COD实验分析

load('/Users/milin/分控模型验证/COD实验数据.RData')
library(tidyverse)

分析不同实验组的情况

Ucus <- cod_df_shiyan1 %>% group_by(variant,state) %>% summarise(n = n())
Ucus <- Ucus %>% filter(state=='pay success')
Ucus
## # A tibble: 4 x 3
## # Groups:   variant [4]
##   variant state           n
##     <int> <chr>       <int>
## 1       0 pay success 18137
## 2       1 pay success 18091
## 3       2 pay success 19412
## 4       3 pay success 19192

可以看出单量差别不是很大

问题1

对照组人员COD下单次数不合常理

Ucus <- cod_df_shiyan1 %>% group_by(variant,user_id,state) %>% summarise(n = n())

Ucusbig1 <- Ucus %>% filter(n!=1)


DT::datatable(Ucus)
DT::datatable(Ucusbig1)

我们可以看到对照组的COD人员下了非常多的订单.在对照组中，有0.1276135的人下了2单及以上。约有3%的人下3单及以上,这3%的人，占比单量为13%。去除掉这3%的数据,再进行分析

require(Hmisc)
Ucus_1 <- Ucus %>% filter((variant %in% c(0,1) &state=='pay success'&n<=2)|(variant %in% c(2,3)))

Ucus_1g <- Ucus_1 %>% group_by(variant,state) %>% summarise(n=sum(n)) %>% ungroup()
Ucus_1g <- Ucus_1g %>% filter(state=='pay success')
Ucus_1g
## # A tibble: 4 x 3
##   variant state           n
##     <int> <chr>       <int>
## 1       0 pay success 15916
## 2       1 pay success 15755
## 3       2 pay success 19412
## 4       3 pay success 19192

可以看到实验组有更多的单量

观察不同组的签收率

Ucus2 <- cod_df_shiyan1 %>% filter(user_id %in% Ucus_1$user_id) %>% group_by(variant,current_status) %>% summarise(n = n())

Ucus2 <- Ucus2 %>% filter(current_status=='delivered')

data.frame(Ucus2$variant,Ucus2$n/Ucus_1g$n)
##   Ucus2.variant Ucus2.n.Ucus_1g.n
## 1             0         0.6878613
## 2             1         0.6852428
## 3             2         0.6708737
## 4             3         0.6675177