PTSD and Distress Among the Ukrainian Population During the War: Does Media Traumatization Play a Mediating Role?
Analytical Code Report — Working Version
Author
Andrii Bova
Published
February 23, 2026
Abstract
Background. Media consumption during armed conflict may amplify the psychological burden imposed by trauma exposure.
Objective. This study tests whether media traumatization mediates the relationship between PTSD symptomatology and psychological distress in a nationally representative Ukrainian wartime sample.
Methods. Data were drawn from a CATI/CAVI survey conducted in October 2023 (N = 2,767). Psychological distress was assessed via a 28-item adaptation of the SCL-90-R (SCL-28), modeled as a hierarchical second-order latent factor. PTSD and media traumatization were each operationalized as single-factor latent constructs. A full-information SEM was estimated using the WLSMV estimator with sampling weights.
Results. The hierarchical SCL-28 structure demonstrated acceptable fit. The indirect effect of PTSD on distress via media traumatization was statistically significant, indicating partial mediation.
Conclusions. War-related media exposure constitutes a secondary pathway through which PTSD translates into broader psychological distress.
# ── Core data wrangling ───────────────────────────────────────────────────────library(here)library(haven)library(dplyr)library(purrr)library(stringr)library(tibble)# ── SEM / CFA ─────────────────────────────────────────────────────────────────library(lavaan)library(semTools)library(lavaanExtra)library(lavaanPlot)library(BifactorIndicesCalculator)library(semPlot)# ── Psychometrics ─────────────────────────────────────────────────────────────library(psych)library(sjmisc)# ── Visualization ─────────────────────────────────────────────────────────────library(ggcorrplot)library(patchwork)library(DiagrammeR)# ── Reporting tables ──────────────────────────────────────────────────────────library(knitr)library(kableExtra)library(gridExtra)
Introduction
The empirical basis of this study is data from a sociological survey conducted within the framework of the Institute of Sociology of the National Academy of Sciences of Ukraine project, “Stress States of the Ukrainian Population in the Context of War: Prevalence, Risk Groups, and Compensation Pathways” (Project PI: Dr. Serhii Dembitskyi). The survey was carried out by the “Rating” group across Ukraine (excluding occupied territories of Crimea and Donbas) from October 6–10, 2023, using the CATI-CAVI method. The sample includes 2,767 respondents (Dembitskyi, 2023), with an overview of the primary survey results presented in Dembitskyi et al. (2025).
The present study represents an independent analytical contribution to this research program. While drawing on the same empirical dataset, the current analysis pursues a distinct research objective: testing a theoretically grounded mediation model using confirmatory factor analysis (CFA) and structural equation modeling (SEM), with the inclusion of additional measurement scales not analyzed in the original publication. Specifically, this study introduces a hierarchical second-order CFA structure for the SCL-28 distress instrument, a bifactor model comparison, and a formal test of media traumatization as a mediator between PTSD and psychological distress.
1 Research Objective and Hypotheses
RQ: Does media traumatization mediate the relationship between PTSD and psychological distress?
H1: PTSD positively predicts psychological distress. H2: PTSD positively predicts media traumatization. H3: The effect of PTSD on distress is partially mediated by media traumatization.
2 Measures
2.1 Psychological Distress (SCL-28)
Psychological distress was assessed using a 28-item adapted version of the Symptom Checklist-90-Revised (SCL-90-R; Derogatis, 1994), covering seven subscales: Depression (DPR), Anxiety (TRE), Somatization (SOM), General Exhaustion (IST), Interpersonal Sensitivity (SNS), Paranoid Ideation (PAR), and Hostility (VRG), with four items per subscale. Respondents rated the frequency of each symptom on a 4-point ordinal scale ranging from 1 (Never) to 4 (Almost constantly).
2.2 PTSD Symptoms
PTSD symptomatology was measured using five items reflecting core re-experiencing symptoms consistent with DSM-5 criteria (American Psychiatric Association, 2013), including recurrent distressing memories, nightmares, flashbacks, psychological distress at trauma-related cues, and physiological reactivity to trauma reminders. Each item was rated on a 5-point ordinal scale from 1 (Never) to 5 (Very often).
