Lab preparation

Creating a version-controlled R-Project with Github

Download repository here: https://github.com/garberadamc/SEM-Lab4

On the Github repository webpage:

fork your own branch of the lab repository
copy the repository web URL address from the clone or download menu

Within R-Studio:

click “NEW PROJECT” (upper right corner of window)
choose option Version Control
choose option Git
paste the repository web URL path copied from the clone or download menu on Github page
choose location of the R-Project (too many nested folders will result in filepath error)

Load packages

library(datapasta)
library(mediation)
library(tidyverse)
library(MplusAutomation)
library(rhdf5)
library(here)
library(kableExtra)
library(gtsummary)

Lab outline

Estimate a mediation model using the {mediation} package
Estimate the same model using the Structural Equation Modeling (SEM) framework with {MplusAutomation}
For the second empirical example, estimate parallel models using the mediation and SEM methods

A quick detour - Equivalent models

Figure. Picture adapted from SEM slides by Sacha Epskamp http://sachaepskamp.com/files/SEM22019/SEM2_2019_Week2_slides.pdf

Have you ever seen the perfect table and want to adapt it for your own research purposes?

Use {datapasta} by copying tables and pasting them automatically as tribbles or dfs

copy a table or data matrix
run the fuction tribble_paste() or df_paste()

tribble_paste()


pasta_table %>% 
  kable(booktabs = T, linesep = "") %>% 
  kable_styling(latex_options = c("striped"), 
                full_width = F,
                position = "left")

The empirical examples of mediation used in this exercise are from the following article

Tingley, D., Yamamoto, T., Hirose, K., Keele, L., & Imai, K. (2014). Mediation: R package for causal mediation analysis.

https://cran.r-project.org/web/packages/mediation/vignettes/mediation.pdf

Data source for example 1

Brader T, Valentino NA, Suhat E (2008). What Triggers Public Opposition to Immigration? Anxiety, Group Cues, and Immigration. American Journal of Political Science, 52(4), 959–978.

https://onlinelibrary.wiley.com/doi/pdf/10.1111/j.1540-5907.2008.00353.x

To see metadata run - ?framing

Read in the framing dataset

set.seed(4212020)

data("framing", package = "mediation")

framing <- droplevels(framing) %>%  # drop factor levels with frequency zero
  mutate(emo = emo - 2)

Take a look at variables used in the mediation model

Name	Labels	Variable status
emo	Measure of subjects’ negative feeling during the experiment (1-10). 1 indicates the most negative feeling.	Mediator (Z)
treat	Framing codition interaction term. News story with conditions tone (Negative/Positive) and ethnic identity cue (Latino/European)	Treatment (X)
cong_mesg	Whether subjects requested sending an anti-immigration message to Congress on their behalf.	Outcome (Y)
age	Age of subject (18-85)	pre-treatment covariate
educ	Education (1-4)	pre-treatment covariate
gender	Gender (Male/Female)	pre-treatment covariate
income	Subjects’ income, measured as a 19-point scale.	pre-treatment covariate

Look at descriptives table for the framing dataset using {gtsummary}

table_data <- framing %>% 
  select(emo, treat, cong_mesg, age, educ, gender, income)

table1 <- tbl_summary(table_data,
                      statistic = list(all_continuous() ~ "{mean} ({sd})"),
                      missing = "no") %>%
  bold_labels() 

table1

Characteristic	N = 265¹
emo	4.97 (2.77)
treat	68 (26%)
cong_mesg	88 (33%)
age	48 (16)
educ
less than high school	20 (7.5%)
high school	92 (35%)
some college	70 (26%)
bachelor's degree or higher	83 (31%)
gender
male	126 (48%)
female	139 (52%)
income	11 (4)
¹ Statistics presented: mean (SD); n (%)

Estimate a mediation model in R using {`mediation`}

step 1: fit a linear model of the mediator (emo) regressed on treatment (treat) and pre-treatment covariates

med_fit <- lm(emo ~ treat + age + educ + gender + income, 
              data = framing)

step 2: fit a general linear model (glm) with the binary outcome variable cong_mesg regressed on treatment (treat), mediator, and pre-treatment covariates

out_fit <- glm(cong_mesg ~ emo + treat + age + educ + gender + income,
               data = framing,
               family = binomial("probit"))

step 3: estimate the mediation effects with bias corrected bootstrapped confidence intervals

med_out <- mediate(med_fit, out_fit, treat = "treat", mediator = "emo",
                   boot = TRUE, boot.ci.type ="bca", sims = 100)

summary(med_out)

