pkgdown/extra-head.html

Skip to contents

Use LikertMakeR to replicate data from SEM results in a published report

Hume Winzar

2026-05-21

Source: vignettes/dwivedi_replication.qmd

tl;dr

  • Goal: Demonstrate how the LikertMakeR package (Winzar 2026) can reconstruct survey data using only published summary statistics.

  • Source study: Dwivedi et al. (2019) reports results from a 43-item survey of 340 respondents measuring 15 constructs.

  • Method:

    • Generate synthetic scale scores with the same means and standard deviations as the published data.
    • Apply the reported correlation matrix to reproduce the relationships among variables.
    • Reconstruct plausible individual survey items using reported reliability (Cronbach’s α\alpha).

  • Validation: The synthetic dataset closely reproduces the published descriptive statistics, correlations, and scale reliabilities.

  • Result: Structural equation modelling of the synthetic data produces relationships consistent with those reported in the original paper.

  • Takeaway: LikertMakeR can approximate realistic datasets from published results, enabling replication exercises, teaching examples, and methodological exploration when raw data are unavailable.

Dwivedi et al. (2019) data replication

Background

Researchers often publish summary statistics - means, standard deviations, correlations, and reliability estimates - without releasing the original data.

In teaching, simulation studies, and methodological work, it can be useful to reconstruct plausible datasets from those published summaries.

Prerequisites

This tutorial assumes a basic familiarity with:

  • Likert-scale survey data
  • correlation matrices
  • structural equation modelling

Required R packages:

The LikertMakeR package provides tools for doing exactly this: generating synthetic Likert-scale datasets that reproduce the statistical properties of reported data.

In this tutorial we demonstrate the process using a published study by Dwivedi et al. (2019). Using only the statistics reported in their paper, we reconstruct a synthetic dataset and show that it reproduces the key statistical relationships reported in the original analysis.

Dwivedi et al. (2019) sought to understand social media platforms from a “brand” perspective through examining the effect of consumers’ emotional attachment on social media consumer-based brand equity (CBBE) using an online survey of 340 Australian social media consumers, as shown in Figure 1.

Figure 1: Hypothesised relationships among variables. (Dwivedi et al. (2019))

Overview of the Reconstruction Process

Reconstructing a dataset from published statistics involves several steps:

  1. Extract summary statistics from the published paper
  2. Reconstruct scale-level distributions (means and standard deviations)
  3. Reconstruct the correlation structure between scales
  4. Generate synthetic Likert-scale item responses consistent with those statistics
  5. Verify that the reconstructed dataset reproduces the reported results

We now walk through each step.

Verifying the Reconstruction

Reconstructing data from summary statistics inevitably involves approximation.

To assess whether the reconstruction is credible, we compare several properties of the synthetic dataset with those reported in the original paper:

  • scale means and standard deviations
  • inter-scale correlations
  • reliability estimates
  • structural equation model results

If these statistics closely match the published values, the synthetic data can be considered a reasonable reconstruction of the original dataset.

Step 1: Extract summary statistics

To recreate a plausible dataset, we need to match the first two statistical moments of each scale:

  • the mean (average level of responses), and
  • the standard deviation (spread of responses).

If synthetic data reproduce these two properties, the reconstructed scales behave similarly to the original ones.

Fifteen constructs were measured with Likert scales constructed from 43 items (questions). Summary statistics from the study are shown in Table 1.

construct name n_items mean sd alpha
Affection (AFF) AFF 2 2.51 0.93 0.87
Connection (CON) CON 2 2.79 0.98 0.87
Passion (PAS) PAS 3 2.62 0.89 0.84
Clarity of Positioning (COP) COP 3 3.52 0.70 0.80
Brand trust (TRT) TRT 4 3.30 0.74 0.88
Consumer satisfaction (SAT) SAT 3 3.87 0.64 0.86
Awareness/associations (AWR) AWR 5 4.10 0.62 0.87
Quality (QUL) QUL 4 3.69 0.66 0.84
Loyalty (LOY) LOY 2 3.56 0.84 0.79
Perceived differentiation (DIF) DIF 2 3.80 0.78 0.85
Self-brand congruence (FIT) FIT 2 3.02 0.83 0.82
Extraversion (EXT) EXT 2 3.09 0.91 0.79
Brand attitude (ATT) ATT 2 3.50 0.81 0.88
Relationship proneness (REL) REL 3 2.90 0.83 0.79
Involvement (INV) INV 4 3.32 0.92 0.79
Table 1: Summary statistics and variable properties. (Dwivedi et al. (2019))

Step 2: Reconstruct scale distributions

With those summary statistics, we can synthesise a dataframe with the same first and second moments as the original, using the LikertMakeR::lfast() function.

Show the code
library(LikertMakeR)

sample_size <- 340
lower <- 1
upper <- 5

set.seed(42)

base_data <- data.frame(
  AFF = lfast(sample_size, mean = 2.51, sd = 0.93, lower, upper, items = 2),
  CON = lfast(sample_size, mean = 2.79, sd = 0.98, lower, upper, items = 2),
  PAS = lfast(sample_size, mean = 2.62, sd = 0.89, lower, upper, items = 3),
  COP = lfast(sample_size, mean = 3.52, sd = 0.70, lower, upper, items = 3),
  TRT = lfast(sample_size, mean = 3.30, sd = 0.74, lower, upper, items = 4),
  SAT = lfast(sample_size, mean = 3.87, sd = 0.64, lower, upper, items = 3),
  AWR = lfast(sample_size, mean = 4.10, sd = 0.62, lower, upper, items = 5),
  QUL = lfast(sample_size, mean = 3.69, sd = 0.66, lower, upper, items = 4),
  LOY = lfast(sample_size, mean = 3.56, sd = 0.84, lower, upper, items = 2),
  DIF = lfast(sample_size, mean = 3.80, sd = 0.78, lower, upper, items = 2),
  FIT = lfast(sample_size, mean = 3.02, sd = 0.83, lower, upper, items = 2),
  EXT = lfast(sample_size, mean = 3.09, sd = 0.91, lower, upper, items = 2),
  ATT = lfast(sample_size, mean = 3.50, sd = 0.81, lower, upper, items = 2),
  REL = lfast(sample_size, mean = 2.90, sd = 0.83, lower, upper, items = 3),
  INV = lfast(sample_size, mean = 3.32, sd = 0.92, lower, upper, items = 4)
)

With this base dataframe, we check the summary statistics and find in Table 2 that the means and standard deviations are the same as in the reported data, to two decimal places.

mean sd
AFF 2.510 0.928
CON 2.790 0.979
PAS 2.621 0.892
COP 3.520 0.700
TRT 3.300 0.738
SAT 3.869 0.640
AWR 4.099 0.620
QUL 3.690 0.661
LOY 3.560 0.841
DIF 3.800 0.781
FIT 3.021 0.830
EXT 3.090 0.910
ATT 3.500 0.812
REL 2.900 0.832
INV 3.321 0.921
Table 2: Synthetic data summary statistics

Step 3: Correlations to match published data

These existing synthetic data have random low correlations. We need to correlate the variables according to the reported data. Dwivedi et al. (2019) reported the correlation presented in Table 3. This is our target correlation matrix.