2.3 Media Traumatization
Media traumatization was operationalized as the degree of psychological impact experienced while consuming war-related media content. Five items captured both intrusive reactions (e.g., involuntary recall of distressing media images) and avoidance-related responses (e.g., deliberate efforts to suppress media-related thoughts), rated on a 5-point ordinal scale from 1 (Not at all) to 5 (Extremely). This operationalization parallels the intrusion and avoidance symptom clusters associated with trauma exposure (Holman et al., 2014).
3 Data Preparation
This section describes the data source, sampling procedure, variable operationalization, and recoding decisions made prior to analysis.
3.1 Data Loading and Initial Processing
Show / Hide Code
# Load data from SPSS filedata_raw <-read_sav(here("stressFin.sav"))# Respondents residing in Ukraine have NA on sd11 (country of residence)data_raw <- data_raw %>%filter(is.na(sd11))cat("Total observations after filtering:", nrow(data_raw), "\n")
Total observations after filtering: 2712
3.2 Helper Functions
Show / Hide Code
# Extract SPSS variable label from attributeget_label <-function(x) { label <-attr(x, "label")if (is.null(label)) return(NA_character_) label}# Build a code → label reference table for a set of variablescreate_variable_labels_table <-function(data, variables) { labels <-sapply(data[variables], get_label)tibble(Variable_Code =names(labels),Question_Text =unname(labels) )}
What this section tests.
Because all SCL-28 items are measured on a 4-point ordinal scale, Pearson correlations underestimate true associations due to categorization artifacts. Polychoric correlations recover the underlying continuous latent-variable correlations and provide the appropriate input for WLSMV-based CFA. The full 28 × 28 polychoric matrix is visualized as a heatmap (no clustering imposed) to allow inspection of inter-item relationships across subscales.
Prerequisite for CFA: Polychoric correlations must be computable (no zero-variance items, no perfect collinearity) and the resulting matrix must be positive definite.
Table 2: Polychoric Correlation Matrix of SCL-28 Items
dpr1
dpr2
dpr3
dpr4
tre1
tre2
tre3
tre4
som1
som2
som3
som4
ist1
ist2
ist3
ist4
sns1
sns2
sns3
sns4
par1
par2
par3
par4
vrg1
vrg2
vrg3
vrg4
dpr1
1.00
0.47
0.42
0.53
0.52
0.50
0.47
0.50
0.44
0.39
0.37
0.46
0.57
0.39
0.46
0.49
0.42
0.34
0.29
0.45
0.29
0.25
0.32
0.37
0.28
0.34
0.35
0.41
dpr2
0.47
1.00
0.45
0.64
0.38
0.41
0.48
0.36
0.33
0.27
0.30
0.28
0.40
0.33
0.34
0.47
0.44
0.30
0.27
0.33
0.33
0.23
0.30
0.38
0.24
0.26
0.35
0.34
dpr3
0.42
0.45
1.00
0.56
0.40
0.39
0.47
0.40
0.35
0.34
0.33
0.33
0.39
0.27
0.29
0.34
0.39
0.27
0.22
0.36
0.28
0.29
0.26
0.38
0.24
0.28
0.31
0.31
dpr4
0.53
0.64
0.56
1.00
0.45
0.42
0.52
0.39
0.36
0.32
0.33
0.33
0.48
0.35
0.42
0.53
0.50
0.36
0.31
0.38
0.32
0.30
0.32
0.44
0.29
0.32
0.46
0.38
tre1
0.52
0.38
0.40
0.45
1.00
0.53
0.49
0.55
0.43
0.48
0.46
0.50
0.50
0.40
0.46
0.39
0.43
0.35
0.29
0.43
0.31
0.28
0.36
0.40
0.28
0.35
0.36
0.44
tre2
0.50
0.41
0.39
0.42
0.53
