用不同的因变量重复回归

Question

我在 Stack 和 google 中搜索了解决方案，none 找到了解决我问题的方法。

我有大约 40 个因变量，我的目标是获得调整均值 (lsmeans)。在考虑了一些协变量之后，我需要对 A 组和 B 组进行调整。我的最终对象应该是一个数据框，其中包含 A 组和 B 组的所有 40 个因变量的预测均值。

这是我试过的，没有成功：

# Examplified here with 2 outcome variables
outcome1 <- c(2, 4, 6, 8, 10, 12, 14, 16)
outcome2 <- c(1, 2, 3, 4, 5, 6, 7, 8)
var1 <- c("a", "a", "a", "a", "b", "b", "b", "b")
var2 <- c(10, 11, 12, 9, 14, 9, 5, 8)
var3 <- c(100, 101, 120, 90, 140, 90, 50, 80)

df <- data.frame(outcome1, outcome2, var1, var2, var3)

dependents <- c(outcome1, outcome2)

library(lsmeans) #install.packages("lsmeans")

results <- list()
for (i in seq_along(dependents) {
    fit <- lm(i ~ var1 + var2 + var3, data= df)
    summary <- summary(lsmeans(fit, "var1"))
    summary$outcome <- i
    results[i] <- summary
    }

Answer 1

有一些错别字之类的，但我想这就是你想要的：

# Examplified here with 2 outcome variables
outcome1 <- c(2, 4, 6, 8, 10, 12, 14, 16)
outcome2 <- c(1, 2, 3, 4, 5, 6, 7, 8)
var1 <- c("a", "a", "a", "a", "b", "b", "b", "b")
var2 <- c(10, 11, 12, 9, 14, 9, 5, 8)
var3 <- c(100, 101, 120, 90, 140, 90, 50, 80)

df <- data.frame(outcome1, outcome2, var1, var2, var3)

dependents <- c("outcome1", "outcome2")

library(lsmeans) #install.packages("lsmeans")

results <- list()
for (i in seq_along(dependents)) {
  eq <- paste(dependents[i],"~ var1 + var2 + var3")
  fit <- lm(as.formula(eq), data= df)
  summary <- summary(lsmeans(fit, "var1"))
  summary$outcome <- i
  results[[i]] <- summary
}

Answer 2

这是另一个使用 lapply 的选项。

dependents <- c('outcome1', 'outcome2')
lst <- lapply(dependents, function(x) {
         fit <- lm(paste(x,'~', 'var1+var2+var3'), data=df)
         summary(lsmeans(fit, 'var1', data=df))})
Map(cbind, lst, outcome = seq_along(dependents))

Answer 3

在更现代的 R 中，the lazyeval package 提供了更好的公式处理函数。

这是我的代码版本：

#load libs
library(tidyverse)
library(lazyeval)
library(lsmeans)

#make data
df = tibble(
  y1 = c(2, 4, 6, 8, 10, 12, 14, 16),
  y2 = c(1, 2, 3, 4, 5, 6, 7, 8),
  var1 = c("a", "a", "a", "a", "b", "b", "b", "b"),
  var2 = c(10, 11, 12, 9, 14, 9, 5, 8),
  var3 = c(100, 101, 120, 90, 140, 90, 50, 80)
)

#outcomes
outcomes = c("y1", "y2")

#fit
results <- list()
for (i in seq_along(outcomes)) {
  #make a formula
  f = i ~ var1 + var2 + var3
  
  #set outcome, must be a symbol explicitly
  f_lhs(f) = as.symbol(outcomes[i])
  
  #fit
  fit <- lm(f, data = df)
  
  #save
  summary <- summary(lsmeans(fit, "var1"))
  results[[i]] = summary
}

#set outcome names
names(results) = outcomes

#print results
results

最后一行打印：

$y1
 var1 lsmean   SE df lower.CL upper.CL
 a       5.5 1.38  4     1.68     9.32
 b      12.5 1.38  4     8.68    16.32

Confidence level used: 0.95 

$y2
 var1 lsmean    SE df lower.CL upper.CL
 a      2.75 0.688  4     0.84     4.66
 b      6.25 0.688  4     4.34     8.16

Confidence level used: 0.95

通常，使用字符串并在拟合前转换为公式会更容易。我在这里使用公式来完成。

用不同的因变量重复回归

Repeat regression with varying dependent variable

loops

regression

r

dplyr