## 
## Causal Mediation Analysis 
## 
## Nonparametric Bootstrap Confidence Intervals with the BCa Method
## 
##                          Estimate 95% CI Lower 95% CI Upper p-value    
## ACME (control)             0.0824       0.0246         0.13  <2e-16 ***
## ACME (treated)             0.0835       0.0239         0.14  <2e-16 ***
## ADE (control)              0.0113      -0.0921         0.12    0.70    
## ADE (treated)              0.0124      -0.1051         0.13    0.70    
## Total Effect               0.0948      -0.0205         0.25    0.16    
## Prop. Mediated (control)   0.8693     419.1265       666.19    0.16    
## Prop. Mediated (treated)   0.8808     369.2224       586.64    0.16    
## ACME (average)             0.0829       0.0224         0.13  <2e-16 ***
## ADE (average)              0.0118      -0.0991         0.12    0.70    
## Prop. Mediated (average)   0.8751     394.1745       626.42    0.16    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Sample Size Used: 265 
## 
## 
## Simulations: 100

Run mediation model 1 using the Structural Equation Modeling framework with {`MplusAutomation`}

m1_mediate  <- mplusObject(
  TITLE = "m1 mediate Lab4", 
  VARIABLE = 
   "usevar =
    cong_mesg emo treat age
    educ gender income;
  
    categorical = cong_mesg; ! outcome is binary",            
  
  ANALYSIS =  "bootstrap = 500; ! set number of bootstrap samples (500 for example purposes)" , 
  
  MODEL = 
   "emo on treat age educ gender income;            ! mediator linear regression 
    cong_mesg on emo treat age educ gender income;  ! outcome GLM regression
    
    Model indirect:
    cong_mesg ind treat;" ,
  
  OUTPUT = 
    "sampstat standardized cinterval (bcbootstrap); ! bias-corrected bootstrap", 
  
  PLOT = "type=plot2;",
  
  usevariables = colnames(framing),   
  rdata = framing)                    

m1_mediate_fit <- mplusModeler(m1_mediate,
                     dataout=here("mplus_files", "Lab4.dat"),       
                    modelout=here("mplus_files", "m1_mediate_Lab4.inp"),
                    check=TRUE, run = TRUE, hashfilename = FALSE)

Model 1 Mplus output

STANDARDIZED TOTAL, TOTAL INDIRECT, SPECIFIC INDIRECT, AND DIRECT EFFECTS

                 Estimate       S.E.  Est./S.E.    P-Value

Effects from TREAT to CONG_MES

  Total                0.109      0.075      1.453      0.146
  Total indirect       0.101      0.031      3.253      0.001

  Specific indirect 1
    CONG_MES
    EMO
    TREAT              0.101      0.031      3.253      0.001

  Direct
    CONG_MES
    TREAT              0.008      0.071      0.119      0.905

Data source for example 2

Vinokur AD, Price RH, Schul Y (1995). Impact of the JOBS Intervention on Unemployed Workers Varying in Risk for Depression. American Journal of Community Psychology, 23(1), 39–74.

https://link.springer.com/content/pdf/10.1007/BF02506922.pdf

To see metadata run - ?jobs

Note: For this example we will ignore the issue of non-compliance addressed in Tingley et al. (2014) as this causal inference topic is beyond the scope of this course.

Read in the data from the job search intervention study (jobs)

data("jobs", package = "mediation")

Take a look at variables used in the mediation model

Name	Label
depress2 (Y)	Measure of depressive symptoms post-treatment.
treat (X)	Indicator variable for whether participant was randomly selected for the JOBS II training program. 1 = assignment to participation.
job_dich (Z)	The job_seek measure recoded into two categories of high and low. 1 = high job search self-efficacy.
sex	Indicator variable for sex. 1 = female
age	Age in years.
marital	Factor with five categories for marital status.
nonwhite	Indicator variable for race. 1 = nonwhite.
educ	Factor with five categories for educational attainment.
income	Factor with five categories for level of income.