AFF CON PAS COP TRT SAT AWR QUL LOY DIF FIT EXT ATT REL INV
AFF 1.00 0.79 0.82 0.47 0.50 0.44 0.35 0.44 0.54 0.46 0.58 0.33 0.66 0.62 0.51
CON 0.79 1.00 0.77 0.43 0.49 0.42 0.36 0.41 0.56 0.41 0.61 0.32 0.65 0.61 0.49
PAS 0.82 0.77 1.00 0.46 0.51 0.46 0.40 0.48 0.55 0.45 0.60 0.36 0.73 0.62 0.48
COP 0.47 0.43 0.46 1.00 0.57 0.54 0.39 0.51 0.50 0.37 0.41 0.24 0.49 0.38 0.31
TRT 0.50 0.49 0.51 0.57 1.00 0.61 0.31 0.61 0.55 0.31 0.52 0.23 0.58 0.43 0.33
SAT 0.44 0.42 0.46 0.54 0.61 1.00 0.53 0.67 0.63 0.40 0.47 0.23 0.63 0.43 0.41
AWR 0.35 0.36 0.40 0.39 0.31 0.53 1.00 0.55 0.49 0.47 0.30 0.29 0.46 0.41 0.48
QUL 0.44 0.41 0.48 0.51 0.61 0.67 0.55 1.00 0.49 0.43 0.43 0.27 0.55 0.44 0.34
LOY 0.54 0.56 0.55 0.50 0.55 0.63 0.49 0.49 1.00 0.47 0.55 0.28 0.63 0.53 0.52
DIF 0.46 0.41 0.45 0.37 0.31 0.40 0.47 0.43 0.47 1.00 0.33 0.36 0.46 0.42 0.42
FIT 0.58 0.61 0.60 0.41 0.52 0.47 0.30 0.43 0.55 0.33 1.00 0.27 0.60 0.51 0.42
EXT 0.33 0.32 0.36 0.24 0.23 0.23 0.29 0.27 0.28 0.36 0.27 1.00 0.39 0.47 0.36
ATT 0.66 0.65 0.73 0.49 0.58 0.63 0.46 0.55 0.63 0.46 0.60 0.39 1.00 0.66 0.55
REL 0.62 0.61 0.62 0.38 0.43 0.43 0.41 0.44 0.53 0.42 0.51 0.47 0.66 1.00 0.71
INV 0.51 0.49 0.48 0.31 0.33 0.41 0.48 0.34 0.52 0.42 0.42 0.36 0.55 0.71 1.00
Table 3: Target correlation matrix. Dwivedi et al. (2019)

We apply this correlation matrix to our existing data so that the variables are correlated according to our target correlations, using the LikertMakeR::lcor() function.

Show the code
correlated_data <- LikertMakeR::lcor(
  data = base_data,
  target = dwivedi_correlations
)
names(correlated_data) <- varnames

And the revised dataframe should be correlated in the same way as the published data. Results are shown in Table 4.

Show the code
synth_correlations <- cor(correlated_data)
dimnames(synth_correlations) <- list(Rows = varnames, Cols = varnames)
AFF CON PAS COP TRT SAT AWR QUL LOY DIF FIT EXT ATT REL INV
AFF 1.00 0.79 0.82 0.47 0.50 0.44 0.35 0.44 0.54 0.46 0.58 0.33 0.66 0.62 0.51
CON 0.79 1.00 0.77 0.43 0.49 0.42 0.36 0.41 0.56 0.41 0.61 0.32 0.65 0.61 0.49
PAS 0.82 0.77 1.00 0.46 0.51 0.46 0.40 0.48 0.55 0.45 0.60 0.36 0.73 0.62 0.48
COP 0.47 0.43 0.46 1.00 0.57 0.54 0.39 0.51 0.50 0.37 0.41 0.24 0.49 0.38 0.31
TRT 0.50 0.49 0.51 0.57 1.00 0.61 0.31 0.61 0.55 0.31 0.52 0.23 0.58 0.43 0.33
SAT 0.44 0.42 0.46 0.54 0.61 1.00 0.53 0.67 0.63 0.40 0.47 0.23 0.63 0.43 0.41
AWR 0.35 0.36 0.40 0.39 0.31 0.53 1.00 0.55 0.49 0.47 0.30 0.29 0.46 0.41 0.48
QUL 0.44 0.41 0.48 0.51 0.61 0.67 0.55 1.00 0.49 0.43 0.43 0.27 0.55 0.44 0.34
LOY 0.54 0.56 0.55 0.50 0.55 0.63 0.49 0.49 1.00 0.47 0.55 0.28 0.63 0.53 0.52
DIF 0.46 0.41 0.45 0.37 0.31 0.40 0.47 0.43 0.47 1.00 0.33 0.36 0.46 0.42 0.42
FIT 0.58 0.61 0.60 0.41 0.52 0.47 0.30 0.43 0.55 0.33 1.00 0.27 0.60 0.51 0.42
EXT 0.33 0.32 0.36 0.24 0.23 0.23 0.29 0.27 0.28 0.36 0.27 1.00 0.39 0.47 0.36
ATT 0.66 0.65 0.73 0.49 0.58 0.63 0.46 0.55 0.63 0.46 0.60 0.39 1.00 0.66 0.55
REL 0.62 0.61 0.62 0.38 0.43 0.43 0.41 0.44 0.53 0.42 0.51 0.47 0.66 1.00 0.71
INV 0.51 0.49 0.48 0.31 0.33 0.41 0.48 0.34 0.52 0.42 0.42 0.36 0.55 0.71 1.00
Table 4: Synthetic data correlation matrix.

