A. Univariate Model Pre-Selection

Fit Indices

The best-fitting model was Model 3: Constant + L2C + C2C, determined via LRT (p = <.001) and fit indices (AIC = 6146.736, BIC = 6176.933).

Fit Indices

The best-fitting model was Model 1: Constant Change Only, determined via LRT (p = 0.705) and fit indices (AIC = 11864.149, BIC = 11886.796).

B. Bivariate Model Results

Model Fit Statistics

Best Fitting Bivariate Model: B4a: +L2C Equal & C2C Equal via nested model Likelihood Ratio Tests and Information Critera (AIC=17846.663, BIC=17925.929). The “No Coupling” model (and all subsequent bivariate models) is constructed using the winning univariate models from above and all of the possible intercept and slope covariances between the two constructs (six in total). Models B2-B5 add the parameters listed in their respective labels to the “No Coupling” model.

Fit Indices

Nested Model Tests

Parameter Estimates

Random Effects

Fixed Effects

Effect Sizes and Model Performance

Standardized Effect Sizes ⓘ

C. Reliable Change Index (RCI) Analysis

Note: Classifications below are based on the bivariate model, which accounts for cross-construct relationships and provides more accurate error estimates for the RCI calculations than the univariate models. Use these classifications; the univariate ones above are provided just for the sake of comparison.

Individual Construct RCI Classifications

Note. RCI analysis includes only subjects with ≥2 valid observations (n = 283). Change scores calculated as difference between earliest and latest observed timepoints for each subject. Those with only one valid observation cannot be classified using RCI methodology.

Note. RCI analysis includes only subjects with ≥2 valid observations (n = 283). Change scores calculated as difference between earliest and latest observed timepoints for each subject. Those with only one valid observation cannot be classified using RCI methodology.

Cross-Tabulation of Change Classifications

Fisher's Exact Test: p = 0.030 (recommended)
χ² test not available (structural zeros present)
Note: 72.0% of cells have expected count < 5, so Fisher's Exact Test is preferred.

Instructions: First click a row OR column header in the table ABOVE (turns light green). Then click any specific row in the table BELOW to see the odds ratio interpretation below for that particular row/column. Red cells = overrepresented, Blue = underrepresented.

Your System's Change Equations and Directionality

Change equation for Self-blame:

Change equation for PTSD:

Model Selection

Univariate Model Pre-Selection

Three nested models were examined for Self-blame: constant change only, constant change + proportional change (adding L2C), and a full model further adding sequential change (constant change + L2C + C2C). Adding proportional change improved fit over the constant change-only baseline (Δχ²(1) = 25.46, p = <.001). Adding sequential change further improved fit (Δχ²(1) = 11.73, p = <.001). The selected Model 3: Constant + L2C + C2C showed these absolute fit statistics: χ²(27) = 238.98, p = <.001; CFI = 0.834; RMSEA = 0.156; SRMR = 0.137; AIC = 6146.7; BIC = 6176.9.

The same three nested models were tested for PTSD. Adding proportional change did not improve fit over the constant change-only baseline (Δχ²(1) = 0.14, p = 0.705). Adding sequential change did not improve fit (Δχ²(1) = 0.00, p = 0.950). The selected Model 1: Constant Change Only showed these absolute fit statistics: χ²(29) = 131.40, p = <.001; CFI = 0.939; RMSEA = 0.105; SRMR = 0.092; AIC = 11864.1; BIC = 11886.8.

Bivariate Model Selection

The best-fitting univariate models were combined into a baseline bivariate model (B1) with no cross-construct coupling, but including all possible intercept and slope covariances between the two constructs. We first tested whether cross-construct proportional coupling (L2C) was needed by comparing B1 (no coupling) against B2 (equal L2C between constructs) and B3 (free L2C parameters). The comparison between equal vs. free L2C parameters (B2 vs B3: Δχ²(1) = 1.40, p = 0.237) favored B2 (equal L2C), leading us to include constrained L2C parameters in subsequent models. We then tested whether adding cross-construct sequential coupling (C2C) improved fit. Adding equal C2C parameters to the equal L2C base (B2 vs B4a: Δχ²(1) = 12.44, p = <.001) significantly improved fit. We then tested whether freeing the C2C parameters further improved fit (B4a vs B5a: Δχ²(1) = 3.58, p = 0.059) and found no significant improvement. The final selected model (B4a) significantly outperformed the no-coupling baseline (Δχ²(4) = 44.50, p = <.001), justifying the inclusion of the retained cross-construct coupling effects. Fit statistics of the final selected model were: χ²(98) = 364.45, p = <.001; CFI = 0.914; RMSEA = 0.092; SRMR = 0.109; AIC = 17846.7; BIC = 17925.9.

Overall Average Trajectories

Self-blame

Based on the model parameters, individuals starting at mean levels showed minimal net change in self-blame over the study period (average change of -0.01 points per wave, total change = -0.03 points). This pattern reflects the combined influence of positive constant change (1.67), proportional damping (L2C = -0.44), positive momentum (C2C = 0.39), and PTSD prior change effects (C2C cross = 0.18).

PTSD

Based on the model parameters, individuals starting at mean levels showed improvement in PTSD by an average of 1.64 points per wave, decreasing from 56.6 to 46.8 over the study period. This pattern reflects the combined influence of negative constant change (-1.67) and self-blame prior change effects (C2C cross = 0.18).

Note: In this coupled system, trajectories represent the net result of both within- and across-construct forces. When net change appears minimal, this often reflects a balance of competing dynamics rather than absence of systematic effects. A stable trajectory in one construct may be maintained by regulatory forces within that construct, by influences from the other construct, or by opposing forces that cancel out. Understanding these system-level dynamics is essential for identifying effective intervention targets.