Look at descriptives of the framing dataset using {gtsummary}

jobs_desc <- jobs %>% 
  select(depress2, job_dich, treat, sex, age, marital, nonwhite, educ, income)

tablej <- tbl_summary(jobs_desc,
                      statistic = list(all_continuous() ~ "{mean} ({sd})"),
                      missing = "no" ) %>%
  bold_labels() 

tablej

step 1: fit a binomial logist model using glm with the binary mediator (job_dich) regressed on treatment (treat) and pre-treatment covariates

jmed_fit <- glm(job_dich ~ treat + sex + age + marital +
                   nonwhite + educ + income,
                 data = jobs, family = binomial)

step 2: fit a linear model with depression score (depress2) regressed on treatment, mediator, and pre-treatment covariates

jout_fit <- lm(depress2 ~ job_dich + treat +
                 sex + age + marital + nonwhite + educ + income,
               data = jobs)

step 3: Estimate the mediation effects with bias corrected bootstrapped confidence intervals.

jmed_out <- mediate(jmed_fit, jout_fit, treat = "treat", mediator = "job_dich",
                   boot = TRUE, boot.ci.type ="bca", sims = 100)

summary(jmed_out)

## 
## Causal Mediation Analysis 
## 
## Nonparametric Bootstrap Confidence Intervals with the BCa Method
## 
##                Estimate 95% CI Lower 95% CI Upper p-value    
## ACME            -0.0145      -0.0571        -0.01  <2e-16 ***
## ADE             -0.0306      -0.1047         0.04    0.56    
## Total Effect    -0.0451      -0.1350         0.03    0.30    
## Prop. Mediated   0.3208     -78.0586         0.15    0.30    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Sample Size Used: 899 
## 
## 
## Simulations: 100

Run mediation model 2 as a SEM model with {`MplusAutomation`}

m2_jmediate  <- mplusObject(
  
  TITLE = "m2 jobs mediate Lab4", 
  
  VARIABLE = 
   "usevar = treat sex
    age marital nonwhite
    educ income depress2 job_dich;
                 
    categorical = job_dich; ! moderator is binary",            
  
  ANALYSIS =  
   "bootstrap = 500; ! set number of bootstrap draws (500 for example purposes)" , 
  
  MODEL = 
   "job_dich on treat sex age marital nonwhite educ income;
   
    depress2 on job_dich treat sex age marital nonwhite educ income;  
    
    Model indirect:
    depress2 ind treat;" ,
  
  OUTPUT = 
    "sampstat standardized cinterval (bcbootstrap); ! bias-corrected bootstrap", 
  
  PLOT = "type=plot2;",
  
  usevariables = colnames(jobs),   
  rdata = jobs)                    

m2_jmediate_fit <- mplusModeler(m2_jmediate,
                     dataout=here("mplus_files", "Lab4_jobs.dat"),       
                    modelout=here("mplus_files", "m2_jmediate_Lab4.inp"),
                    check=TRUE, run = TRUE, hashfilename = FALSE)

Model 2 Mplus output

STANDARDIZED TOTAL, TOTAL INDIRECT, SPECIFIC INDIRECT, AND DIRECT EFFECTS

                 Estimate       S.E.  Est./S.E.    P-Value

Effects from TREAT to DEPRESS2

  Total               -0.043      0.033     -1.306      0.192
  Total indirect      -0.029      0.012     -2.385      0.017

  Specific indirect 1
    DEPRESS2
    JOB_DICH
    TREAT             -0.029      0.012     -2.385      0.017

  Direct
    DEPRESS2
    TREAT             -0.015      0.033     -0.440      0.660

Run model 3 including the mediator*treatement interaction (potential outcomes framework)

For further reading on this topic see chapter 3 of Regression and mediation analysis using Mplus (Muthen et al., 2017)

m3_jmed  <- mplusObject(
  
  TITLE = "m3 MX jobs mediate Lab4", 
  
  VARIABLE = 
   "usevar =
    treat sex age marital nonwhite
    educ income depress2 job_dich mx; ", 
  
  DEFINE = "mx = job_dich*treat;",
  
  ANALYSIS =  "bootstrap = 500; ", 
  
  MODEL = 
   "job_dich on treat sex age marital nonwhite educ income;           
    depress2 on job_dich treat mx sex age marital nonwhite educ income;  
    