The synthetic data correlations appear to be the same, rounded to two decimal places, as the target, published correlations.

To quantify how closely the reconstructed correlations match the published ones, we compute the Frobenius norm of the difference between the two matrices.

In simple terms, this measures the overall discrepancy between the two correlation tables. A value close to zero indicates that the reconstructed correlations closely match those reported in the paper.

The Frobenius Norm is defined as the square root of the sum of the squares of all the matrix entries Equation 1.

F=(∑i=1m∑j=1naij2)1/2(1) F = \left( \sum_{i=1}^m \sum_{j=1}^n a_{ij}^2 \right)^{1/2} \qquad(1)

Frobenius norm of a matrix.

Show the code
mat_diff <- dwivedi_correlations - synth_correlations
frob_diff <- matrixcalc::frobenius.norm(mat_diff)

Calculated Frobenius Norm here is 0.0124366, which is very low for a matrix of this size.

Step 4: Generate synthetic Likert responses

So far we have reconstructed the statistical properties of the scales. The next step is to generate individual Likert item responses consistent with those statistics.

Given the correlated variables, we also have information on the nature of the items that make each variable presented in Table 1. We can use those reliability coefficients to generate plausible individual item responses.

We use the LikertMakeR::makeItemsScale() function.

Show the code
set.seed(42)

aff_items <- makeItemsScale(
  scale = correlated_data$AFF,
  lowerbound = 1, upperbound = 5,
  items = 2,
  alpha = 0.87,
  summated = FALSE,
  progress = FALSE
)
names(aff_items) <- c("aff1", "aff2")

Generating the other 14 sets of scale items works the same way, using alpha values and number of item from Table 1.

Show the code
con_items <- makeItemsScale(correlated_data$CON, 1, 5, 2, 0.87, FALSE, FALSE)
names(con_items) <- c("con1", "con2")

pas_items <- makeItemsScale(correlated_data$PAS, 1, 5, 3, 0.84, FALSE, FALSE)
names(pas_items) <- c("pas1", "pas2", "pas3")

cop_items <- makeItemsScale(correlated_data$COP, 1, 5, 3, 0.80, FALSE, FALSE)
names(cop_items) <- c("cop1", "cop2", "cop3")

trt_items <- makeItemsScale(correlated_data$TRT, 1, 5, 4, 0.88, FALSE, FALSE)
names(trt_items) <- c("trt1", "trt2", "trt3", "trt4")

sat_items <- makeItemsScale(correlated_data$SAT, 1, 5, 3, 0.86, FALSE, FALSE)
names(sat_items) <- c("sat1", "sat2", "sat3")

awr_items <- makeItemsScale(correlated_data$AWR, 1, 5, 5, 0.87, FALSE, FALSE)
names(awr_items) <- c("awr1", "awr2", "awr3", "awr4", "awr5")