1.00
0.52
0.48
0.39
0.41
0.34
0.40
0.43
0.36
0.41
0.36
0.39
0.37
0.33
0.46
0.21
0.22
0.28
0.32
0.23
0.24
0.29
0.34
tre3
0.47
0.48
0.47
0.52
0.49
0.52
1.00
0.42
0.39
0.40
0.36
0.38
0.42
0.39
0.42
0.37
0.47
0.36
0.33
0.40
0.30
0.27
0.33
0.36
0.25
0.29
0.35
0.35
tre4
0.50
0.36
0.40
0.39
0.55
0.48
0.42
1.00
0.42
0.43
0.39
0.40
0.50
0.39
0.45
0.37
0.37
0.36
0.27
0.49
0.31
0.34
0.37
0.35
0.33
0.35
0.31
0.43
som1
0.44
0.33
0.35
0.36
0.43
0.39
0.39
0.42
1.00
0.44
0.50
0.48
0.53
0.38
0.38
0.31
0.30
0.32
0.22
0.39
0.34
0.23
0.30
0.36
0.27
0.29
0.26
0.36
som2
0.39
0.27
0.34
0.32
0.48
0.41
0.40
0.43
0.44
1.00
0.54
0.48
0.40
0.31
0.36
0.25
0.29
0.33
0.24
0.37
0.28
0.25
0.29
0.28
0.23
0.24
0.24
0.31
som3
0.37
0.30
0.33
0.33
0.46
0.34
0.36
0.39
0.50
0.54
1.00
0.52
0.44
0.32
0.37
0.25
0.30
0.27
0.24
0.32
0.28
0.24
0.30
0.33
0.22
0.28
0.34
0.31
som4
0.46
0.28
0.33
0.33
0.50
0.40
0.38
0.40
0.48
0.48
0.52
1.00
0.43
0.30
0.33
0.25
0.29
0.31
0.22
0.34
0.22
0.23
0.30
0.35
0.18
0.27
0.27
0.36
ist1
0.57
0.40
0.39
0.48
0.50
0.43
0.42
0.50
0.53
0.40
0.44
0.43
1.00
0.48
0.48
0.45
0.41
0.36
0.28
0.42
0.27
0.27
0.31
0.35
0.28
0.34
0.36
0.39
ist2
0.39
0.33
0.27
0.35
0.40
0.36
0.39
0.39
0.38
0.31
0.32
0.30
0.48
1.00
0.55
0.38
0.39
0.34
0.34
0.42
0.25
0.26
0.29
0.31
0.32
0.34
0.34
0.36
ist3
0.46
0.34
0.29
0.42
0.46
0.41
0.42
0.45
0.38
0.36
0.37
0.33
0.48
0.55
1.00
0.45
0.42
0.37
0.35
0.41
0.23
0.28
0.30
0.32
0.26
0.33
0.33
0.38
ist4
0.49
0.47
0.34
0.53
0.39
0.36
0.37
0.37
0.31
0.25
0.25
0.25
0.45
0.38
0.45
1.00
0.45
0.32
0.32
0.35
0.30
0.26
0.33
0.34
0.25
0.27
0.33
0.32
sns1
0.42
0.44
0.39
0.50
0.43
0.39
0.47
0.37
0.30
0.29
0.30
0.29
0.41
0.39
0.42
0.45
1.00
0.47
0.47
0.39
0.29
0.29
0.38
0.38
0.22
0.31
0.34
0.34
sns2
0.34
0.30
0.27
0.36
0.35
0.37
0.36
0.36
0.32
0.33
0.27
0.31
0.36
0.34
0.37
0.32
0.47
1.00
0.50
0.44
0.28
0.29
0.35
0.35
0.25
0.27
0.29
0.34
sns3
0.29
0.27
0.22
0.31
0.29
0.33
0.33
0.27
0.22
0.24
0.24
0.22
0.28
0.34
0.35
0.32
0.47
0.50
1.00
0.38
0.23
0.26
0.28
0.27
0.13
0.21
0.27
0.26
sns4
0.45
0.33
0.36
0.38
0.43
0.46
0.40
0.49
0.39
0.37
0.32
0.34
0.42
0.42
0.41
0.35
0.39
0.44
0.38
1.00
0.29
0.28
0.36
0.38
0.35
0.36
0.28
0.40
par1
0.29
0.33
0.28
0.32
0.31
0.21
0.30
0.31
0.34
0.28
0.28
0.22
0.27
0.25
0.23
0.30
0.29
0.28
0.23
0.29
1.00
0.31
0.43
0.40
0.20
0.23
0.25
0.29
par2
0.25
0.23
0.29
0.30
0.28
0.22
0.27
0.34
0.23
0.25
0.24
0.23
0.27
0.26
0.28
0.26
0.29
0.29
0.26
0.28
0.31
1.00
0.31
0.31
0.30
0.24
0.27
0.27
par3
0.32
0.30
0.26
0.32
0.36
0.28
0.33
0.37
0.30
0.29
0.30
0.30
0.31
0.29
0.30
0.33
0.38
0.35
0.28
0.36
0.43
0.31
1.00
0.48
0.27
0.22
0.25
0.32
par4
0.37
0.38
0.38
0.44
0.40
0.32
0.36
0.35
0.36
0.28
0.33
0.35
0.35
0.31
0.32
0.34
0.38
0.35
0.27
0.38
0.40
0.31
0.48
1.00
0.26
0.28
0.37
0.38
vrg1
0.28
0.24
0.24
0.29
0.28
0.23
0.25
0.33
0.27
0.23
0.22
0.18
0.28
0.32
0.26
0.25
0.22
0.25
0.13
0.35
0.20
0.30
0.27
0.26
1.00
0.42
0.36
0.43
vrg2
0.34
0.26
0.28
0.32
0.35
0.24
0.29
0.35
0.29
0.24
0.28
0.27
0.34
0.34
0.33
0.27
0.31
0.27
0.21
0.36
0.23
0.24
0.22
0.28
0.42
1.00
0.52
0.54
vrg3
0.35
0.35
0.31
0.46
0.36
0.29
0.35
0.31
