Pengaruh citra merek terhadap loyalitas konsumen melalui kepercayaan merek (PLS-SEM)

# Contoh kode R: PLS-SEM (seminr)
suppressPackageStartupMessages({
  library(ggplot2)
  library(patchwork)
  library(seminr)
})
set.seed(42)
# ---- Gaya grafik ----
pal <- c("#1f6feb", "#e8590c", "#2f9e44", "#ae3ec9", "#f59f00")
warna <- list(bg = "#f4f6fa", panel = "#ffffff", teks = "#1f2933", redup = "#5f6b78", grid = "#e3e8ee", bar = "#1d3557")
tata <- 1
theme_set(theme_minimal(base_size = 11) + theme(legend.position = "bottom"))

# ---- Data sintetis: skor butir Likert 1-5 dari tiga konstruk laten ----
n <- 200
a <- runif(1, 0.4, 0.7)                                         # jalur X1 -> X2 (mediator)
b <- runif(1, 0.3, 0.55)                                        # jalur X2 -> Y
cp <- runif(1, 0.1, 0.35)                                       # jalur langsung X1 -> Y
e_x1 <- rnorm(n)
e_x2 <- a * e_x1 + rnorm(n, 0, sqrt(1 - a^2))
e_y <- cp * e_x1 + b * e_x2 + rnorm(n, 0, runif(1, 0.55, 0.75))
e_y <- as.numeric(scale(e_y))
# butir Likert: muatan acak 0.65-0.9, ambang dibuat agar rata-rata butir sekitar 3.4-3.8
likert <- function(eta, k) {
  sapply(seq_len(k), function(j) {
    l <- runif(1, 0.65, 0.9)
    s <- l * eta + rnorm(length(eta), 0, sqrt(1 - l^2))
    findInterval(s, sort(c(-2, -1.1, -0.2, 0.9) + runif(1, -0.3, 0.2))) + 1
  })
}
k <- c(4, 3, 4)                                                 # jumlah butir: X1, X2, Y
d <- data.frame(likert(e_x1, k[1]), likert(e_x2, k[2]), likert(e_y, k[3]))
names(d) <- c(paste0("brand_image", "_", 1:k[1]), paste0("brand_trust", "_", 1:k[2]), paste0("brand_loyalty", "_", 1:k[3]))
write.csv(d, "data_contoh.csv", row.names = FALSE)

# ---- Model pengukuran (reflektif, mode A) dan model struktural ----
mm <- constructs(
  composite("brand_image", multi_items("brand_image_", 1:k[1]), weights = mode_A),
  composite("brand_trust", multi_items("brand_trust_", 1:k[2]), weights = mode_A),
  composite("brand_loyalty", multi_items("brand_loyalty_", 1:k[3]), weights = mode_A))
sm <- relationships(
  paths(from = "brand_image", to = c("brand_trust", "brand_loyalty")),
  paths(from = "brand_trust", to = "brand_loyalty"))
pls <- estimate_pls(data = d, measurement_model = mm, structural_model = sm)
s <- summary(pls)
print(s$loadings); print(s$reliability); print(s$validity$htmt); print(s$paths)

# ---- Bootstrap untuk signifikansi jalur dan efek tidak langsung ----
bt <- bootstrap_model(seminr_model = pls, nboot = 400, cores = 1, seed = 123)
sb <- summary(bt)
print(sb$bootstrapped_paths)
ind <- specific_effect_significance(bt, from = "brand_image", through = "brand_trust", to = "brand_loyalty", alpha = 0.05)
print(ind)
bp <- sb$bootstrapped_paths
p_jalur <- 2 * pnorm(-abs(bp[, "T Stat."]))                     # p dua sisi dari statistik t bootstrap
koef <- bp[, "Original Est."]
nm_j <- rownames(bp)                                            # "X1  ->  X2" dst.
j_a <- grep("^brand_image .*-> *brand_trust$", nm_j); j_b <- grep("^brand_trust .*-> *brand_loyalty$", nm_j); j_c <- grep("^brand_image .*-> *brand_loyalty$", nm_j)
rel <- s$reliability
ave <- rel[, "AVE"]; rhoc <- rel[, "rhoC"]
r2 <- s$paths["R^2", c("brand_trust", "brand_loyalty")]
htmt_maks <- max(s$validity$htmt, na.rm = TRUE)
ind_est <- unname(ind["Original Est."]); ind_lo <- unname(ind["2.5% CI"]); ind_hi <- unname(ind["97.5% CI"])