qul_items <- makeItemsScale(correlated_data$QUL, 1, 5, 4, 0.84, FALSE, FALSE)
names(qul_items) <- c("qul1", "qul2", "qul3", "qul4")

loy_items <- makeItemsScale(correlated_data$LOY, 1, 5, 2, 0.79, FALSE, FALSE)
names(loy_items) <- c("loy1", "loy2")

dif_items <- makeItemsScale(correlated_data$DIF, 1, 5, 2, 0.85, FALSE, FALSE)
names(dif_items) <- c("dif1", "dif2")

fit_items <- makeItemsScale(correlated_data$FIT, 1, 5, 2, 0.82, FALSE, FALSE)
names(fit_items) <- c("fit1", "fit2")

ext_items <- makeItemsScale(correlated_data$EXT, 1, 5, 2, 0.79, FALSE, FALSE)
names(ext_items) <- c("ext1", "ext2")

att_items <- makeItemsScale(correlated_data$ATT, 1, 5, 2, 0.88, FALSE, FALSE)
names(att_items) <- c("att1", "att2")

rel_items <- makeItemsScale(correlated_data$REL, 1, 5, 3, 0.79, FALSE, FALSE)
names(rel_items) <- c("rel1", "rel2", "rel3")

inv_items <- makeItemsScale(correlated_data$INV, 1, 5, 4, 0.79, FALSE, FALSE)
names(inv_items) <- c("inv1", "inv2", "inv3", "inv4")

The first ten observations are presented in Table 5. They are all integer responses to standard 1-5 Likert-scale-item questions.

   aff1 aff2 con1 con2 pas1 pas2 pas3 cop1 cop2 cop3 trt1 trt2 trt3 trt4 sat1
1     2    2    2    2    3    3    2    3    3    3    4    3    3    3    4
2     2    3    2    3    3    3    3    4    5    4    3    3    4    3    5
3     3    4    3    3    4    3    3    3    3    3    4    4    4    4    3
4     3    3    4    3    3    3    3    4    4    5    5    4    5    5    4
5     3    3    4    4    2    3    2    4    3    3    3    3    2    3    3
6     2    1    2    1    1    2    3    4    3    3    3    2    3    2    3
7     1    2    2    2    2    2    1    5    5    3    2    3    2    3    3
8     2    1    3    3    2    2    1    2    3    3    4    3    3    3    4
9     3    3    3    3    3    3    2    4    3    4    3    3    4    3    4
10    2    3    3    3    2    2    2    4    4    4    3    2    3    3    4
   sat2 sat3 awr1 awr2 awr3 awr4 awr5 qul1 qul2 qul3 qul4 loy1 loy2 dif1 dif2
1     3    3    4    4    5    4    4    3    4    4    3    4    2    3    3
2     5    4    4    4    5    5    5    5    4    4    4    5    4    4    5
3     4    3    5    4    4    5    4    4    3    5    4    4    4    4    5
4     5    4    4    4    5    4    4    4    4    4    4    5    5    4    5
5     3    2    4    4    5    4    4    4    3    4    3    3    3    3    4
6     4    4    4    4    4    4    4    3    4    3    4    4    2    3    3
7     3    2    3    3    4    3    3    4    4    3    4    3    3    4    4
8     4    3    4    4    4    4    4    4    3    4    4    4    2    3    4
9     4    3    5    5    5    5    4    4    3    3    4    4    3    3    3
10    4    4    4    5    4    4    5    4    3    3    3    5    3    4    5
   fit1 fit2 ext1 ext2 att1 att2 rel1 rel2 rel3 inv1 inv2 inv3 inv4
1     2    3    3    5    4    3    4    3    4    4    3    4    4
2     3    3    3    3    4    4    3    4    4    5    4    5    5
3     4    5    4    4    4    3    2    3    3    4    3    4    3
4     3    4    4    4    4    4    4    5    5    5    3    5    3
5     2    3    4    2    3    3    3    3    2    4    3    4    4
6     2    1    2    2    2    2    2    1    1    3    1    3    2
7     2    2    2    1    2    2    2    2    2    1    1    3    1
8     3    4    3    4    3    3    3    3    2    4    2    5    3
9     2    3    3    3    4    3    4    3    2    5    5    5    5
10    2    4    5    4    4    3    2    3    4    5    4    5    4
Table 5: First ten rows of our synthetic data - all 43 items

