| Title: | Create Phantom Variables in Structural Equation Models for Sensitivity Analyses |
|---|---|
| Description: | Create phantom variables, which are variables that were not observed, for the purpose of sensitivity analyses for structural equation models. The package makes it easier for a user to test different combinations of covariances between the phantom variable(s) and observed variables. The package may be used to assess a model's or effect's sensitivity to temporal bias (e.g., if cross-sectional data were collected) or confounding bias. |
| Authors: | Alexis Georgeson [aut, cre, cph]
|
| Maintainer: | Alexis Georgeson <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 1.0.0.0 |
| Built: | 2025-02-23 03:02:29 UTC |
| Source: | https://github.com/argeorgeson/phantsem |
ghost_par_ests() Selects certain parameter estimates from the output of the sensitivity analysis.
ghost_par_ests(step3, parameter_label, remove_NA = FALSE)ghost_par_ests(step3, parameter_label, remove_NA = FALSE)
step3 |
The object returned from SA_step3. |
parameter_label |
The label used for the parameter in the lavaan code. |
remove_NA |
Remove rows for combinations of phantom variable parameters that resulted in inadmissable solutions in lavaan. |
A dataframe of the parameter estimates from the lavaan model.
# example code covmatrix <- matrix(c( 0.25, 0.95, 0.43, 0.95, 8.87, 2.66, 0.43, 2.66, 10.86 ), nrow = 3, byrow = TRUE) colnames(covmatrix) <- c("X", "M2", "Y2") # lavann syntax for observed model observed <- " M2 ~ X Y2 ~ M2+X " # lavaan output obs_output <- lavaan::sem(model = observed, sample.cov = covmatrix, sample.nobs = 200) # lavaan syntax for phantom variable model phantom <- " M2 ~ M1 + Y1 + a*X Y2 ~ M1 + Y1 + b*M2 + cp*X " Step1 <- SA_step1( lavoutput = obs_output, mod_obs = observed, mod_phant = phantom ) phantom_assignment <- list( "CovM1X" = 0, "CovY1M1" = "CovY2M2", "CovY1X" = 0, "VarM1" = 1, "VarY1" = 1, "CovM1M2" = seq(.4, .6, .1), "CovY1Y2" = "CovM1M2", "CovY1M2" = seq(.2, .4, .1), "CovM1Y2" = "CovY1M2" ) Step2 <- SA_step2( phantom_assignment = phantom_assignment, step1 = Step1 ) Step3 <- SA_step3( step2 = Step2, n = 200 ) b_results <- ghost_par_ests( step3 = Step3, parameter_label = "b", remove_NA = TRUE )# example code covmatrix <- matrix(c( 0.25, 0.95, 0.43, 0.95, 8.87, 2.66, 0.43, 2.66, 10.86 ), nrow = 3, byrow = TRUE) colnames(covmatrix) <- c("X", "M2", "Y2") # lavann syntax for observed model observed <- " M2 ~ X Y2 ~ M2+X " # lavaan output obs_output <- lavaan::sem(model = observed, sample.cov = covmatrix, sample.nobs = 200) # lavaan syntax for phantom variable model phantom <- " M2 ~ M1 + Y1 + a*X Y2 ~ M1 + Y1 + b*M2 + cp*X " Step1 <- SA_step1( lavoutput = obs_output, mod_obs = observed, mod_phant = phantom ) phantom_assignment <- list( "CovM1X" = 0, "CovY1M1" = "CovY2M2", "CovY1X" = 0, "VarM1" = 1, "VarY1" = 1, "CovM1M2" = seq(.4, .6, .1), "CovY1Y2" = "CovM1M2", "CovY1M2" = seq(.2, .4, .1), "CovM1Y2" = "CovY1M2" ) Step2 <- SA_step2( phantom_assignment = phantom_assignment, step1 = Step1 ) Step3 <- SA_step3( step2 = Step2, n = 200 ) b_results <- ghost_par_ests( step3 = Step3, parameter_label = "b", remove_NA = TRUE )
SA_lookup() is used to look up the sensitivity analysis results for a two-wave mediation model when provided with the cross-sectional correlations.
