I am an assistant professor at KU. My primary research interests include model fit and model selection, growth curve modeling, structural equation modeling, multilevel (mixed) modeling, and missing data analysis. I am also interested in the application of these methods in education, social, developmental and health related studies.
MY CV
SEEDMC stands for search for efficient designs using Monte Carlo simulation. SEEDMC implements a systematic procedure proposed in Wu, Jia, Rhemtulla, and Little (2015) to search for efficient complete data and planned missing data designs for growth-curve modeling. SEEDMC creates the design pool, and uses functions in the R package MplusAutomation (Hallquist & Wiley, 2014) to automate the Monte Carlo simulation in Mplus (Muthén & Muthén, 1998-2012). The current version of the package (i.e., SEEDMC 1.0.0) can accommodate user specified unconditional linear and quadratic GCMs, budget and sample size constraints, and output efficient designs for any single effect and multiple effects based on a selected threshold.
Wu. W., Jia*, F., Rhemtulla, M., & Little, T. D. (2015). Search for efficient complete and planned missing data designs for analysis of change. Behavioral Research Methods. doi: 10.3758/s13428-015-0629-5.The package can be downloaded here. SEEDMC_1.0.0 or on GitHub https://github.com/fjia/SEEDMC
To install the package locally, click "Packages -> Install package(s) from local zip file...". Before running the functions, make sure that the working directory is appropriately set using setwd(), because the package will save all Mplus outputs and the R result under the working directory. The help documents for the package can be accessed by running help(package = "SEEDMC") in R.
The citation for the package is
Jia, F., & Wu, W. (2015). SEEDMC: SEarch for Efficient Designs using Monte Carlo Simulation. R package version 1.0.0.
The procedure can accommodate any mediation models that can be analyzed using Mplus (e.g., longitudinal mediation models or latent mediation models). A simple example is used to illustrate the steps involved in the procedure. In this example, there are six variables: y (outcome variable), m (mediator), x (predictor), a1, a2, and a3 are auxiliary variables. The dataset used in the example can be downloaded here.
Step 1: obtain m (e.g., 50) imputations and k (e.g., 100) bootstrap samples for each imputation. There are in total m*k (e.g., 5,000) samples. Write out the imputed datasets and the bootstrap samples as external text files which will be used in step 2 for data analysis. Two text files are also created to index the imputed datasets (implist.dat) and the bootstrap datasets (blist.dat). The step is implemented using the sas macro % mibootstep1. The macro can be downloaded here.
Below are the macro variables you need to specify in the macro.
%let nimp = 50; *number of imputed datasets;
%Let nboot = 100; *number of bootstrap samples;
%let folder = R:\users\wwei\medmiss\example; * the folder where the original data is;
%let varlist = y x m a1 a2 a3; * the list of variables that are included in the imputation model;
Step 2: fit the mediation model using Mplus to the imputed datasets. Record the final point estimate of ab. The example mplus syntax for this step is here.
Step 3: fit the mediation model using Mplus to each of the bootstrap samples. The bootstrap datasets are read in as if they were simulated datasets. The parameter estimates from each of the bootstrap datasets is saved to an external dataset using the savedata command in Mplus. In this step, one needs to figure out the number of variables/columns in the saved dataset and the column(s) that contains the parameter estimates for the target mediation effect. The example mplus syntax for this step is here.
In the example, the estimates are saved in a dataset called sample.dat. The 9th column in this dataset contains the estimates for the mediation effect.
Step 4. Computed the bias corrected confidence interval for the target mediation effect based on the outcomes from steps 2 and 3. This step is implemented using the sas macro %bcci which will print out the lower (bc_low) and upper limit (bc_up) of the interval. The macro can be downloaded here.
Note that this macro is written to compute the bias corrected confidence interval for only one mediation effect at a time. If there are multiple mediation effects in the model, you just need to run the macro for each of the effects by modifying the values for the the macro variables ab and abe correspondingly.
Below are the macro variables you need to specify
%let savedata = R:\users\wwei\medmiss\example\sample.dat; * the dataset contains the parameter estimates from each of the boostrap samples;
%let varlist2 = x1 - x33; * the list of variables in the dataset from step 3;
%let ab = x9; * the variable that represents the mediation effect. In the example, it is the 9th variable;
%let abe = 0.153; * the point estimate of ab, this value is obtained from step 2;
%let alpha = 0.95; *confidence level;
Reference.
Wu, W., & Jia, F. (2013). A new procedure to test mediation with missing data through nonparametric bootstrapping and multiple imputation. Multivariate Behavioral Research, 48(5), 663-691.
Karl Wuensch's SAS Programs Page