The resulting dataset contains simulated item responses that follow the same statistical structure as the original study. Most constructs achieved Cronbach’s alpha values close to those specified, as we can see in Table 6.

construct original_alphas synthetic_alphas
AFF 0.87 0.871
CON 0.87 0.870
PAS 0.84 0.840
COP 0.80 0.799
TRT 0.88 0.879
SAT 0.86 0.859
AWR 0.87 0.869
QUL 0.84 0.839
LOY 0.79 0.790
DIF 0.85 0.851
FIT 0.82 0.819
EXT 0.79 0.789
ATT 0.88 0.881
REL 0.79 0.791
INV 0.79 0.788
Table 6: Comparison between Cronbach’s alphas for original published constructs and synthetic constructs.

Step 5: Reproduce relationship structure analysis

Structural Equation Model

Now that we have a complete synthetic data set, we can analyse the data to see if we can find the same relationships found by Dwivedi et al.

original results

The original published results from Dwivedi et al. are reproduced in Table 7.

Estimated path β 95% CI Hypothesis support
Panel A: Hypothesised model results
Emotional brand attachment → Brand credibility 0.72 0.60-0.82 Supported
Emotional brand attachment → Consumer satisfaction 0.54 0.42-0.64 Supported
Emotional brand attachment → Consumer-based brand equity (CBBE) 0.21 0.03-0.40 Not supported
Brand credibility → CBBE 0.31 0.11-0.54 Supported
Consumer satisfaction → CBBE 0.63 0.50-0.76 Supported
Total indirect effect
Emotional brand attachment (EBA) → CBBE 0.57 0.42-0.77
Specific indirect effects
EBA → CBBE via Brand credibility 0.09 0.03-0.18 Supported
EBA → CBBE via Consumer satisfaction 0.14 0.09-0.21 Supported
Panel B: Alternative model results
Emotional brand attachment → Brand credibility 0.74 0.61-0.84 Supported
Emotional brand attachment → Consumer satisfaction 0.56 0.44-0.68 Supported
Emotional brand attachment → Brand loyalty 0.40 0.17-0.61 Supported
Brand credibility → Brand loyalty 0.14 0.07-0.45 Not supported
Consumer satisfaction → Brand loyalty 0.35 0.19-0.53 Supported
Specific indirect effects
EBA → LOY via Consumer satisfaction 0.12 0.06-0.20 Supported
Table 7: Parameter estimates of the hypothesised and alternative models - Table IV from Dwivedi et al. (2019).

Results from synthetic data

The authors of the original paper used the AMOS program, which is a part of SPSS. I haven’t used SPSS/AMOS for a very long time, so I’ll run my analysis using the lavaan package for R. By default, AMOS uses Maximum Likelihood (ML) estimation, while lavaan uses Robust Maximum Likelihood (MLR). With ML specified and bootstrapped confidence interval estimates, we should get comparable output.

Show the code
# model definition
model <- "
# first-order constructs
    AFF =~ aff1 + aff2
    CON =~ con1 + con2
    PAS =~ pas1 + pas2 + pas3
    AWR =~ awr1 + awr2 + awr3 + awr4 + awr5
    QUL =~ qul1 + qul2 + qul3 + qul4
    LOY =~ loy1 + loy2
    COP =~ cop1 + cop2 + cop3
    TRT =~ trt1 + trt2 + trt3 + trt4
    SAT =~ sat1 + sat2 + sat3
# second-order constructs
    Attachment =~  AFF + CON + PAS
    Credibility =~ TRT + COP
    BrandEquity =~ AWR + QUL + LOY
# structural model
    Credibility ~ H1*Attachment
    SAT ~ H2*Attachment
    BrandEquity ~ H3*Attachment + H4*Credibility + H5*SAT
# allow correlation between mediators
   Credibility ~~ SAT
# Specific indirect effects
  ind_Attachment_via_Credibility := H1 * H4
  ind_Attachment_via_SAT := H2 * H5
# Total indirect effect
  total_ind_Attachment := (H1 * H4) + (H2 * H5)
# Total effect
  total_Attachment := H3 + total_ind_Attachment
"