The sections below provide more detail: The Underlying Forces section breaks down each individual component (constant change, proportional effects, momentum, and cross-construct coupling), while the Integrated Story section explains how these forces interact dynamically to produce the observed patterns.

The Underlying Forces

Growth Parameters

Self-blame

• Constant change: There was a significant positive constant change component (b = 1.67, SE = 0.54, p = 0.002, β = 2.44), contributing upward pressure (worsening direction) to change scores. This represents the systematic additive change that occurs each wave before accounting for any proportional change and momentum effects that may be present in the model.
• Constant change variance: People showed significant individual differences in their constant change component (σ² = 0.47, SE = 0.11, p = <.001), meaning the baseline rate of change varied across individuals.
• Intercept variance: There was significant variability in starting points (σ² = 3.73, SE = 0.37, p = <.001) indicating that people began with different levels of self-blame.
• Intercept-constant change covariance: Higher starting levels were associated with larger constant change components (b = 0.94, SE = 0.15, p = <.001), indicating that initial levels predicted systematic additive change tendencies.

PTSD

• Constant change: There was a significant negative constant change component (b = -1.67, SE = 0.15, p = <.001, β = -0.87), contributing downward pressure (improvement direction) to change scores. This represents the systematic additive change that occurs each wave before accounting for any proportional change and momentum effects that may be present in the model.
• Constant change variance: People showed significant individual differences in their constant change component (σ² = 3.64, SE = 0.49, p = <.001), meaning the baseline rate of change varied across individuals.
• Intercept variance: There was significant variability in starting points (σ² = 95.23, SE = 8.97, p = <.001) indicating that people began with different levels of PTSD.
• Intercept-constant change covariance: Starting levels were not significantly related to constant change (b = 1.82, SE = 1.50, p = 0.224).

Developmental Coupling

This section examines how growth parameters are related across constructs, independent of the regulatory effects described in the two sections below. Note that these covariances describe relationships amongst initial levels and constant change parameters rather than observed trajectories themselves. Understanding these relationships reveals whether constructs share common developmental origins and whether initial severity in one domain predicts systematic change tendencies in the other.

• Initial Level Coupling (Intercept-Intercept Covariance): Starting levels were positively related (cov = 5.77, p = <.001). This indicates that individuals starting with more severe self-blame also tend to have higher initial PTSD, suggesting these problems co-occur and may share common risk factors or vulnerability pathways.
• Constant Change Component Coupling (Slope-Slope Covariance): Constant change components were not significantly related (cov = -0.06, p = 0.732), indicating that individual differences in systematic additive change tendencies are independent across domains. Whatever drives baseline change pressure in self-blame operates separately from what drives change in PTSD.
• Cross-Domain Predictive Coupling (Self-blame Initial Level → PTSD Constant Change): Initial self-blame did not significantly predict constant change tendencies in PTSD (cov = 0.54, p = 0.098), indicating that starting levels in one domain do not systematically relate to baseline change pressures in the other.
• Cross-Domain Predictive Coupling (PTSD Initial Level → Self-blame Constant Change): Initial PTSD did not significantly predict constant change tendencies in self-blame (cov = 1.72, p = 0.080), indicating that starting levels in one domain do not systematically relate to baseline change pressures in the other.

For how these constructs dynamically influence each other's trajectories through moment-to-moment regulatory processes, see the Cross-Construct Effects section below.

Regulatory Effects

Within-Construct Effects

Self-blame

• Proportional Change (L2C): There was a significant buffering effect where higher current levels of self-blame predicted smaller subsequent increases (b = -0.44, SE = 0.05, p = <.001, β = -1.37). This creates self-correcting dynamics where those with more problems tend to improve naturally over time (recovery effects).
• Sequential Change (C2C): There was a significant momentum effect where recent increases in self-blame predicted continued increases in subsequent periods (b = 0.39, SE = 0.13, p = 0.003, β = 0.47). This creates cascading dynamics where changes build upon themselves - worsening creates downward spirals that accelerate decline.

PTSD

• Proportional Change (L2C): How current levels influence future changes. Not estimated in the selected model.
• Sequential Change (C2C): How recent changes influence future changes. Not estimated in the selected model.

Cross-Construct Effects

• Self-blame Level → PTSD Change (Proportional Cross-Construct Coupling): Current levels of self-blame did not significantly predict subsequent changes in PTSD through proportional coupling (b = 0.01, SE = 0.01, p = 0.590, β = 0.01), suggesting that momentary levels in one domain do not directly influence the rate of change in the other through level-dependent mechanisms.
• Self-blame Change → PTSD Change (Sequential Cross-Construct Coupling): Despite statistical significance, the sequential cross-construct effect was negligible (b = 0.18, SE = 0.06, p = 0.001, β = 0.01), meaning self-blame's worsening had minimal practical impact on PTSD's trajectory.
• PTSD Level → Self-blame Change (Proportional Cross-Construct Coupling): Current levels of PTSD did not significantly predict subsequent changes in self-blame through proportional coupling (b = 0.01, SE = 0.01, p = 0.590, β = 0.11), suggesting that momentary levels in one domain do not directly influence the rate of change in the other through level-dependent mechanisms.
• PTSD Change → Self-blame Change (Sequential Cross-Construct Coupling): As PTSD improves but self-blame worsens, recent improvement in PTSD predicted subsequent improvement in self-blame (b = 0.18, SE = 0.06, p = 0.001, β = 0.77). This reveals a protective countervailing force where PTSD's improvement generates momentum in the opposite direction of self-blame's overall worsening trajectory. As PTSD improves, it creates cascading forces that partially resist self-blame's deterioration, slowing its rate of decline. While insufficient to reverse self-blame's dominant worsening trend, this cross-construct influence provides meaningful buffering—self-blame would worsen faster without PTSD's improving momentum.