# ---- Grafik ----
bintang <- function(p) ifelse(p < 0.001, "***", ifelse(p < 0.01, "**", ifelse(p < 0.05, "*", " ns")))
bungkus <- function(s) paste(head(strwrap(s, 17), 2), collapse = "\n")   # nama konstruk maks. 2 baris
nd <- data.frame(x = c(0, 1, 2), y = c(0, 1, 0), lab = c(bungkus("brand_image"), bungkus("brand_trust"), bungkus("brand_loyalty")),
                 r2 = c("", sprintf("R2 = %.2f", r2[1]), sprintf("R2 = %.2f", r2[2])))
pan <- data.frame(x = c(0.2, 1.2, 0.36), y = c(0.2, 0.8, 0), xend = c(0.8, 1.8, 1.64), yend = c(0.8, 0.2, 0),
                  lab = sprintf("%.2f%s", koef[c(j_a, j_b, j_c)], bintang(p_jalur[c(j_a, j_b, j_c)])),
                  lx = c(0.33, 1.67, 1), ly = c(0.6, 0.6, -0.1))
p1 <- ggplot() +
  geom_segment(data = pan, aes(x, y, xend = xend, yend = yend), colour = warna$redup, linewidth = 0.9,
               arrow = arrow(length = unit(0.14, "inches"), type = "closed")) +
  geom_text(data = pan, aes(lx, ly, label = lab), colour = warna$teks, size = 3.4, fontface = "bold") +
  geom_tile(data = nd, aes(x, y), width = 0.72, height = 0.38, fill = pal[1:3]) +
  geom_text(data = nd, aes(x, y + 0.04, label = lab), colour = warna$panel, fontface = "bold", size = 3.1, lineheight = 0.9) +
  geom_text(data = nd, aes(x, y - 0.13, label = r2), colour = warna$panel, size = 2.8) +
  annotate("text", x = 1, y = -0.34, label = sprintf("Indirect effect = %.3f [%.3f, %.3f]", ind_est, ind_lo, ind_hi),
           colour = warna$teks, size = 3.2) +
  coord_cartesian(xlim = c(-0.4, 2.4), ylim = c(-0.42, 1.22)) +
  theme_void() + theme(plot.background = element_rect(fill = warna$panel, colour = NA),
                       plot.title = element_text(face = "bold", colour = warna$teks, margin = margin(4, 0, 4, 6))) +
  labs(title = "PLS path model (bootstrap, 400 resamples)")
ld <- s$loadings
dl <- data.frame(butir = rownames(ld), muatan = apply(ld, 1, function(v) v[v != 0][1]))
# nama konstruk pendek (<= 18 huruf): kata sambung dibuang dan kata panjang disingkat bila perlu
singkat <- function(s, w = 18) {
  k <- strsplit(s, " ")[[1]]
  if (nchar(s) > w) k <- k[!tolower(k) %in% c("of", "to", "the", "and", "for", "in", "on", "with")]
  i <- length(k)
  while (nchar(paste(k, collapse = " ")) > w && i >= 1) {
    if (nchar(k[i]) > 5) k[i] <- paste0(substr(k[i], 1, 4), ".")
    i <- i - 1
  }
  substr(paste(k, collapse = " "), 1, w)
}
nm_k <- make.unique(unname(sapply(c("brand_image", "brand_trust", "brand_loyalty"), singkat)), sep = " ")
dl$konstruk <- rep(nm_k, k)
dl$label <- paste0(dl$konstruk, " #", sub(".*_", "", dl$butir))
dl$label <- factor(dl$label, levels = rev(unique(dl$label)))
dl$konstruk <- factor(dl$konstruk, levels = nm_k)
p2 <- ggplot(dl, aes(muatan, label, fill = konstruk)) +
  geom_col(width = 0.7) +
  geom_vline(xintercept = 0.7, linetype = "dashed", colour = warna$teks) +
  geom_text(aes(label = sprintf("%.2f", muatan)), hjust = -0.15, colour = warna$teks, size = 2.7) +
  scale_fill_manual(values = pal[1:3], name = NULL) +
  scale_x_continuous(limits = c(0, 1.1), breaks = seq(0, 1, 0.2)) +
  labs(title = "Outer loadings (threshold 0.7)", x = "Loading", y = NULL)
dq <- data.frame(konstruk = factor(rep(nm_k, 2), levels = nm_k),
                 ukuran = rep(c("AVE", "rhoC"), each = 3), nilai = c(ave[c("brand_image", "brand_trust", "brand_loyalty")], rhoc[c("brand_image", "brand_trust", "brand_loyalty")]))
p3 <- ggplot(dq, aes(konstruk, nilai, fill = ukuran)) +
  geom_col(position = position_dodge(0.75), width = 0.7) +
  geom_hline(yintercept = c(0.5, 0.7), linetype = "dashed", colour = warna$redup) +
  scale_fill_manual(values = pal[4:5], name = NULL) +
  scale_y_continuous(limits = c(0, 1)) +
  labs(title = "Convergent validity and reliability", x = NULL, y = "Value")
p_hasil <- if (tata == 1) p1 / (p2 | p3) + plot_layout(heights = c(1.1, 1)) else (p1 | p2) / p3 + plot_layout(heights = c(1.4, 1))

print(p_hasil)