set.seed(42)

# run the model
sem_result <- lavaan::sem(
  model,
  data = all_items,
  estimator = "ML",
  se = "bootstrap",
  bootstrap = 2000
)

# extract standardised results
std_results <- lavaan::standardizedSolution(sem_result)
# Keep only structural and indirect effects
std_structural <- std_results[
  std_results$op %in% c("~", ":="),
]
# Keep only regressions (~) and defined parameters (:=)
std_results <- std_results[std_results$op %in% c("~", ":="), ]
# Get bootstrap CI from parameterEstimates
boot_pe <- lavaan::parameterEstimates(
  sem_result,
  standardized = TRUE,
  ci = TRUE,
  boot.ci.type = "bca.simple"
)
boot_pe <- boot_pe[boot_pe$op == ":=", ]

In this paper, we confine ourselves to the main theoretical model shown in Panel A. Parameter estimates of the hypothesised model are summarised in Table 8. We can see that not all the figures are exactly the same as in the original, but they are very close and the conclusions are the same.

Estimated path β 95% CI
Direct effects
Attachment → Credibility 0.71 [0.63, 0.79]
Attachment → SAT 0.53 [0.45, 0.61]
Attachment → BrandEquity 0.22 [0.08, 0.37]
Credibility → BrandEquity 0.26 [-0.06, 0.59]
SAT → BrandEquity 0.60 [0.35, 0.86]
Indirect and total effects
ind_Attachment_via_Credibility 0.19 [-0.05, 0.42]
ind_Attachment_via_SAT 0.32 [0.18, 0.46]
total_ind_Attachment 0.51 [0.37, 0.64]
total_Attachment 0.73 [0.66, 0.8]
Table 8: Parameter estimates of the hypothesised model - From synthetic data.

A graphic representation of the structural component of the SEM model created by the synthetic data is presented in Figure 2.

Figure 2: Structural model of Synthetic data - reproducing original data.

Conclusion

This exercise set out to see if we can use functions in the LikertMakeR package to “reproduce” raw data based on summary statistics presented in a published report. We find that results are surprisingly strong - producing data that give similar output to those in the original report, and giving the same conclusions. Of course we cannot expect the data to give exactly the same results - after all, they are artificial and subject to rounding of the reported output.

What we have learned

This reconstruction demonstrates that synthetic Likert-scale data can be generated from published summary statistics with high fidelity.

Even though the original raw data are unavailable, the reconstructed dataset reproduces:

  • the reported scale distributions
  • the correlation structure between constructs
  • the reliability estimates
  • the results of the structural equation model

This makes synthetic reconstruction useful for:

  • teaching statistical methods
  • replicating published analyses
  • conducting methodological simulations when raw data are unavailable.

References

Dwivedi, Abhishek, Lester W. Johnson, Dean Charles Wilkie, and Luciana De Araujo-Gil. 2019. ‘Consumer Emotional Brand Attachment with Social Media Brands and Social Media Brand Equity’. European Journal of Marketing 53 (6): 1176–204. https://doi.org/10.1108/EJM-09-2016-0511.
Winzar, Hume. 2026. LikertMakeR: Synthesise and Correlate Likert-Scale and Related Rating-Scale Data Based on Summary Statistics. Version 2.0.0. The Comprehensive R Archive Network (CRAN). https://CRAN.R-project.org/package=LikertMakeR.