Integrated Story

Self-blame and PTSD showed opposing constant change parameters (b = 1.67, p = 0.002 vs. b = -1.67, p = <.001), establishing divergent underlying baseline change tendencies. Two of six covariances were significant, indicating selective developmental coupling at the growth factor level. Individuals with higher initial self-blame also had higher initial PTSD (cov = 5.77, p = <.001), suggesting shared vulnerability or co-occurrence at baseline. Higher baseline self-blame predicted faster worsening (cov = 0.94, p = <.001), though baseline PTSD did not predict its own constant change component (cov = 1.82, p = 0.224). Cross-construct intercept-slope covariances were non-significant (self-blame → PTSD: cov = 0.54, p = 0.098; PTSD → self-blame: cov = 1.72, p = 0.080), indicating that baseline levels did not predict cross-construct constant change components. Constant change components were independent (cov = -0.06, p = 0.732), indicating that baseline change tendencies followed separate timelines. These patterns reveal developmental interdependence through both growth factor coupling and moment-to-moment cross-lagged regulatory influences. The following sections examine how constant change parameters, within-construct regulation, and cross-construct influences combined to produce each construct's model-implied trajectory.

Self-blame's model-implied trajectory reflected the combined influence of multiple competing forces: the positive constant change component (b = 1.67, SE = 0.54, p = 0.002, β = 2.44), large proportional damping (b = -0.44, SE = 0.05, p = <.001, β = -1.37), medium sequential momentum (b = 0.39, SE = 0.13, p = 0.003, β = 0.47), and large dampening cross-construct sequential influence from PTSD (b = 0.18, SE = 0.06, p = 0.001, β = 0.77). The proportional damping created homeostatic resistance where higher levels triggered corrective forces pulling back toward baseline, establishing self-limiting dynamics, while the sequential momentum added inertial properties where recent worsening built forward-moving pressure for continued change in the same direction. These opposing forces created tension, with the strong proportional effect (β = -1.37) resisting the substantial sequential effect (β = 0.47), and the proportional mechanism exerting greater influence. The strong cross-construct sequential influence meant that as PTSD improved, it pulled self-blame toward improvement, creating cascading momentum across constructs. The net result was a modulated trajectory where the positive constant change parameter (β = 2.44) dominated, but regulatory forces created meaningful deviations from pure linearity, producing worsening that was shaped by both internal dynamics and external cross-construct influences. These cross-construct influences had substantial predictive value, as the bivariate model captured regulatory dynamics—particularly the interplay of multiple forces producing non-linear patterns—that a univariate model treating self-blame in isolation would miss, demonstrating that even modest standardized effects can meaningfully shape developmental trajectories within complex regulatory systems.

PTSD's model-implied trajectory reflected the combined influence of two forces: the negative constant change component (b = -1.67, SE = 0.15, p = <.001, β = -0.87) and negligible amplifying cross-construct sequential influence from self-blame (b = 0.18, SE = 0.06, p = 0.001, β = 0.01). Despite statistical significance, the cross-construct sequential influence was negligible (β = 0.01), meaning self-blame's worsening had minimal practical impact on PTSD's trajectory. The net result was a trajectory that remained close to linear improvement, as the dominant constant change parameter (β = -0.87) overwhelmed regulatory forces, producing steady directional change with minimal modulation. The negligible cross-construct influence was reflected in the bivariate model's predictions closely matching those from a univariate model, confirming that PTSD's trajectory was effectively autonomous and unaffected by self-blame's dynamics.

At the system level, these constructs exhibited bidirectional coupling, though PTSD's influence on self-blame (β = 0.77) was substantially stronger than the reverse direction (β = 0.01), creating an asymmetric feedback architecture. This asymmetric coupling had meaningful predictive consequences: the bivariate model substantially improved trajectory prediction for self-blame by capturing regulatory dynamics that univariate approaches would miss, while adding negligible predictive value for PTSD, which operated essentially autonomously. This asymmetry identifies PTSD as the primary intervention leverage point, where improvements would produce both direct benefits and cascading effects on self-blame, while targeting self-blame alone would have more limited cross-domain impact.

Reliable Change Analysis

The Reliable Change Index (RCI; Jacobsen & Truax (1991)) identifies individuals whose changes in Self-blame from their earliest to latest timepoint exceed measurement error, distinguishing meaningful change from random fluctuation based on the study's measurement precision. For this study, that value is ±2.97 for the 95% confidence threshold (distinguishing reliable improvement from reliable deterioration) and ±2.48 for the 90% confidence threshold (distinguishing probable improvement from probable deterioration). Based on these thresholds, 8.48% of the sample showed significant reduction (avg. Δ = -4.42 points), 84.45% remained stable (avg. Δ = 0.10 points), and 7.07% experienced significant worsening (avg. Δ = 3.90 points). The balanced response ratio of 1.2:1 shows no significant deviation from chance (binomial test: p = 0.652).

Similarly, for PTSD, the RCI thresholds are ±16.20 (95% CI) and ±13.55 (90% CI). 23.32% of the sample showed significant reduction (avg. Δ = -25.68 points), 7.07% showed probable reduction (avg. Δ = -14.95 points), 65.02% remained stable (avg. Δ = -2.08 points), 2.83% showed probable worsening (avg. Δ = 14.88 points), and 1.77% experienced significant worsening (avg. Δ = 19.20 points). The response ratio of 6.6:1 indicates individuals were significantly more likely to show meaningful improvement than deterioration (binomial test: p = < .001).

