Fully Latent Principal Stratification (FLPS)1

Fully Latent Principal Stratification (FLPS) is an extension of principal stratification.

Installation

Install the latest release from CRAN or git repository:

devtools::install_github("sooyongl/flps")
install.packages("flps")
## Version: 1.0.0
## 
## It is a demo.
## Acknowledgements. It is supported by the Institute of Education Sciences, U.S. Department of Education, through Grant R305D210036.

  • Documentation is available here.

  • For compiling errors on Windows, see the relevant guide.

Basic working example

Load Example Data

  • binary: a data frame containing all the data for FLPS. It is used in runFLPS function.
  • This data will be converted to a list of data for rstan package.
  • For latent variable models, Rasch, 2PL, GRM, SEM (one-factor CFA), and mixture models (LCA and LPA) are available.
  • Multilevel structure will be supported soon.
data(binary)
# Input data frame
data.table::data.table(binary)
##       schid   id sex race pretest stdscore    cm_sex   cm_race cm_pretest
##    1:     1 2383   0    1      20  -0.3296 0.4406780 0.9322034   14.81356
##    2:     1 2384   1    0       8   1.1597 0.4406780 0.9322034   14.81356
##    3:     1 2385   0    1      14  -0.7385 0.4406780 0.9322034   14.81356
##    4:     1 2387   0    1      12  -1.3518 0.4406780 0.9322034   14.81356
##    5:     1 2388   0    1       6  -1.2057 0.4406780 0.9322034   14.81356
##   ---                                                                    
## 4762:    63 4761   0    1      16   0.6457 0.4860558 0.3944223   17.03187
## 4763:    63 4763   0    0      11   0.3168 0.4860558 0.3944223   17.03187
## 4764:    63 4764   0    0      18  -0.4114 0.4860558 0.3944223   17.03187
## 4765:    63 4765   1    0      31   2.2116 0.4860558 0.3944223   17.03187
## 4766:    63 4766   1    1      13  -0.0668 0.4860558 0.3944223   17.03187
##       cm_stdscore trt      Y q1 q2 q3 q4 q5 q6 q7 q8 q9 q10 q11 q12 q13 q14 q15
##    1:  -0.4068119   1 -0.487  0 NA  1  1  1  1  1 NA NA  NA  NA  NA  NA  NA  NA
##    2:  -0.4068119   1  0.487  1 NA  1  1  1  1  1  1  1   1   1   1  NA  NA  NA
##    3:  -0.4068119   1 -1.073  0 NA  1  1  0 NA NA NA NA  NA  NA  NA  NA  NA  NA
##    4:  -0.4068119   1 -1.087  0 NA  1  1  1 NA NA NA NA  NA  NA  NA  NA  NA  NA
##    5:  -0.4068119   1 -1.171  0 NA  1  1 NA NA NA NA NA  NA  NA  NA  NA  NA  NA
##   ---                                                                          
## 4762:   0.0997502   0 -0.114  0 NA NA NA NA NA NA NA NA  NA  NA  NA  NA  NA  NA
## 4763:   0.0997502   0 -0.438  0 NA NA NA NA NA NA NA NA  NA  NA  NA  NA  NA  NA
## 4764:   0.0997502   0 -1.047  0 NA NA NA NA NA NA NA NA  NA  NA  NA  NA  NA  NA
## 4765:   0.0997502   0  0.765  0 NA NA NA NA NA NA NA NA  NA  NA  NA  NA  NA  NA
## 4766:   0.0997502   0  0.539  0 NA NA NA NA NA NA NA NA  NA  NA  NA  NA  NA  NA
##       q16 q17 q18 q19 q20
##    1:  NA  NA  NA  NA  NA
##    2:  NA  NA  NA  NA  NA
##    3:  NA  NA  NA  NA  NA
##    4:  NA  NA  NA  NA  NA
##    5:  NA  NA  NA  NA  NA
##   ---                    
## 4762:  NA  NA  NA  NA  NA
## 4763:  NA  NA  NA  NA  NA
## 4764:  NA  NA  NA  NA  NA
## 4765:  NA  NA  NA  NA  NA
## 4766:  NA  NA  NA  NA  NA

Model Fitting with FLPS

  • runFLPS() internally transforms binary data into the format suitable for rstan, subsequently executing FLPS.

  • To avoid re-compiling the Stan code each time, pre-compile it using modelBuilder(), which stores the stanmodel object in the flps directory, accelerating subsequent analyses.

  • Once the Stan model is compiled, use importModel() to bring in the compiled Stan code. This code can then be provided to the compiled_stan argument in runFLPS. If this step is omitted, runFLPS() will compile the Stan code during each execution of FLPS.

modelBuilder(type = "rasch")
complied_stan <- importModel(type = "rasch")
  • In case of errors, try the latest rstan and StanHeaders packages.
remove.packages(c("rstan", "StanHeaders"))
install.packages("rstan", repos = c("https://mc-stan.org/r-packages/", getOption("repos")))

Now, execute your FLPS model. Given the time-intensive nature of the process, chains and iterations have been initially limited to 1 and 5000, respectively. It is advisable to increase these values for your specific research needs.

# Subset of data: 1000 students
binary <- binary[c(sample(which(binary$trt == 1), 250), 
                   sample(which(binary$trt == 0), 250)),]

res <- runFLPS(
  inp_data = binary,
  # complied_stan = complied # if necessary
  outcome = "Y",
  trt = "trt",
  covariate = c("sex","race","pretest","stdscore"),
  lv_type = "rasch",
  lv_model = "F =~ q1 + q2 + q3 + q4 + q5 + q6 + q7 + q8 + q9 + q10",
  stan_options = list(iter = 5000, cores = 1, chains = 2)
)

Results

Retrieve summaries and visualize results with the following:

summary(res, type = "causal")

The flps_plot() shows the plot related to FLPS models

flps_plot(res, type = "causal")

flps_plot(res, type = "latent")

  1. Acknowledgements. This package is supported by the Institute of Education Sciences, U.S. Department of Education, through Grant R305D210036.↩︎