0.26
0.24
0.34
0.27
0.36
0.34
0.33
0.33
0.34
0.29
0.27
0.28
0.25
0.27
0.25
0.37
0.36
0.52
1.00
0.52
vrg4
0.41
0.34
0.31
0.38
0.44
0.34
0.35
0.43
0.36
0.31
0.31
0.36
0.39
0.36
0.38
0.32
0.34
0.34
0.26
0.40
0.29
0.27
0.32
0.38
0.43
0.54
0.52
1.00
5.2 Single Factor (M1 - CMB check)
What this section tests. All 28 SCL-28 items were collected via a single self-report questionnaire, raising the possibility of Common Method Bias (CMB). Harman’s single-factor CFA test evaluates whether a single unmeasured factor can account for the majority of variance across all indicators. Poor model fit (CFI < .95, RMSEA > .08) provides evidence against a dominant common method factor.
Limitation note. Harman’s test is a necessary but not sufficient test for CMB. It identifies the worst case but cannot rule out partial method effects. Results should be interpreted alongside substantive design considerations.
Table 3: Harman’s Single-Factor CFA — Fit Indices (Common Method Bias Test)
Fit Index
Value
Acceptable Threshold
χ² (Scaled)
3914.724
—
df
350.000
—
p-value
0.000
—
CFI (Robust)
0.813
≥ .95
TLI (Robust)
0.798
≥ .95
RMSEA (Robust)
0.080
≤ .06
RMSEA 90% CI Lower
0.078
—
RMSEA 90% CI Upper
0.082
—
SRMR
0.056
≤ .08
Note. Poor fit of the single-factor model (CFI < .95, RMSEA > .08) indicates that a dominant common method factor does not account for the majority of item covariance, providing evidence against pervasive CMB.
6 Measurement Model Validation
6.1 Single-Factor Models for Individual Subscales
What this section tests. Before evaluating the full hierarchical structure, we verify the unidimensionality of each subscale in isolation. A well-fitting single-factor model (CFI ≥ .95, RMSEA ≤ .06) confirms that the four items within each domain are locally unidimensional — a prerequisite for their use as first-order factors in the hierarchical model.
Note. All loadings estimated with robust WLSMV; sampling weights applied.
6.2 Correlated First-Order Factors Model (M1 — Baseline)
What this section tests. The correlated first-order factors model (M1) treats the seven SCL-28 subscales as distinct but inter-correlated latent constructs, without imposing any higher-order structure. This model serves two purposes: (1) it provides a baseline against which the more parsimonious hierarchical model (M2) can be formally compared; and (2) it allows direct inspection of inter-factor correlations and reliability/validity indices — all of which must be assessed before imposing second-order structure.
Discriminant validity criterion. If inter-factor correlations systematically exceed √AVE (Fornell–Larcker criterion) or HTMT ratios exceed .85, the constructs may not be sufficiently distinct to warrant separate measurement.
What this section reports. Reliability and validity are evaluated on the correlated first-order factors model (M1), which is the methodologically appropriate baseline. Indices calculated on a hierarchical model may be distorted by the additional constraints imposed at the second-order level (Hair et al., 2022; Kline, 2016).