However, these individual patterns do not tell the full story. The joint classification analysis reveals a significant association. Both the traditional independence test (Fisher's exact test: p = 0.028) and the Stuart-Maxwell test (χ² = 61.24, df = 4, p < .001), which properly accounts for the paired nature of the data (same individuals classified on both constructs), converge on the same conclusion, indicating that changes in these constructs are statistically interdependent rather than occurring independently. This interdependence is driven primarily by the following patterns in the crosstabulation: more individuals than expected by chance (3.9%) showed reliable reduction in Self-blame combined with reliable reduction in PTSD (observed = 11, expected = 5.6, std. residual = 2.28, OR = 16.16 [95% CI: 8.64, 30.23] relative to no reliable change in either construct).

Overall, 61.8% of individuals showed concordant change patterns (both constructs changing in the same direction or both remaining stable), while 38.2% showed discordant patterns (improving on one construct while deteriorating or remaining stable on the other). The significant coupling of change patterns suggests that Self-blame and PTSD may be functionally related, responding to similar mechanisms or intervention targets. Individuals experiencing positive change in one domain are more likely to experience positive change in the other, indicating potential synergistic treatment effects.

Effect Sizes & Relative Dominance

Parameter Ranking

Self-blame: With respect to the within-construct predictors, the strongest effect was the overall trajectory (β = 2.44, 95% CI [0.89, 3.99]). It was 1.78× stronger than but not significantly different (Wald z = 1.41, p = 0.159) from the proportional change effect (L2C_within) (β = 1.37, 95% CI [1.18, 1.56]). It was 5.19× stronger than but not significantly different (Wald z = 0.85, p = 0.395) from the sequential change effect (C2C_within) (β = 0.47, 95% CI [0.36, 0.58]). The proportional change effect (L2C_within) was 2.92× stronger than and significantly different (Wald z = 6.83, p < .001) from the sequential change effect (C2C_within). Considering cross-construct influences, PTSD also predicted self-blame change through sequential coupling (C2C_cross: β = 0.77, 95% CI [0.49, 1.04]). Similarly, self-blame's own momentum (C2C_within) was 0.61× as strong as but not significantly different (Wald z = 1.35, p = 0.176) from PTSD's momentum (C2C_cross). The following cross-construct effects were also in the model but were not statistically significant: L2C_cross (β = 0.11, 95% CI [-0.10, 0.32], p > .05). The complete hierarchy of effects predicting self-blame change was: (1) self-blame's trajectory (β = 2.44), (2) self-blame's proportional effect (L2C_within) (β = 1.37), (3) PTSD's sequential influence (C2C_cross) (β = 0.77), (4) self-blame's sequential effect (C2C_within) (β = 0.47).

PTSD: With respect to the within-construct predictors, the only significant predictor was the overall trajectory (β = 0.87, 95% CI [0.72, 1.03]). Considering cross-construct influences, self-blame also predicted PTSD change through sequential coupling (C2C_cross: β = 0.01, 95% CI [0.01, 0.02]). The following cross-construct effects were also in the model but were not statistically significant: L2C_cross (β = 0.01, 95% CI [-0.00, 0.01], p > .05). The complete hierarchy of effects predicting PTSD change was: (1) PTSD's trajectory (β = 0.87), (2) self-blame's sequential influence (C2C_cross) (β = 0.01).

System Architecture & Dynamics ⓘ

Self-blame: Variance partitioning reveals that internal dynamics account for approximately 93.2% of explained variance in change (averaged across timepoints), while cross-construct influences from PTSD account for 6.8%. This 13.6:1 internal:external ratio indicates that self-blame is highly autonomous. Internal processes—trajectory, proportional change mechanisms (L2C), and sequential dynamics (C2C)—dominate the change architecture. Cross-construct influences from PTSD are statistically significant but account for a minor proportion of variance.

PTSD: Variance partitioning analysis indicates that internal dynamics account for approximately 100.0% of explained variance in change (averaged across timepoints), while cross-construct influences from self-blame account for 0.0%. The internal:external ratio of 3985.4:1 indicates that PTSD is highly autonomous. Although cross-construct effects reach statistical significance (p < .05), their contribution to explained variance is minimal. PTSD change is predominantly driven by within-construct processes: its trajectory, proportional effects (how current levels predict subsequent change), and sequential momentum (how recent change predicts future change).

System asymmetry: PTSD shows substantially greater autonomy (3985.4:1) compared to self-blame (13.6:1), suggesting a hierarchical system where PTSD operates more independently while self-blame is more susceptible to external influence. This asymmetry reveals the system's architecture: PTSD has stronger internal structure and is less easily perturbed, while self-blame is more reactive and malleable. Understanding this hierarchy is crucial for efficient intervention—the most leverage comes from targeting the element with the right combination of autonomy and outgoing influence.

Cross-construct influence patterns: Bidirectional asymmetric influence: PTSD → self-blame is 55.52× stronger than self-blame → PTSD. Both directions of influence exist, but they are not equivalent. This creates an imbalanced feedback loop where one direction dominates the dynamics.

System characterization: This is an asymmetric feedback system. Both constructs influence each other (bidirectional coupling), but the relationship is unbalanced in two ways: (1) one shows greater autonomy than the other, and (2) one direction of influence (PTSD → self-blame) is stronger than the reverse. This creates complex interdependence—neither construct is fully independent, but they do not have equivalent roles in the system.

