跨多个列循环计算描述性统计信息

编程问答 2022-07-27

问题描述

我正在为一个变量生成随机数，并将该过程重复几次。我想在循环的每次迭代中为每个组（value ，group1，group2）计算group3的平均值。我想存储结果，以便以后估算循环所有迭代中每个组的平均份额。

require(tidyverse)

set.seed(21)
group1 <- sample(c("A1","A2","B1","B2","C1","C2"),1000,TRUE)
group2 <- sample(c("G1","G2","G4"),TRUE)
group3 <- sample(c("D1","D2"),TRUE)
prob <- runif(1000,1)
df <- as.data.frame(cbind(group1,group2,group3,prob))

df$prob <- as.numeric(df$prob)

for (i in 1:15) {
  
  df <- df %>%
    mutate(value = rbinom(nrow(df),1,prob = prob))

  # [INSERT CALculaTION OF MEAN FOR EACH GROUP VARIABLE AND STORE IT]
}

# [INSERT CALculaTION OF MEAN ACROSS ALL IteraTIONS]

我的主要问题是如何有效地估计多个变量中value 的平均值，并以一种平滑的方式存储结果。

谢谢。

要澄清：

我希望最终结果看起来像这样：

col "group1_A1" "Group1_A2" "group1_B1" "group1_B2" "group1_C1" "group1_C2" "group2_G1" "group2_G2" "group2_G4" "group3_D1" "group3_D2"
x1  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x2  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x3  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x4  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x4  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x5  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x6  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x7  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x8  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x9  "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"    
x10 "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"     "x_bar"

其中三组子组的平均值替换"x_bar，而每一行是一个迭代计算的平均值。一个简单的解决方案是使用dplyr的{​​{1}}，但是我想找到一个解决方案，所以我要遍历所有三个分组变量。

将其放在上下文中：假设变量group_by是死亡的概率。 prob是表示6个年龄段的变量，group1表示社会经济地位，group2是性别。然后，我想看看谁最有可能死。为此，我随机生成一个伯努利变量，该变量取决于group3的概率。为了消除一些随机性，我重复此过程15次，然后想知道每个死亡的社会人口统计学组的份额有多大（在变量prob上收到1的值。对于每次迭代，我都希望计算死亡人数的组别（因此，有多少男性死亡，有多少老年人死亡）。很抱歉，我没有提出一个更快乐的例子。

解决方法

这是一种使用某些tidyverse函数的方法。

library(dplyr)
library(tidyr)
df2 <- df %>% 
  pivot_longer(starts_with("group") ) %>%
  mutate(group = paste0(name,"_",value)) %>%
  select(group)

for (i in 1:15) {
  
  df2 <- df %>%
    mutate(value = rbinom(nrow(df),1,prob = prob)) %>%
    pivot_longer(starts_with("group"),values_to = "val" ) %>%
    mutate(group = paste0(name,val)) %>%
    group_by(group) %>%
    summarise(mean = mean(value,na.rm = TRUE)) %>%
    rename_with(.cols = mean,.fn = ~ paste0("mean",i)) %>%
    inner_join(df2,by = c("group" = "group"))
}
df2 %>%
  pivot_longer(starts_with("mean"),names_to = "trial",names_prefix = "mean") %>%
  distinct() %>%
  pivot_wider(id_cols = mean,names_from = "group",values_from = "value")
# A tibble: 15 x 12
   trial group1_A1 group1_A2 group1_B1 group1_B2 group1_C1 group1_C2 group2_G1 group2_G2 group2_G4 group3_D1 group3_D2
   <chr>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>     <dbl>
 1 15        0.519     0.514     0.516     0.519     0.551     0.533     0.529     0.542     0.507     0.518     0.533
 2 14        0.481     0.486     0.503     0.536     0.487     0.550     0.526     0.493     0.507     0.495     0.520
 3 13        0.506     0.541     0.477     0.470     0.572     0.556     0.575     0.499     0.496     0.486     0.552
 4 12        0.519     0.534     0.497     0.557     0.604     0.509     0.549     0.522     0.548     0.548     0.531
 5 11        0.5       0.568     0.458     0.481     0.497     0.562     0.542     0.496     0.496     0.467     0.548
 6 10        0.525     0.466     0.503     0.503     0.535     0.580     0.581     0.490     0.496     0.488     0.548
 7 9         0.494     0.547     0.490     0.448     0.610     0.598     0.578     0.504     0.519     0.501     0.560
 8 8         0.538     0.554     0.471     0.530     0.599     0.538     0.545     0.516     0.559     0.565     0.518
 9 7         0.525     0.588     0.548     0.475     0.535     0.568     0.601     0.499     0.522     0.507     0.565
10 6         0.513     0.527     0.529     0.546     0.561     0.503     0.562     0.513     0.522     0.510     0.550
11 5         0.462     0.493     0.503     0.508     0.513     0.568     0.513     0.493     0.522     0.507     0.510
12 4         0.506     0.5       0.452     0.481     0.599     0.556     0.545     0.516     0.496     0.520     0.516
13 3         0.525     0.466     0.503     0.525     0.567     0.556     0.529     0.550     0.499     0.497     0.552
14 2         0.462     0.554     0.471     0.514     0.519     0.574     0.536     0.516     0.499     0.520     0.512
15 1         0.506     0.541     0.497     0.470     0.519     0.544     0.510     0.519     0.507     0.510     0.514

这是您的第一部分-data.frame，其中每一行都是一个试验，每个组的均值。

您的第二部分如下：

df2 %>%
  pivot_longer(starts_with("mean"),names_prefix = "mean") %>%
  distinct() %>%
  group_by(group) %>%
  summarize(mean = mean(value))
# A tibble: 11 x 2
   group      mean
   <chr>     <dbl>
 1 group1_A1 0.505
 2 group1_A2 0.525
 3 group1_B1 0.495
 4 group1_B2 0.504
 5 group1_C1 0.551
 6 group1_C2 0.553
 7 group2_G1 0.548
 8 group2_G2 0.511
 9 group2_G4 0.513
10 group3_D1 0.509
11 group3_D2 0.535
, 
library(dplyr)
df <- df %>%
    mutate(value = rbinom(nrow(df),prob = prob)) %>%
    summarize_if(is.numeric,mean,na.rm = TRUE) #summarize_all also works if all columns are numeric anyway
data-managementdplyrdplyrrr