Ordinal alpha (α_ord): appropriate α for ordinal data, based on polychoric correlations.
McDonald’s ω: model-based composite reliability; does not assume tau-equivalence. Threshold: ω ≥ .70 acceptable; ω ≥ .80 good.
DPR TRE SOM IST SNS PAR VRG
0.748 0.748 0.740 0.732 0.700 0.662 0.702
6.4 Convergent Validity: AVE — M1
What this section tests. AVE = Σλ² / (Σλ² + Σε²), where λ are standardized loadings and ε are item error variances. AVE ≥ .50 means the factor explains more variance in its indicators than error — the Fornell–Larcker convergent validity criterion. Assessed on M1 (correlated factors), before imposing second-order constraints.
Table 10: Convergent Validity: AVE and Composite Reliability — M1
Subscale
AVE
CR_omega
AVE ≥ .50
CR ≥ .70
DPR
0.519
0.748
✓
✓
TRE
0.499
0.748
✗
✓
SOM
0.492
0.740
✗
✓
IST
0.471
0.732
✗
✓
SNS
0.444
0.700
✗
✓
PAR
0.385
0.662
✗
✗
VRG
0.471
0.702
✗
✓
Note. AVE ≥ .50 indicates convergent validity (Fornell & Larcker, 1981). If AVE < .50 but CR > .60, convergent validity may still be adequate (Bagozzi & Yi, 1988). All indices estimated from M1 (correlated first-order factors).
6.5 Discriminant Validity: HTMT and Fornell–Larcker — M1
What this section tests.
(1) Fornell–Larcker criterion. AVE of each factor must exceed the squared inter-factor correlation (φ²) with every other factor.
(2) HTMT ratio (Henseler et al., 2015): HTMT < .85 (strict) or < .90 (liberal) indicates adequate discriminant validity. HTMT is considered a more sensitive criterion than Fornell–Larcker for ordinal SEM. Both tests use M1 estimates.
Table 12: Heterotrait–Monotrait Ratio (HTMT) for Discriminant Validity — M1
DPR
TRE
SOM
IST
SNS
PAR
VRG
DPR
1.000
0.879
0.678
0.841
0.737
0.724
0.660
TRE
0.879
1.000
0.823
0.866
0.791
0.711
0.688
SOM
0.678
0.823
1.000
0.720
0.621
0.642
0.578
IST
0.841
0.866
0.720
1.000
0.817
0.705
0.698
SNS
0.737
0.791
0.621
0.817
1.000
0.760
0.631
PAR
0.724
0.711
0.642
0.705
0.760
1.000
0.659
VRG
0.660
0.688
0.578
0.698
0.631
0.659
1.000
Note. HTMT < .85 indicates strict discriminant validity (Henseler et al., 2015). HTMT ≥ .90 suggests the two constructs may not be empirically distinct.
6.6 Hierarchical Second-Order CFA Model (M2 — Primary)
What this section tests. The hierarchical model (M2) posits that the covariance among the seven first-order subscales is fully explained by a single superordinate latent factor — General Psychological Distress (SCL28). M2 is more parsimonious than M1 (14 fewer parameters), trading some local fit for theoretical elegance.
CFA was conducted using the WLSMV estimator, appropriate for ordinal indicators with non-normal distributions (Flora & Curran, 2004). Sampling weights were incorporated via the sampling.weights argument in lavaan (Rosseel, 2012). Model fit was evaluated using robust fit indices: CFI ≥ .95, TLI ≥ .95, RMSEA ≤ .06, SRMR ≤ .08 (Hu & Bentler, 1999).