The asymmetry suggests a prioritized dual-target strategy. Focus primary intervention efforts on the stronger driver (the construct with greater outgoing influence), but do not ignore the other construct entirely—its reciprocal influence, though weaker, still matters. Consider a 70-30 or 60-40 resource allocation rather than an all-or-nothing choice. The bidirectional coupling means improvements in either construct will eventually propagate to the other, but starting with the dominant driver will produce faster and more stable system-wide change. Once the primary driver shows improvement, the weaker reciprocal influence can help maintain gains through positive feedback, even if intervention intensity on the driver is reduced.

Model Performance

Model performance is assessed by examining the proportion of variance explained (R²) in each construct's observed scores across timepoints. Higher R² indicates the specified mechanisms (trajectory, L2C, C2C, and cross-construct coupling) successfully account for observed change patterns.

Construct-Specific Performance

Self-blame: The model explains a very large proportion of variance (avg. R² = 0.736, range: 0.697–0.765) in self-blame scores across timepoints, and this explanatory power is stable across timepoints (SD = 0.026). This indicates the model performs consistently well across all measurement occasions, providing reliable estimates of the change mechanisms driving self-blame.

PTSD: The model explains a very large proportion of variance (avg. R² = 0.805, range: 0.736–0.881) in PTSD change, though this explanatory power is moderately variable across timepoints (SD = 0.055). While average performance is good, the variation suggests the model fits some timepoints better than others, which may reflect time-varying influences on PTSD change or measurement issues at specific intervals.

Comparative Assessment

Differential explanatory power: The model explains significantly more variance in PTSD change (avg. R² = 0.805) than in self-blame change (avg. R² = 0.736), with a mean difference of 6.9 percentage points (t(6) = 3.61, p = 0.011). This asymmetry suggests PTSD change follows a more predictable, systematic pattern that is better captured by the specified mechanisms (trajectory, L2C, C2C), while self-blame change may involve additional unmodeled influences or greater measurement error.

Comparable temporal stability: Both constructs show similar stability in prediction across time (Self-blame SD = 0.026; PTSD SD = 0.055), F = 0.22, p = 0.090. The model performs with consistent reliability for both constructs across all timepoints.

Model Figure

Full Model Output

Model Fit Statistics

Parameter Estimates

Lavaan & Mplus Model Code

Click each button to expand or collapse the full model syntax for the best-fitting bivariate model, annotated for clarity.

# Define Latent Variables
lsb0 =~ 1*sb0
lsb1 =~ 1*sb1
lsb2 =~ 1*sb2
lsb3 =~ 1*sb3
lsb4 =~ 1*sb4
lsb5 =~ 1*sb5
lsb6 =~ 1*sb6
isb0 =~ 1*lsb0
s_sb =~ 1*dsb0 + 1*dsb1 + 1*dsb2 + 1*dsb3 + 1*dsb4 + 1*dsb5
lpcls0 =~ 1*pcls0
lpcls1 =~ 1*pcls1
lpcls2 =~ 1*pcls2
lpcls3 =~ 1*pcls3
lpcls4 =~ 1*pcls4
lpcls5 =~ 1*pcls5
lpcls6 =~ 1*pcls6
ipcls0 =~ 1*lpcls0
s_pcls =~ 1*dpcls0 + 1*dpcls1 + 1*dpcls2 + 1*dpcls3 + 1*dpcls4 + 1*dpcls5
# Define Change Scores
dsb0 =~ 1*lsb1
dsb1 =~ 1*lsb2
dsb2 =~ 1*lsb3
dsb3 =~ 1*lsb4
dsb4 =~ 1*lsb5
dsb5 =~ 1*lsb6
dpcls0 =~ 1*lpcls1
dpcls1 =~ 1*lpcls2
dpcls2 =~ 1*lpcls3
dpcls3 =~ 1*lpcls4
dpcls4 =~ 1*lpcls5
dpcls5 =~ 1*lpcls6
# Define Growth Factors
isb0 =~ 1*lsb0
s_sb =~ 1*dsb0 + 1*dsb1 + 1*dsb2 + 1*dsb3 + 1*dsb4 + 1*dsb5
ipcls0 =~ 1*lpcls0
s_pcls =~ 1*dpcls0 + 1*dpcls1 + 1*dpcls2 + 1*dpcls3 + 1*dpcls4 + 1*dpcls5
# Set Autoregressive Paths
lsb0 =~ 1*sb0
lsb1 =~ 1*sb1
lsb2 =~ 1*sb2
lsb3 =~ 1*sb3
lsb4 =~ 1*sb4
lsb5 =~ 1*sb5
lsb6 =~ 1*sb6
lsb1 ~ 1*lsb0
lsb2 ~ 1*lsb1
lsb3 ~ 1*lsb2
lsb4 ~ 1*lsb3
lsb5 ~ 1*lsb4
lsb6 ~ 1*lsb5
dsb0 =~ 1*lsb1
dsb1 =~ 1*lsb2
dsb2 =~ 1*lsb3
dsb3 =~ 1*lsb4
dsb4 =~ 1*lsb5
dsb5 =~ 1*lsb6
isb0 =~ 1*lsb0
s_sb =~ 1*dsb0 + 1*dsb1 + 1*dsb2 + 1*dsb3 + 1*dsb4 + 1*dsb5
lpcls0 =~ 1*pcls0
lpcls1 =~ 1*pcls1
lpcls2 =~ 1*pcls2
lpcls3 =~ 1*pcls3
lpcls4 =~ 1*pcls4
lpcls5 =~ 1*pcls5
lpcls6 =~ 1*pcls6
lpcls1 ~ 1*lpcls0
lpcls2 ~ 1*lpcls1
lpcls3 ~ 1*lpcls2
lpcls4 ~ 1*lpcls3
lpcls5 ~ 1*lpcls4
lpcls6 ~ 1*lpcls5
dpcls0 =~ 1*lpcls1
dpcls1 =~ 1*lpcls2
dpcls2 =~ 1*lpcls3
dpcls3 =~ 1*lpcls4
dpcls4 =~ 1*lpcls5
dpcls5 =~ 1*lpcls6
ipcls0 =~ 1*lpcls0
s_pcls =~ 1*dpcls0 + 1*dpcls1 + 1*dpcls2 + 1*dpcls3 + 1*dpcls4 + 1*dpcls5
# Proportional Change: Self-blame level → Self-blame change
dsb0 ~ l2c_sb*lsb0
dsb1 ~ l2c_sb*lsb1
dsb2 ~ l2c_sb*lsb2
dsb3 ~ l2c_sb*lsb3
dsb4 ~ l2c_sb*lsb4
dsb5 ~ l2c_sb*lsb5
# Cross-Construct Proportional Change: PTSD level → Self-blame change
dsb0 ~ l2c_cross*lpcls0
dsb1 ~ l2c_cross*lpcls1
dsb2 ~ l2c_cross*lpcls2
dsb3 ~ l2c_cross*lpcls3
dsb4 ~ l2c_cross*lpcls4
dsb5 ~ l2c_cross*lpcls5
# Cross-Construct Proportional Change: Self-blame level → PTSD change
dpcls0 ~ l2c_cross*lsb0
dpcls1 ~ l2c_cross*lsb1
dpcls2 ~ l2c_cross*lsb2
dpcls3 ~ l2c_cross*lsb3
dpcls4 ~ l2c_cross*lsb4
dpcls5 ~ l2c_cross*lsb5
# Sequential Change: Self-blame change → Self-blame change
dsb1 ~ c2c_sb*dsb0
dsb2 ~ c2c_sb*dsb1
dsb3 ~ c2c_sb*dsb2
dsb4 ~ c2c_sb*dsb3
dsb5 ~ c2c_sb*dsb4
# Cross-Construct Sequential Change: PTSD change → Self-blame change
dsb1 ~ c2c_cross*dpcls0
dsb2 ~ c2c_cross*dpcls1
dsb3 ~ c2c_cross*dpcls2
dsb4 ~ c2c_cross*dpcls3
dsb5 ~ c2c_cross*dpcls4
# Cross-Construct Sequential Change: Self-blame change → PTSD change
dpcls1 ~ c2c_cross*dsb0
dpcls2 ~ c2c_cross*dsb1
dpcls3 ~ c2c_cross*dsb2
dpcls4 ~ c2c_cross*dsb3
dpcls5 ~ c2c_cross*dsb4
# Residual Variances
sb0 ~~ ra*sb0
sb1 ~~ ra*sb1
sb2 ~~ ra*sb2
sb3 ~~ ra*sb3
sb4 ~~ ra*sb4
sb5 ~~ ra*sb5
sb6 ~~ ra*sb6
pcls0 ~~ rb*pcls0
pcls1 ~~ rb*pcls1
pcls2 ~~ rb*pcls2
pcls3 ~~ rb*pcls3
pcls4 ~~ rb*pcls4
pcls5 ~~ rb*pcls5
pcls6 ~~ rb*pcls6
# Residual Covariances
lsb0 ~~ 0*lsb0
lsb1 ~~ 0*lsb1
lsb2 ~~ 0*lsb2
lsb3 ~~ 0*lsb3
lsb4 ~~ 0*lsb4
lsb5 ~~ 0*lsb5
lsb6 ~~ 0*lsb6
dsb0 ~~ 0*dsb0
dsb1 ~~ 0*dsb1
dsb2 ~~ 0*dsb2
dsb3 ~~ 0*dsb3
dsb4 ~~ 0*dsb4
dsb5 ~~ 0*dsb5
dsb0 ~~ 0*dsb1 + 0*dsb2 + 0*dsb3 + 0*dsb4 + 0*dsb5
dsb1 ~~ 0*dsb2 + 0*dsb3 + 0*dsb4 + 0*dsb5
dsb2 ~~ 0*dsb3 + 0*dsb4 + 0*dsb5
dsb3 ~~ 0*dsb4 + 0*dsb5
dsb4 ~~ 0*dsb5
lpcls0 ~~ 0*lpcls0
lpcls1 ~~ 0*lpcls1
lpcls2 ~~ 0*lpcls2
lpcls3 ~~ 0*lpcls3
lpcls4 ~~ 0*lpcls4
lpcls5 ~~ 0*lpcls5
lpcls6 ~~ 0*lpcls6
dpcls0 ~~ 0*dpcls0
dpcls1 ~~ 0*dpcls1
dpcls2 ~~ 0*dpcls2
dpcls3 ~~ 0*dpcls3
dpcls4 ~~ 0*dpcls4
dpcls5 ~~ 0*dpcls5
dpcls0 ~~ 0*dpcls1 + 0*dpcls2 + 0*dpcls3 + 0*dpcls4 + 0*dpcls5
dpcls1 ~~ 0*dpcls2 + 0*dpcls3 + 0*dpcls4 + 0*dpcls5
dpcls2 ~~ 0*dpcls3 + 0*dpcls4 + 0*dpcls5
dpcls3 ~~ 0*dpcls4 + 0*dpcls5
dpcls4 ~~ 0*dpcls5
sb0 ~~ wc*pcls0
sb1 ~~ wc*pcls1
sb2 ~~ wc*pcls2
sb3 ~~ wc*pcls3
sb4 ~~ wc*pcls4
sb5 ~~ wc*pcls5
sb6 ~~ wc*pcls6
# Fixed Zero Constraints
sb0 ~ 0*1
sb1 ~ 0*1
sb2 ~ 0*1
sb3 ~ 0*1
sb4 ~ 0*1
sb5 ~ 0*1
sb6 ~ 0*1
lsb0 ~ 0*1
lsb1 ~ 0*1
lsb2 ~ 0*1
lsb3 ~ 0*1
lsb4 ~ 0*1
lsb5 ~ 0*1
lsb6 ~ 0*1
dsb0 ~ 0*1
dsb1 ~ 0*1
dsb2 ~ 0*1
dsb3 ~ 0*1
dsb4 ~ 0*1
dsb5 ~ 0*1
lsb0 ~~ 0*lsb0
lsb1 ~~ 0*lsb1
lsb2 ~~ 0*lsb2
lsb3 ~~ 0*lsb3
lsb4 ~~ 0*lsb4
lsb5 ~~ 0*lsb5
lsb6 ~~ 0*lsb6
dsb0 ~~ 0*dsb0
dsb1 ~~ 0*dsb1
dsb2 ~~ 0*dsb2
dsb3 ~~ 0*dsb3
dsb4 ~~ 0*dsb4
dsb5 ~~ 0*dsb5
dsb0 ~~ 0*dsb1 + 0*dsb2 + 0*dsb3 + 0*dsb4 + 0*dsb5
dsb1 ~~ 0*dsb2 + 0*dsb3 + 0*dsb4 + 0*dsb5
dsb2 ~~ 0*dsb3 + 0*dsb4 + 0*dsb5
dsb3 ~~ 0*dsb4 + 0*dsb5
dsb4 ~~ 0*dsb5
pcls0 ~ 0*1
pcls1 ~ 0*1
pcls2 ~ 0*1
pcls3 ~ 0*1
pcls4 ~ 0*1
pcls5 ~ 0*1
pcls6 ~ 0*1
lpcls0 ~ 0*1
lpcls1 ~ 0*1
lpcls2 ~ 0*1
lpcls3 ~ 0*1
lpcls4 ~ 0*1
lpcls5 ~ 0*1
lpcls6 ~ 0*1
dpcls0 ~ 0*1
dpcls1 ~ 0*1
dpcls2 ~ 0*1
dpcls3 ~ 0*1
dpcls4 ~ 0*1
dpcls5 ~ 0*1
lpcls0 ~~ 0*lpcls0
lpcls1 ~~ 0*lpcls1
lpcls2 ~~ 0*lpcls2
lpcls3 ~~ 0*lpcls3
lpcls4 ~~ 0*lpcls4
lpcls5 ~~ 0*lpcls5
lpcls6 ~~ 0*lpcls6
dpcls0 ~~ 0*dpcls0
dpcls1 ~~ 0*dpcls1
dpcls2 ~~ 0*dpcls2
dpcls3 ~~ 0*dpcls3
dpcls4 ~~ 0*dpcls4
dpcls5 ~~ 0*dpcls5
dpcls0 ~~ 0*dpcls1 + 0*dpcls2 + 0*dpcls3 + 0*dpcls4 + 0*dpcls5
dpcls1 ~~ 0*dpcls2 + 0*dpcls3 + 0*dpcls4 + 0*dpcls5
dpcls2 ~~ 0*dpcls3 + 0*dpcls4 + 0*dpcls5
dpcls3 ~~ 0*dpcls4 + 0*dpcls5
dpcls4 ~~ 0*dpcls5