SA_lookup(CorXM, CorXY, CorMY)SA_lookup(CorXM, CorXY, CorMY)
CorXM |
The observed correlation between predictor X and mediator M. |
CorXY |
The observed correlation between predictor X and outcome Y. |
CorMY |
The observed correlation between mediator M and outcome Y. |
Results of a sensitivity analysis with varying cross-lagged and autoregressive correlations.
# specify correlations xm <- .2 xy <- .3 my <- .4 output <- SA_lookup( CorXM = xm, CorXY = xy, CorMY = my )# specify correlations xm <- .2 xy <- .3 my <- .4 output <- SA_lookup( CorXM = xm, CorXY = xy, CorMY = my )
SA_step1() is used to identify the phantom variables and generate names for their covariance parameters. The output of this function will be used in SA_step2().
SA_step1(lavoutput, mod_obs, mod_phant)SA_step1(lavoutput, mod_obs, mod_phant)
lavoutput |
The lavaan output object output from lavaan functions sem() or lavaan() when fitting your observed model. |
mod_obs |
A lavaan syntax for the observed model. |
mod_phant |
A lavaan syntax for the phantom variable model. |
a list containing the names of all phantom covariance parameters.
# covariance matrix covmatrix <- matrix(c( 0.25, 0.95, 0.43, 0.95, 8.87, 2.66, 0.43, 2.66, 10.86 ), nrow = 3, byrow = TRUE) colnames(covmatrix) <- c("X", "M2", "Y2") # lavann syntax for observed model observed <- " M2 ~ X Y2 ~ M2+X " # lavaan output obs_output <- lavaan::sem(model = observed, sample.cov = covmatrix, sample.nobs = 200) # lavaan syntax for phantom variable model phantom <- " M2 ~ M1 + Y1 + a*X Y2 ~ M1 + Y1 + b*M2 + cp*X " Step1 <- SA_step1( lavoutput = obs_output, mod_obs = observed, mod_phant = phantom )# covariance matrix covmatrix <- matrix(c( 0.25, 0.95, 0.43, 0.95, 8.87, 2.66, 0.43, 2.66, 10.86 ), nrow = 3, byrow = TRUE) colnames(covmatrix) <- c("X", "M2", "Y2") # lavann syntax for observed model observed <- " M2 ~ X Y2 ~ M2+X " # lavaan output obs_output <- lavaan::sem(model = observed, sample.cov = covmatrix, sample.nobs = 200) # lavaan syntax for phantom variable model phantom <- " M2 ~ M1 + Y1 + a*X Y2 ~ M1 + Y1 + b*M2 + cp*X " Step1 <- SA_step1( lavoutput = obs_output, mod_obs = observed, mod_phant = phantom )
SA_step2() is used to assign values to the phantom covariances. There are three options for assigning values to the phantom covariances: 1. Fix phantom covariance to a numeric value (i.e., 0 or 1), 2. Fix phantom covariance to be equal to another covariance, or 3. Test different values for the phantom covariance.
SA_step2(phantom_assignment, step1)SA_step2(phantom_assignment, step1)
phantom_assignment |
A list of all phantom parameter names (copied from |
step1 |
The output object created in |
A list containing test covariance matrices that the phantom model will be fit to.
