中介分析学习笔记

library(mediation)  
library(readxl)  # 用于读取Excel文件  
library(officer)  # 用于创建Word文档  
library(ggplot2)  # 用于绘图  

set.seed(123456)
#single
data <- read_excel("C:/Users/lenovo/Desktop/new_analysis_teenfood_mediation.xlsx")  
model_mediator <- lm(fgf21 ~ BMI, data = data)  
model_outcome <- lm(var29 ~ BMI + fgf21, data = data) 
mediation_result <- mediate(model_mediator, model_outcome, treat = "BMI", mediator = "fgf21", boot = TRUE, sims = 1000)  
summary(mediation_result)
summary(model_outcome)
plot(mediation_result)

以上代码是以BMI为自变量，fgf21为中介变量，var29为因变量，探讨fgf21在BMI和var29中的中介作用。种子点可以自己设置。在R中运行以后，生成的结果如下：

Causal Mediation Analysis 

Nonparametric Bootstrap Confidence Intervals with the Percentile Method

               Estimate 95% CI Lower 95% CI Upper p-value    
ACME            0.03878      0.01058         0.07   0.006 ** 
ADE             0.07757      0.00284         0.16   0.042 *  
Total Effect    0.11635      0.04723         0.19  <2e-16 ***
Prop. Mediated  0.33327      0.08242         0.95   0.006 ** 
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Sample Size Used: 104 


Simulations: 1000 

> summary(model_outcome)

Call:
lm(formula = var29 ~ BMI + fgf21, data = data)

Residuals:
    Min      1Q  Median      3Q     Max 
-3.4973 -0.8360  0.0974  0.8952  2.9218 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept) -5.31089    1.00836  -5.267 7.86e-07 ***
BMI          0.07757    0.03494   2.220  0.02862 *  
fgf21        0.54664    0.18866   2.898  0.00461 ** 
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 1.445 on 101 degrees of freedom
  (因为不存在，2个观察量被删除了)
Multiple R-squared:  0.1748,	Adjusted R-squared:  0.1585 
F-statistic:  10.7 on 2 and 101 DF,  p-value: 6.113e-05

重点需要关注以下几个指标：

ACME表示中介效应的估计值，ADE表示直接效应的估计值，总效应：Total effect，需要重点关注这三个指标是否符合以下要求：①置信区间是否横跨0，置信区间不包含零 = 显著效应存在 (效果真实存在)。置信区间包含零 = 效应不显著 (效果可能为零，无法确定效果存在)。②他们的P值是否有显著性意义。  置信区间不包括0，且P值显著，则说明是有中介效应的。

承接上面，一般结果的显示方式是：

对应的结果数据如下：需要的数值为，ACME、ADE的值就是estimate的值，置信区间，以及P值。总的中介效应的prop.mediated的Estimate的数值。

summary(mediation_result)
Causal Mediation Analysis 

Nonparametric Bootstrap Confidence Intervals with the Percentile Method

               Estimate 95% CI Lower 95% CI Upper p-value    
ACME            0.08916      0.00477         0.23   0.034 *  
ADE             0.20424      0.05567         0.35   0.016 *  
Total Effect    0.29340      0.13355         0.47  <2e-16 ***
Prop. Mediated  0.30389      0.02118         0.74   0.034 *  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Sample Size Used: 22 


Simulations: 1000 
> summary(model_outcome)
Call:
lm(formula = Acc_L ~ t2d + fgf21, data = data)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.52623 -0.11906  0.04607  0.14424  0.32314 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)   
(Intercept) -0.91085    0.26044  -3.497  0.00241 **
t2d          0.20424    0.09486   2.153  0.04437 * 
fgf21        0.13238    0.06213   2.131  0.04640 * 
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.1996 on 19 degrees of freedom
  (因为不存在，4个观察量被删除了)
Multiple R-squared:  0.4636,	Adjusted R-squared:  0.4071 
F-statistic: 8.209 on 2 and 19 DF,  p-value: 0.002695

供参考的资料：中介效应那些事——ab与c'符号相异的两种情况、中介与遮掩混合效应解读 - 知乎