Question

我正在运行几个回归模型，并希望将结果放在数据框中。我尝试了几种不成功的方法。

# Generate some fake data: three responses and eight covariates 
df <- data.frame(y1 = rnorm(50), y2 = rnorm(50), y3 = rnorm(50), 
                 aa = rnorm(50), ab = rnorm(50), ca = rnorm(50), 
                 cd = rpois(50, 30), v8 = runif(50)) 

# Create a vector of covariate names 
xs <- c("aa","ab", "ca","cd","v8") 
# Initialize a list whose length is that of the vector xs 
rl1 <- vector('list', 5) 
rl2 <- vector('list', 5) 
rl3 <- vector('list', 5) 

for(i in 1:5) { fm1<- as.formula(paste('y1', xs[i], sep = '~')) 
fm2 <- as.formula(paste('y2', xs[i], sep = '~')) 
fm3 <- as.formula(paste('y3', xs[i], sep = '~')) 
rl1[[i]] <- summary(lm(fm1, data = df) )
rl2[[i]] <- summary(lm(fm2, data = df) )
rl3[[i]] <- summary(lm(fm3, data = df) )
} 

rl1
rl2

###
#The goal is to:
#create a data frame for results that looks like this 

Variable    t value     Pr(>|t|)        
aa      -1.032      0.3070 
ab      -0.121      0.9043
ca       0.041      0.967
cd       0.429      0.670
v8      -1.204      0.2346 

and create a data frame for results that looks like this 
Variable    rl1.t value r2.t value  
aa      -1.032      -1.846 
ab      -0.121      -0.230
ca       0.041      1.347
cd       0.429      0.211
v8      -1.204      0.214

最后，我发现这个代码以这种方式运行回归，我想表示赞赏并表示感谢，尽管我不记得来源。

Answer 1

您提到的方法使用了许多软件包（dplyr，tidyr，purrr和broom）。所有这些都方便地位于tidyverse元数据包中，这有助于安装所有这些软件包。

tidyverse包具有基于整洁数据和整洁数据工具的工作流程。您可以阅读更多相关信息here。

您的示例，使用tidyverse软件包如下：

library(tidyverse)
library(broom) # is part of the tidyverse but has to be attached on its own.

df <- data.frame(y1 = rnorm(50), y2 = rnorm(50), y3 = rnorm(50), 
                 aa = rnorm(50), ab = rnorm(50), ca = rnorm(50), 
                 cd = rpois(50, 30), v8 = runif(50)) 

xs <- c("aa","ab", "ca","cd","v8") 

results <- tibble(covariates = xs, data = list(df)) %>% 
       mutate(fm1 = map2(data, covariates, ~lm(as.formula(paste('y1', .y, sep = '~')),data = .x)), 
              fm2 = map2(data, covariates, ~lm(as.formula(paste('y2', .y, sep = '~')),data = .x)),
              fm3 = map2(data, covariates, ~lm(as.formula(paste('y3', .y, sep = '~')),data = .x))) %>% 
       gather(model, result, fm1:fm3) %>% 
       mutate(stats = map(result, tidy)) 


results %>% 
  select(-data, -result) %>% 
  spread(model, stats) %>% 
  unnest(.sep = "_")

# A tibble: 10 × 16
   covariates    fm1_term fm1_estimate fm1_std.error fm1_statistic fm1_p.value    fm2_term fm2_estimate
        <chr>       <chr>        <dbl>         <dbl>         <dbl>       <dbl>       <chr>        <dbl>
1          aa (Intercept)  0.186250282    0.10901934     1.7084150  0.09401936 (Intercept)  -0.13702761
2          aa          aa -0.122391949    0.09736266    -1.2570727  0.21480983          aa   0.07086972
3          ab (Intercept)  0.235819869    0.11000625     2.1436951  0.03715205 (Intercept)  -0.07707552
4          ab          ab  0.210060898    0.11165387     1.8813580  0.06599560          ab   0.26121398
5          ca (Intercept)  0.192849006    0.11315134     1.7043458  0.09478244 (Intercept)  -0.13487163
6          ca          ca  0.028244577    0.11597189     0.2435467  0.80861978          ca   0.01340660
7          cd (Intercept)  0.336090343    0.67322033     0.4992279  0.61989940 (Intercept)  -1.50990351
8          cd          cd -0.004866459    0.02170098    -0.2242506  0.82351414          cd   0.04484769
9          v8 (Intercept)  0.037664728    0.18697545     0.2014421  0.84120415 (Intercept)   0.12866580
10         v8          v8  0.353757262    0.35827371     0.9873939  0.32840091          v8  -0.62992103
# ... with 8 more variables: fm2_std.error <dbl>, fm2_statistic <dbl>, fm2_p.value <dbl>, fm3_term <chr>,
#   fm3_estimate <dbl>, fm3_std.error <dbl>, fm3_statistic <dbl>, fm3_p.value <dbl>

在此示例中，步骤如下：

创建一个covariate名称的tibble并为每个协变量嵌套数据。
为每个协变量创建模型fm1到fm3，作为单独的列表列，将线性模型映射到每个协变量。
将其重塑为长格式（键值对）
创建统计信息列，并通过将broom::tidy映射到结果列进行整理。
将整件物品存放在一个小组中

然后，您可以使用results tibble通过unnest获取所需内容，并通过dplyr和tidyr将其操作为您想要的内容。

将回归汇总结果放在数据框R中

1 个答案: