: EXPERIMENT 261
3: UNASSIGNED 362 # excluded from all analysis below
# rows of data (or events) per arm
arm N
1: UNASSIGNED 1052345 # excluded from analysis below
2: CONTROL 4595391
3: EXPERIMENT 4962959
```
no. of events per user
- 3 types of events: web browsing (tab_nav), prompt_shown, prompt_closed
```r
# total prompt events per condition
condition N
1: -0.5 3268 # control
2: 0.5 3214 # treatment
```
![[1683570285.png]]
# descriptives
```r
condition n_prompts
1: -0.5 3268
2: 0.5 3214
X-squared = 0.44986, df = 1, p-value = 0.5024
# min prompt/days: 1
# max prompts/days: 104
# no. of prompts shown don't differ by condition
> feglm(n_prompts ~ condition, dprompts_user, family = "quasipoisson", vcov = hc3)
GLM estimation, family = quasipoisson, Dep. Var.: n_prompts
Observations: 494
Standard-errors: Custom
Estimate Std. Error t value Pr(>|t|)
(Intercept) 2.575823 0.069498 37.063282 < 2.2e-16 ***
condition -0.130144 0.138996 -0.936314 0.34957
```
![[1683831540.png]]
when users were prompted, they were generally on good domains (based on PC1 score - x-axis)
![[1683832005.png]]
```r
# no. of share responses per condition
# NA: no response
# 0: not share
# 0.5: not sure
# 1.0: share
share condition N
1: NA -0.5 2611
2: NA 0.5 2220
3: 0.0 -0.5 208
4: 0.0 0.5 281
5: 0.5 -0.5 90
6: 0.5 0.5 188
7: 1.0 -0.5 359
8: 1.0 0.5 525
```
![[1683832672.png]]
![[1683832766.png]]
no. of prompts with and without sharing responses don't differ significantly between conditions
- quasipoisson count model
```r
# DV: no. of prompts with sharing responses
> feglm(n_prompts_responded ~ condition, dprompts_avg, family = "quasipoisson", vcov = hc3)
GLM estimation, family = quasipoisson, Dep. Var.: n_prompts_responded
Observations: 494
Standard-errors: Custom
Estimate Std. Error t value Pr(>|t|)
(Intercept) 1.036646 0.186968 5.54452 4.8162e-08 ***
condition0.5 0.300571 0.245786 1.22290 2.2195e-01
# DV no. of prompts WITHOUT sharing responses
GLM estimation, family = quasipoisson, Dep. Var.: n_prompts_noresponse
Observations: 494
Standard-errors: Custom
Estimate Std. Error t value Pr(>|t|)
(Intercept) 2.416450 0.099101 24.38377 < 2.2e-16 ***
condition0.5 -0.275708 0.149469 -1.84459 0.065699 . # almost significant
```
# how much to winsorize? .99 or .95?
pc1 threshold 60
- count_bad: raw count
- cound_bad_win*: winsorize threshold 95 etc.
![[1683834794.png]]
# no. of prompts/users per day
- 2022-11-07 to 2023-04-10
- each day, 20-30 users were prompted
![[1684258686.png]]
# no. of prompts/users with response per day
- sparse data: 1 to 20 users per day
![[1684259837.png]]
# no. of prompts/users with response per day, for domains with PC1 < .5
- even sparser data...
![[1684260801.png]]