Figure 3: Path Diagram — M2: Hierarchical Second-Order CFA (Standardized)
Show / Hide Code
save_png(pl_m2, "cfa_path_diagram.png")
6.7 Bifactor Model and Indices (M3)
What this section tests. The bifactor model (M3) is an alternative parameterization in which each indicator loads simultaneously on (a) a general factor G capturing global distress and (b) one subscale-specific factor capturing residual domain variance orthogonal to G. Bifactor indices (Rodriguez et al., 2016) quantify the relative strength of G versus specific factors:
ECV — Explained Common Variance: proportion of common variance due to G; > .70 supports total score use
OmegaH — Hierarchical Omega: reliability of total score attributable to G alone; > .70 required
PUC — Proportion of Uncontaminated Correlations: > .70 means item correlations are mostly G-driven
H — Construct Replicability: factor stability across samples; > .70 desirable
Table 16: Factor-Level Dimensionality and Reliability Indices — Bifactor Model
ECV_SS
ECV_SG
ECV_GS
Omega
OmegaH
H
FD
DPR
0.268
0.043
0.732
0.774
0.164
0.463
0.812
TRE
0.252
0.044
0.748
0.800
0.115
0.590
0.959
SOM
0.359
0.051
0.641
0.739
0.250
0.481
0.750
IST
0.194
0.027
0.806
0.728
0.111
0.321
0.652
SNS
0.331
0.046
0.669
0.716
0.209
0.469
0.733
PAR
0.370
0.041
0.630
0.642
0.218
0.416
0.691
VRG
0.457
0.062
0.543
0.724
0.321
0.537
0.778
G
0.687
0.687
0.687
0.936
0.886
0.940
0.959
Note.ECV_SS: Proportion of common variance within a subscale explained by that specific factor; ECV_GS: Proportion due to G. OmegaH: Hierarchical omega for the specific factor (reliable variance above and beyond G). H: Construct replicability; FD: Factor determinacy. OmegaH < .50 for specific factors indicates subscale scores lack reliable unique variance (Rodriguez et al., 2016).
Table 17: Item-Level Explained Common Variance (IECV) — Bifactor Model
IECV
dpr1
0.988
dpr2
0.694
dpr3
0.824
dpr4
0.550
tre1
0.979
tre2
0.405
tre3
0.965
tre4
0.992
som1
0.804
som2
0.649
som3
0.498
som4
0.663
ist1
0.975
ist2
0.708
ist3
0.646
ist4
0.964
sns1
0.823
sns2
0.614
sns3
0.372
sns4
0.944
par1
0.546
par2
0.828
par3
0.521
par4
0.717
vrg1
0.586
vrg2
0.408
vrg3
0.582
vrg4
0.622
Note. IECV = Item Explained Common Variance (proportion of an item’s common variance attributable to G). IECV > .70 indicates the item primarily reflects general distress; IECV < .50 indicates more unique subscale variance than general variance.
Table 18: Comparison of Competing Measurement Models (Robust Fit Indices)
Model
χ² (Scaled)
df
CFI (Robust)
RMSEA (Robust)
SRMR
M1: Single Factor (CMB)
3915
350
0.813
0.080
0.056
M2: Correlated Factors
1876
329
0.912
0.057
0.037
M3: Hierarchical (Primary)
1983
343
0.906
0.057
0.040
M4: Bifactor
1536
322
0.935
0.049
0.035
7 Model Fit Evaluation: Hierarchical Second-Order CFA (M2)
7.1 Global Fit Indices
Reporting strategy. We report robust fit indices (CFI, TLI, RMSEA, SRMR) derived from the WLSMV estimator. Robust indices are preferred over standard DWLS indices for large samples with ordinal data, as standard indices inflate CFI/TLI under large N (DiStefano & Morgan, 2014).
Note. MI = expected decrease in χ² if the parameter is freed; EPC = expected parameter change. Post-hoc modifications require theoretical justification and cross-validation.
8 Higher-Order Reliability: Hierarchical Second-Order CFA (M2)
Rationale. Reliability and convergent/discriminant validity of the first-order subscales were assessed on the correlated factors model (M1) — see Section above. For the hierarchical model (M2), we report only the higher-order reliability coefficients, which quantify how well the second-order general factor (SCL28) accounts for variance in the subscale composites (Zinbarg et al., 2005).
8.1 Internal Consistency: Ordinal Alpha and McDonald’s Categorical Omega
What this section reports.
(1) Ordinal alpha (α_ord), computed from polychoric correlations, is the appropriate analogue of Cronbach’s α for ordinal data.
(2) McDonald’s categorical omega (ω), derived from the CFA model, reflects the proportion of total score variance attributable to the common factor. Unlike α, ω does not assume tau-equivalence and is the preferred reliability index for congeneric models. Threshold: ω ≥ .70 acceptable; ω ≥ .80 good.
Note. These coefficients pertain to the second-order factor (SCL28). Reliability and validity of the first-order subscales are reported in the Correlated Factors section (M1), which is the methodologically appropriate baseline (Hair et al., 2022).
