context("impute_covariance_matrix") set.seed(20190513) K <- 20 N <- sum(1:K) dat <- data.frame( study = rep(LETTERS[1:K], 1:K), yi = rnorm(N), vi = rchisq(N, df = 2), ti = sample(1:(10 * N), N), si = sample(letters[1:3], N, replace = TRUE) ) dat$v_study <- unsplit(tapply(dat$vi, dat$study, mean), dat$study) dat_scramble <- dat[sample.int(nrow(dat)),] test_that("impute_covariance_matrix() is deprecated.", { expect_warning( impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6) ) }) test_that("impute_covariance_matrix error messages and missing argument handling are correct.", { expect_error(impute_covariance_matrix(vi = dat$vi, cluster = dat$study)) expect_error(impute_covariance_matrix(vi = dat$vi, cluster = dat$study, ti = dat$ti)) expect_error(impute_covariance_matrix(vi = dat$vi, cluster = dat$study, ar1 = 0.8)) V1 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6) V2 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6, ti = dat$ti) V3 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6, ti = dat$ti, ar1 = 0) V4 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, ti = dat$ti, ar1 = 0.5) V5 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.0, ti = dat$ti, ar1 = 0.5) expect_equal(V1, V2) expect_equal(V1, V3) expect_equal(V4, V5) }) test_that("impute_covariance_matrix returns correct correlations.", { r <- 0.7 V_single_r <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = r) r_list <- rbeta(K, 2, 2) V_multiple_r <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = r_list) check_correlation <- function(M, r) if (nrow(M) > 1) max(abs(cov2cor(M)[lower.tri(M)] - r)) else 0 check_singles <- sapply(V_single_r, check_correlation, r = r) expect_true(all(check_singles < 10^-14)) check_multiples <- Map(check_correlation, M = V_multiple_r, r = r_list) expect_true(all(check_multiples < 10^-14)) dat_scramble <- dat[sample(nrow(dat)),] V_mat <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = r) expect_equal(dat_scramble$vi, diag(V_mat)) V_resorted <- V_mat[order(dat_scramble$study), order(dat_scramble$study)] dat_unscramble <- dat_scramble[order(dat_scramble$study),] V_unscramble <- impute_covariance_matrix(vi = dat_unscramble$vi, cluster = dat_unscramble$study, r = r) expect_equal(V_resorted, unblock(V_unscramble)) }) test_that("impute_covariance_matrix works with AR1 argument.", { ar1 <- 0.5 V_mat <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, ti = 1:nrow(dat), ar1 = ar1) r_mat <- lapply(V_mat, cov2cor) r_sums <- sapply(r_mat, sum) check <- sapply(2:K, function(k) k + 2 * sum((1:(k-1)) * ar1^((k-1):1))) expect_equal(r_sums[-1], check, check.attributes = FALSE) V_mat_big <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, ti = 1:nrow(dat), ar1 = ar1, return_list = FALSE) r_mat_big <- cov2cor(V_mat_big) expect_equal(sum(r_mat_big), sum(check) + 1) v_sums <- sapply(V_mat, sum) expect_equal(sum(v_sums), sum(V_mat_big)) }) test_that("impute_covariance_matrix works with subgroup argument.", { X <- model.matrix(~ si + 0, data = dat) X_list <- by(X, dat$study, as.matrix) check_diag <- function(v, x) { XVX <- t(x) %*% v %*% x all.equal(sum(XVX), sum(diag(XVX))) } check_all_diag <- function(v_list,x_list) { all(mapply(check_diag , v = v_list, x = x_list)) } V1 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6, subgroup = dat$si) V2 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6, ti = dat$ti, subgroup = dat$si) V3 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6, ti = dat$ti, ar1 = 0, subgroup = dat$si) V4 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, ti = dat$ti, ar1 = 0.5, subgroup = dat$si) V5 <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.0, ti = dat$ti, ar1 = 0.5, subgroup = dat$si) expect_true(check_all_diag(V1, X_list)) expect_true(check_all_diag(V2, X_list)) expect_true(check_all_diag(V3, X_list)) expect_true(check_all_diag(V4, X_list)) expect_true(check_all_diag(V5, X_list)) X_scramble <- model.matrix(~ si + 0, data = dat_scramble) V1 <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = 0.6, subgroup = dat_scramble$si) V2 <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = 0.6, ti = dat_scramble$ti, subgroup = dat_scramble$si) V3 <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = 0.6, ti = dat_scramble$ti, ar1 = 0, subgroup = dat_scramble$si) V4 <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, ti = dat_scramble$ti, ar1 = 0.5, subgroup = dat_scramble$si) V5 <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = 0.0, ti = dat_scramble$ti, ar1 = 0.5, subgroup = dat_scramble$si) expect_true(check_diag(V1, X_scramble)) expect_true(check_diag(V2, X_scramble)) expect_true(check_diag(V3, X_scramble)) expect_true(check_diag(V4, X_scramble)) expect_true(check_diag(V5, X_scramble)) }) test_that("impute_covariance_matrix works with smooth argument.", { V1 <- impute_covariance_matrix(vi = dat$v_study, cluster = dat$study, r = 0.6) V2 <- impute_covariance_matrix(vi = dat$v_study, cluster = dat$study, r = 0.6, ti = dat$ti) V3 <- impute_covariance_matrix(vi = dat$v_study, cluster = dat$study, r = 0.6, ti = dat$ti, ar1 = 0) V4 <- impute_covariance_matrix(vi = dat$v_study, cluster = dat$study, ti = dat$ti, ar1 = 0.5) V5 <- impute_covariance_matrix(vi = dat$v_study, cluster = dat$study, r = 0.0, ti = dat$ti, ar1 = 0.5) V1s <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6, smooth_vi = TRUE) V2s <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6, ti = dat$ti, smooth_vi = TRUE) V3s <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.6, ti = dat$ti, ar1 = 0, smooth_vi = TRUE) V4s <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, ti = dat$ti, ar1 = 0.5, smooth_vi = TRUE) V5s <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = 0.0, ti = dat$ti, ar1 = 0.5, smooth_vi = TRUE) expect_equal(V1, V1s) expect_equal(V2, V2s) expect_equal(V3, V3s) expect_equal(V4, V4s) expect_equal(V5, V5s) V1 <- impute_covariance_matrix(vi = dat_scramble$v_study, cluster = dat_scramble$study, r = 0.6) V2 <- impute_covariance_matrix(vi = dat_scramble$v_study, cluster = dat_scramble$study, r = 0.6, ti = dat_scramble$ti) V3 <- impute_covariance_matrix(vi = dat_scramble$v_study, cluster = dat_scramble$study, r = 0.6, ti = dat_scramble$ti, ar1 = 0) V4 <- impute_covariance_matrix(vi = dat_scramble$v_study, cluster = dat_scramble$study, ti = dat_scramble$ti, ar1 = 0.5) V5 <- impute_covariance_matrix(vi = dat_scramble$v_study, cluster = dat_scramble$study, r = 0.0, ti = dat_scramble$ti, ar1 = 0.5) V1s <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = 0.6, smooth_vi = TRUE) V2s <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = 0.6, ti = dat_scramble$ti, smooth_vi = TRUE) V3s <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = 0.6, ti = dat_scramble$ti, ar1 = 0, smooth_vi = TRUE) V4s <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, ti = dat_scramble$ti, ar1 = 0.5, smooth_vi = TRUE) V5s <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = 0.0, ti = dat_scramble$ti, ar1 = 0.5, smooth_vi = TRUE) expect_equal(V1, V1s) expect_equal(V2, V2s) expect_equal(V3, V3s) expect_equal(V4, V4s) expect_equal(V5, V5s) }) test_that("impute_covariance_matrix works with unobserved factors.", { K <- 10 N <- sum(1:K) dat <- data.frame(study = rep(LETTERS[1:K], 1:K), yi = rnorm(N), vi = rchisq(N, df = 2)) levels(dat$study) <- LETTERS[1:(K + 3)] r <- 0.7 V_single_r <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = r) r_list <- rbeta(K, 2, 2) V_multiple_r <- impute_covariance_matrix(vi = dat$vi, cluster = dat$study, r = r_list) check_correlation <- function(M, r) if (nrow(M) > 1) max(abs(cov2cor(M)[lower.tri(M)] - r)) else 0 check_singles <- sapply(V_single_r, check_correlation, r = r) expect_true(all(check_singles < 10^-14)) dat_scramble <- dat[sample(nrow(dat)),] V_mat <- impute_covariance_matrix(vi = dat_scramble$vi, cluster = dat_scramble$study, r = r) expect_equal(dat_scramble$vi, diag(V_mat)) V_resorted <- V_mat[order(dat_scramble$study), order(dat_scramble$study)] dat_unscramble <- dat_scramble[order(dat_scramble$study),] V_unscramble <- impute_covariance_matrix(vi = dat_unscramble$vi, cluster = dat_unscramble$study, r = r) expect_equal(V_resorted, unblock(V_unscramble)) }) test_that("impute_covariance_matrix works with missing variances.", { skip_if_not_installed("robumeta") data(corrdat, package = "robumeta") dat_miss <- corrdat dat_miss$var[sample.int(nrow(corrdat), size = round(nrow(corrdat) / 10))] <- NA V_missing <- impute_covariance_matrix(dat_miss$var, cluster = dat_miss$studyid, r = 0.8, return_list = FALSE) non_missing_rows <- !is.na(dat_miss$var) V_missing <- V_missing[non_missing_rows, non_missing_rows] dat_complete <- subset(dat_miss, !is.na(var)) V_complete <- impute_covariance_matrix(dat_complete$var, cluster = dat_complete$studyid, r = 0.8, return_list = FALSE) expect_equal(V_missing, V_complete) }) test_that("pattern_covariance_matrix works.", { skip_if_not_installed("robumeta") skip_if_not_installed("metafor") data(oswald2013, package = "robumeta") dat <- metafor::escalc(data = oswald2013, measure = "ZCOR", ri = R, ni = N) # make a patterned correlation matrix p_levels <- levels(dat$Crit.Cat) r_pattern <- 0.7^as.matrix(dist(1:length(p_levels))) diag(r_pattern) <- seq(0.75, 0.95, length.out = 6) rownames(r_pattern) <- colnames(r_pattern) <- p_levels # impute the covariance matrix using patterned correlations V_list <- pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern, smooth_vi = TRUE) r_pattern1 <- r_pattern2 <- r_pattern # Recreate a constant covariance matrix r_pattern1[1:6,1:6] <- 0.6 V_pattern_const <- pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern1, smooth_vi = FALSE) V_impute_const <- impute_covariance_matrix(vi = dat$vi, cluster = dat$Study, r = 0.6) expect_equal(V_pattern_const, V_impute_const, check.attributes = FALSE) # Patterns work with excluded categories exclude <- 3:5 r_pattern2[3:5,] <- 0.3 r_pattern2[,3:5] <- 0.3 V_pattern_exclude <- pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern[-(3:5),-(3:5)], r = 0.3, smooth_vi = FALSE) V_pattern_full <- pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern2, smooth_vi = FALSE) expect_equal(V_pattern_exclude, V_pattern_full) # Patterns work with extra categories levs <- c(p_levels, LETTERS[1:5]) r_pattern3 <- matrix(-4, nrow = length(levs), ncol = length(levs), dimnames = list(levs, levs)) r_pattern3[p_levels,p_levels] <- r_pattern[p_levels, p_levels] V_pattern_extra <- pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern3, r = 0.3, smooth_vi = TRUE) expect_equal(V_pattern_extra, V_list) V_pattern_extra_exclude <- pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern3[-(3:5),-(3:5)], r = 0.3, smooth_vi = FALSE) expect_equal(V_pattern_extra_exclude, V_pattern_full) VS_pattern_extra_exclude <- pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern3[-(3:5),-(3:5)], r = 0.3, subgroup = dat$IAT.Focus, smooth_vi = FALSE, return_list = FALSE) VS_pattern_full <- pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern2, subgroup = dat$IAT.Focus, smooth_vi = FALSE, return_list = FALSE) expect_equal(VS_pattern_extra_exclude, VS_pattern_full) expect_warning( pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern3[-(3:5),-(3:5)], r = 4, subgroup = dat$IAT.Focus, smooth_vi = FALSE, return_list = FALSE) ) V_npd <- pattern_covariance_matrix(vi = dat$vi, cluster = dat$Study, pattern_level = dat$Crit.Cat, r_pattern = r_pattern3[-(3:5),-(3:5)], r = 4, subgroup = dat$IAT.Focus, smooth_vi = FALSE, return_list = FALSE, check = FALSE) expect_true(inherits(V_npd, "matrix")) # pattern_covariance_matrix works with missing entries dat_miss <- dat dat_miss$vi[sample.int(nrow(dat), size = round(nrow(dat) / 10))] <- NA V_missing <- pattern_covariance_matrix(dat_miss$vi, cluster = dat_miss$Study, pattern_level = dat_miss$Crit.Cat, r_pattern = r_pattern, return_list = FALSE) non_missing_rows <- !is.na(dat_miss$vi) V_missing <- V_missing[non_missing_rows, non_missing_rows] dat_complete <- subset(dat_miss, !is.na(vi)) V_complete <- pattern_covariance_matrix(dat_complete$vi, cluster = dat_complete$Study, pattern_level = dat_complete$Crit.Cat, r_pattern = r_pattern, return_list = FALSE) expect_equal(V_missing, V_complete) dat_miss$Crit.Cat[sample.int(nrow(dat), size = round(nrow(dat) / 10))] <- NA expect_error( pattern_covariance_matrix(dat_miss$vi, cluster = dat_miss$Study, pattern_level = dat_miss$Crit.Cat, r_pattern = r_pattern, return_list = TRUE, check_PD = FALSE) ) })