    Model indirect:
    depress2 MOD job_dich mx treat; ",
  
  OUTPUT = 
    "sampstat cinterval(bootstrap); ", 
  
  usevariables = colnames(jobs),   
  rdata = jobs)                    

m3_jmed_fit <- mplusModeler(m3_jmed,
                     dataout=here("mplus_files", "Lab4_jobs.dat"),       
                    modelout=here("mplus_files", "m3_jmediate_Lab4.inp"),
                    check=TRUE, run = TRUE, hashfilename = FALSE)

Model 3 Mplus output

TOTAL, INDIRECT, AND DIRECT EFFECTS BASED ON COUNTERFACTUALS (CAUSALLY-DEFINED EFFECTS)

Effects from TREAT to DEPRESS2

                 Estimate       S.E.  Est./S.E.    P-Value

  Tot natural IE      -0.026      0.011     -2.357      0.018
  Pure natural DE     -0.022      0.055     -0.401      0.688
  Total effect        -0.048      0.055     -0.878      0.380

 Other effects

  Pure natural IE     -0.023      0.012     -1.938      0.053
  Tot natural DE      -0.026      0.052     -0.494      0.621
  Total effect        -0.048      0.055     -0.878      0.380

References

Brader T, Valentino NA, Suhat E (2008). What Triggers Public Opposition to Immigration? Anxiety, Group Cues, and Immigration. American Journal of Political Science, 52(4), 959–978.

Hallquist, M. N., & Wiley, J. F. (2018). MplusAutomation: An R Package for Facilitating Large-Scale Latent Variable Analyses in Mplus. Structural equation modeling: a multidisciplinary journal, 25(4), 621-638.

Ingels, S. J., Pratt, D. J., Herget, D. R., Burns, L. J., Dever, J. A., Ottem, R., … & Leinwand, S. (2011). High School Longitudinal Study of 2009 (HSLS: 09): Base-Year Data File Documentation. NCES 2011-328. National Center for Education Statistics.

Muthén, B. O., Muthén, L. K., & Asparouhov, T. (2017). Regression and mediation analysis using Mplus. Los Angeles, CA: Muthén & Muthén.

Muthén, L.K. and Muthén, B.O. (1998-2017). Mplus User’s Guide. Eighth Edition. Los Angeles, CA: Muthén & Muthén

R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/

Tingley, D., Yamamoto, T., Hirose, K., Keele, L., & Imai, K. (2014). Mediation: R package for causal mediation analysis.

Vinokur AD, Price RH, Schul Y (1995). Impact of the JOBS Intervention on Unemployed Workers Varying in Risk for Depression. American Journal of Community Psychology, 23(1), 39–74.

Wickham et al., (2019). Welcome to the tidyverse. Journal of Open Source Software, 4(43), 1686, https://doi.org/10.21105/joss.01686

Lab 4 - Mediation

Structural Equation Modeling ED 216F - Instructor: Karen Nylund-Gibson

Adam Garber

April 24, 2020

Lab preparation

Creating a version-controlled R-Project with Github

Load packages

Lab outline

A quick detour - Equivalent models

Have you ever seen the perfect table and want to adapt it for your own research purposes?

The empirical examples of mediation used in this exercise are from the following article

Data source for example 1

Estimate a mediation model in R using {`mediation`}

Run mediation model 1 using the Structural Equation Modeling framework with {`MplusAutomation`}

Data source for example 2

Run mediation model 2 as a SEM model with {`MplusAutomation`}

Run model 3 including the mediator*treatement interaction (potential outcomes framework)

References

Lab preparation

Creating a version-controlled R-Project with Github

Load packages

Lab outline

A quick detour - Equivalent models

Have you ever seen the perfect table and want to adapt it for your own research purposes?

The empirical examples of mediation used in this exercise are from the following article

Data source for example 1

Estimate a mediation model in R using {mediation}

Run mediation model 1 using the Structural Equation Modeling framework with {MplusAutomation}

Data source for example 2

Run mediation model 2 as a SEM model with {MplusAutomation}

Run model 3 including the mediator*treatement interaction (potential outcomes framework)

References

Estimate a mediation model in R using {`mediation`}

Run mediation model 1 using the Structural Equation Modeling framework with {`MplusAutomation`}

Run mediation model 2 as a SEM model with {`MplusAutomation`}