What this section reports. For hierarchical models, variance in item scores originates from both first-order subscale factors and the second-order general factor. Three specialized coefficients are reported (Zinbarg et al., 2005):
omegaL1: Total reliability of the SCL-28 sum score (all common factors combined).
omegaL2: Variance attributable solely to SCL28. omegaL2 > .70 justifies using the total SCL-28 score.
partialOmegaL1: Incremental reliability of subscale factors beyond the general factor.
Show / Hide Code
reliability_L2 <- semTools::reliabilityL2(fit.cfa, "SCL28")data.frame(Coefficient =c("omegaL1","omegaL2","partialOmegaL1"),Value =round(as.numeric(reliability_L2), 3),Interpretation =c("Total reliability of the SCL-28 sum score (all common factors)","Reliability attributable to the general distress factor (SCL28) alone","Subscale-level reliability after controlling for the general factor" )) %>%kable(align =c("l","c","l"), booktabs =TRUE) %>%kable_styling(bootstrap_options =c("striped","hover"),full_width =FALSE, font_size =13)
Total reliability of the SCL-28 sum score (all common factors)
omegaL2
0.961
Reliability attributable to the general distress factor (SCL28) alone
partialOmegaL1
0.947
Subscale-level reliability after controlling for the general factor
Note. Computed via semTools::reliabilityL2() (Zinbarg et al., 2005). omegaL2 > .70 justifies the use of the total SCL-28 score as a reliable index of general distress.
Note. g = general psychological distress factor (Schmid-Leiman orthogonalized). F1*–F7* = domain-specific residual factors after removing g. h2 = communality; p2 = proportion of communality due to g.
9.2 Omega-Hierarchical
Show / Hide Code
# --- 3. Omega-Hierarchical and Related Metrics ---omega_sl <- psych::omega(m = poly_matrix,nfactors =7,n.obs =sum(df$wt),rotate ="oblimin",fm ="wls",plot =TRUE)
message("Categorical Omega (ω) for Media Traumatization:")compRelSEM(fit.cfa.mtrvm)
MTRVM
0.798
12 Mediation Analysis
A structural equation modeling framework was used to test the hypothesized mediation model. All three paths (a, b, c’) were estimated simultaneously in a single WLSMV model to avoid bias from sequential estimation. The indirect effect (a × b) was obtained analytically from the product of path coefficients.
Model structure. X — PTSD (independent variable) M — Media Traumatization (mediator) Y — Psychological Distress / SCL-28 (dependent variable)
Show / Hide Code
cat("Setup model")
Setup model
Show / Hide Code
# ── Model specification ────────────────────────────────────────────────────────model.sem <-' # Measurement model DPR =~ dpr1 + dpr2 + dpr3 + dpr4 TRE =~ tre1 + tre2 + tre3 + tre4 SOM =~ som1 + som2 + som3 + som4 IST =~ ist1 + ist2 + ist3 + ist4 SNS =~ sns1 + sns2 + sns3 + sns4 PAR =~ par1 + par2 + par3 + par4 VRG =~ vrg1 + vrg2 + vrg3 + vrg4 SCL28 =~ DPR + TRE + SOM + IST + SNS + PAR + VRG PTSR =~ ptsr1 + ptsr2 + ptsr3 + ptsr4 + ptsr5 ptsr2 ~~ ptsr3 MTRVM =~ mtrvm1 + mtrvm2 + mtrvm3 + mtrvm4 + mtrvm5 mtrvm1 ~~ mtrvm2 # Structural model (mediation) SCL28 ~ c_prime*PTSR + b*MTRVM MTRVM ~ a*PTSR # Defined parameters direct := c_prime indirect := a * b total := c_prime + (a * b) prop_mediated := (a * b) / (c_prime + (a * b))'# ── Fit model ─────────────────────────────────────────────────────────────────fit.sem <-sem(model = model.sem,data = df,estimator ="WLSMV",ordered =TRUE,sampling.weights ="wt",std.lv =TRUE,se ="robust")# ── All parameter estimates — single source for all tables ────────────────────params.sem <-parameterEstimates(fit.sem, standardized =TRUE, ci =TRUE)# ── Effects table (defined