## ----include = FALSE---------------------------------------------------------- knitr::opts_chunk$set( collapse = TRUE, comment = "#>", message = FALSE ) has_magmaan <- requireNamespace("magmaan", quietly = TRUE) && utils::packageVersion("magmaan") >= "0.2.0" ## ----------------------------------------------------------------------------- library(covsim) model <- "F1 =~ x1 + x2 + x3 F2 =~ x4 + x5 + x6" L <- matrix(0, 6, 2) L[1:3, 1] <- L[4:6, 2] <- 0.7 sigma <- L %*% matrix(c(1, 0.4, 0.4, 1), 2) %*% t(L) diag(sigma) <- 1 dimnames(sigma) <- list(paste0("x", 1:6), paste0("x", 1:6)) skew <- rep(2, 6) kurt <- rep(7, 6) # excess kurtosis ## ----------------------------------------------------------------------------- dists <- list( IG = calibrate_ig(sigma, skew, kurt), PLSIM = calibrate_plsim(sigma, skew, kurt, verbose = FALSE), VM = calibrate_vm(sigma, skew, kurt) ) ## ----------------------------------------------------------------------------- moments <- function(x) c(skew = mean(scale(x)^3), exkurt = mean(scale(x)^4) - 3) set.seed(1) round(sapply(dists, function(d) moments(simulate(d, N = 1e5)[[1]][, 1])), 2) ## ----eval = FALSE------------------------------------------------------------- # margins <- rep(list(list(distr = "gamma", shape = 1)), 6) # skewness 2 # dists$VITA <- calibrate_vita(margins, sigma, Nmax = 1e5, verbose = FALSE) ## ----------------------------------------------------------------------------- samples <- simulate(dists$PLSIM, nsim = 50, N = 200, seed = 1) length(samples) head(samples[[1]], 3) ## ----------------------------------------------------------------------------- set.seed(1) old_way <- rPLSIM(200, sigma, skew, kurt, reps = 50, verbose = FALSE) length(old_way$samples) ## ----------------------------------------------------------------------------- fit_tests <- function(x) { fit <- tryCatch(lavaan::sem(model, data = as.data.frame(x), estimator = "MLR"), error = function(e) NULL) if (is.null(fit)) return(c(pvalue = NA, pvalue.scaled = NA)) lavaan::fitMeasures(fit, c("pvalue", "pvalue.scaled")) } by_hand <- lapply(dists, function(cal) { samples <- simulate(cal, nsim = 50, N = 200, seed = 1) t(sapply(samples, fit_tests)) }) head(by_hand$VM, 3) ## ----------------------------------------------------------------------------- round(sapply(by_hand, function(p) colMeans(p < 0.05, na.rm = TRUE)), 2) ## ----------------------------------------------------------------------------- chisq_tests <- function(fit) { lavaan::fitMeasures(fit, c("chisq", "pvalue", "chisq.scaled", "pvalue.scaled")) } res <- simulate_estimate(dists, model, n = c(200, 500), reps = 50, extractor = chisq_tests, estimator = "MLR", seed = 1) head(res) ## ----------------------------------------------------------------------------- table(res$converged, useNA = "ifany") ## ----------------------------------------------------------------------------- reject <- function(p) mean(p < 0.05, na.rm = TRUE) aggregate(cbind(ML = pvalue, MLR = pvalue.scaled) ~ dist + n, data = res, FUN = reject) ## ----------------------------------------------------------------------------- fit_indices <- function(fit) { lavaan::fitMeasures(fit, c("rmsea", "rmsea.robust", "cfi", "cfi.robust")) } res_fi <- simulate_estimate(dists, model, n = 200, reps = 50, extractor = fit_indices, estimator = "MLR", seed = 1) aggregate(cbind(rmsea, rmsea.robust, cfi, cfi.robust) ~ dist, data = res_fi, FUN = mean) ## ----------------------------------------------------------------------------- loading_x2 <- function(fit) { pe <- lavaan::parameterEstimates(fit) pe <- pe[pe$lhs == "F1" & pe$op == "=~" & pe$rhs == "x2", ] c(est = pe$est, se = pe$se, lower = pe$ci.lower, upper = pe$ci.upper) } res_ML <- simulate_estimate(dists, model, n = 200, reps = 50, extractor = loading_x2, seed = 2) res_MLR <- simulate_estimate(dists, model, n = 200, reps = 50, extractor = loading_x2, estimator = "MLR", seed = 2) summarize <- function(r) { c(bias = mean(r$est) - 1, empirical_sd = sd(r$est), mean_se = mean(r$se), coverage = mean(r$lower < 1 & 1 < r$upper)) } tab <- rbind(sapply(split(res_ML, res_ML$dist), summarize), sapply(split(res_MLR, res_MLR$dist), summarize)) rownames(tab) <- paste(rep(c("ML", "MLR"), each = 4), rownames(tab)) round(tab, 3) ## ----------------------------------------------------------------------------- attr(res, "info") ## ----eval = FALSE------------------------------------------------------------- # res4 <- simulate_estimate(dists, model, n = c(200, 500), reps = 50, # extractor = chisq_tests, estimator = "MLR", seed = 1, # cores = 4) # all.equal(res, res4) # TRUE (on Windows, cores is set to 1) ## ----------------------------------------------------------------------------- has_magmaan ## ----eval = has_magmaan------------------------------------------------------- # mag_tests <- function(fit) { # tests <- summary(fit)$tests # setNames(tests$pvalue, paste(tests$test, tests$reference, sep = "_")) # } # # res_mag <- simulate_estimate(dists, model, n = c(200, 500), reps = 50, # extractor = mag_tests, fit = "magmaan", seed = 1) # head(res_mag) # aggregate(cbind(score_peba4, lr_peba4) ~ dist + n, data = res_mag, # FUN = reject) ## ----eval = has_magmaan------------------------------------------------------- # both <- merge(res[, c("dist", "n", "rep", "pvalue.scaled")], # res_mag[, c("dist", "n", "rep", "score_peba4")]) # aggregate(cbind(MLR = pvalue.scaled, PEBA4 = score_peba4) ~ dist, data = both, # FUN = reject) ## ----eval = has_magmaan------------------------------------------------------- # mag_loading_x2 <- function(fit) { # cf <- summary(fit)$coefficients # cf <- cf[cf$lhs == "F1" & cf$op == "=~" & cf$rhs == "x2", ] # c(est = cf$est, se = cf$se, lower = cf$ci.lower, upper = cf$ci.upper) # } # mag_fit_indices <- function(fit) { # fm <- magmaan::fit_measures(fit, lavaan_compat = "MLR") # setNames(fm$estimate, fm$index)[c("rmsea", "rmsea.robust", "cfi", "cfi.robust")] # } # # res_mag2 <- simulate_estimate(dists, model, n = 200, reps = 50, fit = "magmaan", # extractor = function(fit) c(mag_loading_x2(fit), # mag_fit_indices(fit)), # seed = 2) # round(sapply(split(res_mag2, res_mag2$dist), summarize), 3) ## ----eval = has_magmaan------------------------------------------------------- # t_lavaan <- system.time( # simulate_estimate(dists$VM, model, n = 500, reps = 50, # extractor = chisq_tests, estimator = "MLR", seed = 1))[["elapsed"]] # t_magmaan <- system.time( # simulate_estimate(dists$VM, model, n = 500, reps = 50, # extractor = mag_tests, fit = "magmaan", seed = 1))[["elapsed"]] # c(lavaan = t_lavaan, magmaan = t_magmaan, ratio = t_lavaan / t_magmaan) ## ----eval = FALSE------------------------------------------------------------- # remotes::install_github("JonasMoss/magmaan", subdir = "r-package") # remotes::install_github("JonasMoss/magmaan", subdir = "r-magmaan") ## ----------------------------------------------------------------------------- fit_wls <- function(data, model) lavaan::sem(model, data = data, estimator = "WLS") res_wls <- simulate_estimate(dists$IG, model, n = 500, reps = 20, extractor = chisq_tests, fit = fit_wls, seed = 1) reject(res_wls$pvalue) .