covmatrix <- matrix(c( 0.25, 0.95, 0.43, 0.95, 8.87, 2.66, 0.43, 2.66, 10.86 ), nrow = 3, byrow = TRUE) colnames(covmatrix) <- c("X", "M2", "Y2") # lavann syntax for observed model observed <- " M2 ~ X Y2 ~ M2+X " # lavaan output obs_output <- lavaan::sem(model = observed, sample.cov = covmatrix, sample.nobs = 200) # lavaan syntax for phantom variable model phantom <- " M2 ~ M1 + Y1 + a*X Y2 ~ M1 + Y1 + b*M2 + cp*X " Step1 <- SA_step1( lavoutput = obs_output, mod_obs = observed, mod_phant = phantom ) phantom_assignment <- list( "CovM1X" = 0, "CovY1M1" = "CovY2M2", "CovY1X" = 0, "VarM1" = 1, "VarY1" = 1, "CovM1M2" = seq(0, .6, .1), "CovY1Y2" = "CovM1M2", "CovY1M2" = seq(-.6, .6, .1), "CovM1Y2" = "CovY1M2" ) Step2 <- SA_step2( phantom_assignment = phantom_assignment, step1 = Step1 )covmatrix <- matrix(c( 0.25, 0.95, 0.43, 0.95, 8.87, 2.66, 0.43, 2.66, 10.86 ), nrow = 3, byrow = TRUE) colnames(covmatrix) <- c("X", "M2", "Y2") # lavann syntax for observed model observed <- " M2 ~ X Y2 ~ M2+X " # lavaan output obs_output <- lavaan::sem(model = observed, sample.cov = covmatrix, sample.nobs = 200) # lavaan syntax for phantom variable model phantom <- " M2 ~ M1 + Y1 + a*X Y2 ~ M1 + Y1 + b*M2 + cp*X " Step1 <- SA_step1( lavoutput = obs_output, mod_obs = observed, mod_phant = phantom ) phantom_assignment <- list( "CovM1X" = 0, "CovY1M1" = "CovY2M2", "CovY1X" = 0, "VarM1" = 1, "VarY1" = 1, "CovM1M2" = seq(0, .6, .1), "CovY1Y2" = "CovM1M2", "CovY1M2" = seq(-.6, .6, .1), "CovM1Y2" = "CovY1M2" ) Step2 <- SA_step2( phantom_assignment = phantom_assignment, step1 = Step1 )
SA_step3() computes the parameter estimates in your phantom model defined in step 1 for the different values provided.
SA_step3(step2, n)SA_step3(step2, n)
step2 |
The object returned from SA_step2. |
n |
The sample size. |
A list of parameter estimates from each test covariance matrix.
#' @examples covmatrix <- matrix(c( 0.25, 0.95, 0.43, 0.95, 8.87, 2.66, 0.43, 2.66, 10.86 ), nrow = 3, byrow = TRUE) colnames(covmatrix) <- c("X", "M2", "Y2") # lavann syntax for observed model observed <- " M2 ~ X Y2 ~ M2+X " # lavaan output obs_output <- lavaan::sem(model = observed, sample.cov = covmatrix, sample.nobs = 200) # lavaan syntax for phantom variable model phantom <- " M2 ~ M1 + Y1 + a*X Y2 ~ M1 + Y1 + b*M2 + cp*X " Step1 <- SA_step1( lavoutput = obs_output, mod_obs = observed, mod_phant = phantom ) phantom_assignment <- list( "CovM1X" = 0, "CovY1M1" = "CovY2M2", "CovY1X" = 0, "VarM1" = 1, "VarY1" = 1, "CovM1M2" = seq(.4, .6, .1), "CovY1Y2" = "CovM1M2", "CovY1M2" = seq(.1, .3, .1), "CovM1Y2" = "CovY1M2" ) Step2 <- SA_step2( phantom_assignment = phantom_assignment, step1 = Step1 ) Step3 <- SA_step3( step2 = Step2, n = 200 )#' @examples covmatrix <- matrix(c( 0.25, 0.95, 0.43, 0.95, 8.87, 2.66, 0.43, 2.66, 10.86 ), nrow = 3, byrow = TRUE) colnames(covmatrix) <- c("X", "M2", "Y2") # lavann syntax for observed model observed <- " M2 ~ X Y2 ~ M2+X " # lavaan output obs_output <- lavaan::sem(model = observed, sample.cov = covmatrix, sample.nobs = 200) # lavaan syntax for phantom variable model phantom <- " M2 ~ M1 + Y1 + a*X Y2 ~ M1 + Y1 + b*M2 + cp*X " Step1 <- SA_step1( lavoutput = obs_output, mod_obs = observed, mod_phant = phantom ) phantom_assignment <- list( "CovM1X" = 0, "CovY1M1" = "CovY2M2", "CovY1X" = 0, "VarM1" = 1, "VarY1" = 1, "CovM1M2" = seq(.4, .6, .1), "CovY1Y2" = "CovM1M2", "CovY1M2" = seq(.1, .3, .1), "CovM1Y2" = "CovY1M2" ) Step2 <- SA_step2( phantom_assignment = phantom_assignment, step1 = Step1 ) Step3 <- SA_step3( step2 = Step2, n = 200 )