parameters) ───────────────────────────────────────effects.sem <- params.sem %>%filter(op ==":=") %>%select(label, est, se, z, pvalue, ci.lower, ci.upper, std.all) %>%mutate(Effect_Name =case_when( label =="direct"~"Direct Effect (c')", label =="indirect"~"Indirect Effect (a x b)", label =="total"~"Total Effect (c' + ab)", label =="prop_mediated"~"Proportion Mediated",TRUE~ label ),p_fmt =case_when( pvalue < .001~"< .001",TRUE~sprintf("%.3f", pvalue) ),sig =case_when( pvalue < .001~"***", pvalue < .01~"**", pvalue < .05~"*",TRUE~"" ),ci_fmt =sprintf("[%.3f, %.3f]", ci.lower, ci.upper) )# ── Proportion mediated footnote string (built once, reused) ──────────────────prop_row <- effects.sem %>%filter(label =="prop_mediated")prop_footnote <-sprintf("Proportion mediated = %.1f%% (95%% CI: [%.1f%%, %.1f%%], z = %.2f, p < .001).", prop_row$est *100, prop_row$ci.lower *100, prop_row$ci.upper *100, prop_row$z)
Table 32: Structural Regression Coefficients — SEM Mediation Model (WLSMV, Robust SE)
Path
B
SE
z
95% CI
beta (std)
c' (PTSD -> Distress, Direct)
1.171
0.055
21.28
[1.063, 1.279]
0.723
***
b (Media Trauma -> Distress)
0.116
0.030
3.93
[0.058, 0.174]
0.094
***
a (PTSD -> Media Trauma)
0.844
0.037
23.06
[0.772, 0.915]
0.645
***
Note: *** p < .001 ** p < .01 * p < .05. B = unstandardized coefficient; beta (std) = fully standardized coefficient; 95% CI estimated via the delta method (robust SE).
Show / Hide Code
effects.sem %>%filter(label !="prop_mediated") %>%select(Effect_Name, est, se, z, ci_fmt, std.all, sig) %>%kable(digits =3,col.names =c("Effect", "B", "SE", "z", "95% CI", "beta (std)", ""),align ="lrrrrrc") %>%kable_styling(bootstrap_options =c("striped", "hover"), full_width =FALSE, font_size =13) %>%column_spec(1, bold =TRUE, width ="7cm") %>%column_spec(5, width ="3.5cm") %>%footnote(general =paste("*** p < .001 ** p < .01 * p < .05.","B = unstandardized coefficient; beta (std) = fully standardized coefficient;","95% CI estimated via the delta method (robust SE).", prop_footnote ),general_title ="Note:",footnote_as_chunk =TRUE )
Table 33: Decomposition of Mediation Effects with 95% Confidence Intervals (Delta Method)
Effect
B
SE
z
95% CI
beta (std)
Direct Effect (c')
1.171
0.055
21.28
[1.063, 1.279]
0.723
***
Indirect Effect (a x b)
0.098
0.025
3.96
[0.049, 0.146]
0.060
***
Total Effect (c' + ab)
1.269
0.047
27.05
[1.177, 1.361]
0.783
***
Note: *** p < .001 ** p < .01 * p < .05. B = unstandardized coefficient; beta (std) = fully standardized coefficient; 95% CI estimated via the delta method (robust SE). Proportion mediated = 7.7% (95% CI: [3.8%, 11.6%], z = 3.87, p < .001).
The indirect effect is statistically significant (p < .001; 95% CI: [0.049, 0.146]).
Media Traumatization mediates 7.7% of the total PTSD -> Distress effect (effect size: small).
The substantial direct path (beta = 0.723) indicates partial mediation.
Дембіцький, С., Головаха, Є., & Степаненко, В. (2025). Стресові стани населення України в контексті війни: досвід соціологічного вивчення. Інститут соціології НАН України.
How to cite (APA 7th ed.):
Bova, A. A. (2026). PTSD and distress among the Ukrainian population during the war: Does media traumatization play a mediating role? [Analytical Code Report — Working Version]. RPubs. https://rpubs.com/abova/mediation-analysis
BibTeX:
@misc{bova2026,author = {Bova, Andrii A.},title = {PTSD and Distress Among the Ukrainian Population During the War: Does Media Traumatization Play a Mediating Role?},subtitle = {Analytical Code Report — Working Version},year = {2026},url = {https://rpubs.com/abova/mediation-analysis}}
