# Campaign summary Ran campaign for 6 days - total spend: $1465.70 CAD - audience reach: 6661 (of ~12k) - campaign daily budget: 10k (we can increase this) - impression frequency cap: max 10 impressions/day/user (we can increase this) - impressions: 48442 # 6 potential outcome measures - newsguard `ng`, newsguard predicted `ng_pred` - factcheck `fc`, factcheck predicted `fc_pred` - misinfome (iffy) `misinfome`, misinfome predicted `misinfome_pred` For all outcomes, **larger values = higher quality domains**. See [[20210813_001605 domain news ratings correlations|here]] for plots and correlations between outcomes. # Timeline ![[Pasted image 20210917112509.png|377]] # Users per condition - control: 11200/12288 (matched/uploaded) - treatment: 11511/12235 Features used for blocking/randomization: ```r features <- c("active_days", # no. of active days (Aug 22 to Aug 31) "n_retweet", # no. of retweets with links to bad sources (Aug 22 to Aug 31) "n_tweet", # no. of tweets with links to bad sources (Aug 22 to Aug 31) "followers_count", "favourites_count", "statuses_count", "friends_count", "friend_follow_ratio", "days_since_create", "misinfome_pred") # median badness (Aug 22 to Aug 31) ``` # Many users have missing values/outcomes No. of users without any links/domains during campaign ```r condition ng_mean_1 ng_pred_mean_1 fc_mean_1 fc_pred_mean_1 misinfome_pred_mean_1 misinfome_mean_1 1: c 4487 4486 6265 4486 4274 7442 2: t 4467 4466 6231 4466 4248 7325 ``` No. of users without a single tweet during campaign ```r > dt1[n_total_1 == 0, .N, keyby = condition] condition N 1: c 2186 2: t 2165 ``` No. of users whose timelines cannot be retrieved (e.g., protected). Will take a look again! ```r condition N 1: c 302 2: t 291 ``` # Results ## Approach 1: Fill missing values during campaign with **mean** during pre-campaign 6-day period ([[20210916_102017 campaign 0001#Timeline|time 0 window]] above) 6 outcomes, so 6 models/columns `outcome_mean_time1 ~ condition` ![[Pasted image 20210917132238.png]] `outcome_mean_time1 ~ condition + outcome_mean_time0` ![[Pasted image 20210917132302.png]] ## Approach 2: Exclude users with missing values `outcome_mean_time1 ~ condition` ![[Pasted image 20210917121928.png]] `outcome_mean_time1 ~ condition + outcome_mean_time0` ![[Pasted image 20210917122008.png]] ## Approach 3: Fill missing values during campaign with **mean of control** during 6-day campaign period `outcome_mean_time1 ~ condition` ![[Pasted image 20210917132434.png]] `outcome_mean_time1 ~ condition + outcome_mean_time0` ![[Pasted image 20210917132516.png]]