TITLE: Mplus Equivalent of Best-Fitting Bivariate LDCS Model;
DATA: FILE = your_data.dat;
VARIABLE: NAMES = your_var_names;
USEVARIABLES = [your_vars];
ANALYSIS: ESTIMATOR = ML;

MODEL:

!Define Latent Variables
  lsb0 BY sb0@1;
  lsb1 BY sb1@1;
  lsb2 BY sb2@1;
  lsb3 BY sb3@1;
  lsb4 BY sb4@1;
  lsb5 BY sb5@1;
  lsb6 BY sb6@1;
  dsb0 BY lsb1@1;
  dsb1 BY lsb2@1;
  dsb2 BY lsb3@1;
  dsb3 BY lsb4@1;
  dsb4 BY lsb5@1;
  dsb5 BY lsb6@1;
  isb0 BY lsb0@1;
  s_sb BY dsb0@1 dsb1@1 dsb2@1 dsb3@1 dsb4@1 dsb5@1;
  lpcls0 BY pcls0@1;
  lpcls1 BY pcls1@1;
  lpcls2 BY pcls2@1;
  lpcls3 BY pcls3@1;
  lpcls4 BY pcls4@1;
  lpcls5 BY pcls5@1;
  lpcls6 BY pcls6@1;
  dpcls0 BY lpcls1@1;
  dpcls1 BY lpcls2@1;
  dpcls2 BY lpcls3@1;
  dpcls3 BY lpcls4@1;
  dpcls4 BY lpcls5@1;
  dpcls5 BY lpcls6@1;
  ipcls0 BY lpcls0@1;
  s_pcls BY dpcls0@1 dpcls1@1 dpcls2@1 dpcls3@1 dpcls4@1 dpcls5@1;

!Define Change Scores
  dsb0 BY lsb1@1;
  dsb1 BY lsb2@1;
  dsb2 BY lsb3@1;
  dsb3 BY lsb4@1;
  dsb4 BY lsb5@1;
  dsb5 BY lsb6@1;
  dpcls0 BY lpcls1@1;
  dpcls1 BY lpcls2@1;
  dpcls2 BY lpcls3@1;
  dpcls3 BY lpcls4@1;
  dpcls4 BY lpcls5@1;
  dpcls5 BY lpcls6@1;

