Experiment Analysis:Context:
Twitter 让其他公司投放广告。 For this exercise, imagine we have a single ad product where advertisers pay us each each time a userclicks on their ad. 每一个 campaign有 a budget. An advertiser never has to pay more than their budget, so if we were to spendmore than the campaign’s budget, we would not be able to bill the advertiser for the additionalspend. 这个叫 overspending . 很难avoid overspending because there is a delay between when we send ads to users and when they actually click on those ads. Since Twitter only charges advertisers for actual clicks on their ads, the charges can enter into thesystem after some (random) delay. For example, suppose a campaign has $10 of budgetremaining and the ad serving systems serves out 1000 ads, expecting 10 of them to generate aclick resulting in $1 of revenue each. If, however, 20 ads end up generating clicks, we wto support his suspicions, or show that any differences in budgets arelikely due to random fluctuations.
我的回答(feedback 说我的答案不对，想跟大家讨论一下）：
1. Created dummy variable to indicate overspend: If overspend>0 then 1, else 0. (Overspend = spend -budget)
large sample so using z-test
最后我是test whether this overspend dummy variable has more count in treatment than in control
2. 对budget in treatment and control做了一个 z-test 发现p =0.15 所以是random fluctuations.
请大家说说你会怎么解答， 为什么我的回答错了 谢谢！附件是面试官给出用来分析这个题的数据， 要用python or R 来分析。
| 
狗仔卡
收藏
淘帖
好苗
杂草
匿名用户-U1VGY
提升卡
置顶卡
关贴卡
开贴卡
变色卡
千斤顶