!Define Growth Factors
  isb0 BY lsb0@1;
  s_sb BY dsb0@1 dsb1@1 dsb2@1 dsb3@1 dsb4@1 dsb5@1;
  ipcls0 BY lpcls0@1;
  s_pcls BY dpcls0@1 dpcls1@1 dpcls2@1 dpcls3@1 dpcls4@1 dpcls5@1;

!Set Autoregressive Paths
  lsb0 BY sb0@1;
  lsb1 BY sb1@1;
  lsb2 BY sb2@1;
  lsb3 BY sb3@1;
  lsb4 BY sb4@1;
  lsb5 BY sb5@1;
  lsb6 BY sb6@1;
  lsb1 ON lsb0@1;
  lsb2 ON lsb1@1;
  lsb3 ON lsb2@1;
  lsb4 ON lsb3@1;
  lsb5 ON lsb4@1;
  lsb6 ON lsb5@1;
  dsb0 BY lsb1@1;
  dsb1 BY lsb2@1;
  dsb2 BY lsb3@1;
  dsb3 BY lsb4@1;
  dsb4 BY lsb5@1;
  dsb5 BY lsb6@1;
  isb0 BY lsb0@1;
  s_sb BY dsb0@1 dsb1@1 dsb2@1 dsb3@1 dsb4@1 dsb5@1;
  lpcls0 BY pcls0@1;
  lpcls1 BY pcls1@1;
  lpcls2 BY pcls2@1;
  lpcls3 BY pcls3@1;
  lpcls4 BY pcls4@1;
  lpcls5 BY pcls5@1;
  lpcls6 BY pcls6@1;
  lpcls1 ON lpcls0@1;
  lpcls2 ON lpcls1@1;
  lpcls3 ON lpcls2@1;
  lpcls4 ON lpcls3@1;
  lpcls5 ON lpcls4@1;
  lpcls6 ON lpcls5@1;
  dpcls0 BY lpcls1@1;
  dpcls1 BY lpcls2@1;
  dpcls2 BY lpcls3@1;
  dpcls3 BY lpcls4@1;
  dpcls4 BY lpcls5@1;
  dpcls5 BY lpcls6@1;
  ipcls0 BY lpcls0@1;
  s_pcls BY dpcls0@1 dpcls1@1 dpcls2@1 dpcls3@1 dpcls4@1 dpcls5@1;

!Proportional Change: Self-blame level → Self-blame change
  dsb0 ON lsb0 (l2c_sb);
  dsb1 ON lsb1 (l2c_sb);
  dsb2 ON lsb2 (l2c_sb);
  dsb3 ON lsb3 (l2c_sb);
  dsb4 ON lsb4 (l2c_sb);
  dsb5 ON lsb5 (l2c_sb);

!Cross-Construct Proportional Change: PTSD level → Self-blame change
  dsb0 ON lpcls0 (l2c_cross);
  dsb1 ON lpcls1 (l2c_cross);
  dsb2 ON lpcls2 (l2c_cross);
  dsb3 ON lpcls3 (l2c_cross);
  dsb4 ON lpcls4 (l2c_cross);
  dsb5 ON lpcls5 (l2c_cross);

!Cross-Construct Proportional Change: Self-blame level → PTSD change
  dpcls0 ON lsb0 (l2c_cross);
  dpcls1 ON lsb1 (l2c_cross);
  dpcls2 ON lsb2 (l2c_cross);
  dpcls3 ON lsb3 (l2c_cross);
  dpcls4 ON lsb4 (l2c_cross);
  dpcls5 ON lsb5 (l2c_cross);

!Sequential Change: Self-blame change → Self-blame change
  dsb1 ON dsb0 (c2c_sb);
  dsb2 ON dsb1 (c2c_sb);
  dsb3 ON dsb2 (c2c_sb);
  dsb4 ON dsb3 (c2c_sb);
  dsb5 ON dsb4 (c2c_sb);

!Cross-Construct Sequential Change: PTSD change → Self-blame change
  dsb1 ON dpcls0 (c2c_cross);
  dsb2 ON dpcls1 (c2c_cross);
  dsb3 ON dpcls2 (c2c_cross);
  dsb4 ON dpcls3 (c2c_cross);
  dsb5 ON dpcls4 (c2c_cross);

!Cross-Construct Sequential Change: Self-blame change → PTSD change
  dpcls1 ON dsb0 (c2c_cross);
  dpcls2 ON dsb1 (c2c_cross);
  dpcls3 ON dsb2 (c2c_cross);
  dpcls4 ON dsb3 (c2c_cross);
  dpcls5 ON dsb4 (c2c_cross);

!Residual Variances / Covariances
  isb0  WITH isb0;
  s_sb  WITH s_sb;
  isb0  WITH s_sb;
  lsb0 WITH lsb0@0;
  lsb1 WITH lsb1@0;
  lsb2 WITH lsb2@0;
  lsb3 WITH lsb3@0;
  lsb4 WITH lsb4@0;
  lsb5 WITH lsb5@0;
  lsb6 WITH lsb6@0;
  dsb0 WITH dsb0@0;
  dsb1 WITH dsb1@0;
  dsb2 WITH dsb2@0;
  dsb3 WITH dsb3@0;
  dsb4 WITH dsb4@0;
  dsb5 WITH dsb5@0;
  sb0 WITH sb0 (ra);
  sb1 WITH sb1 (ra);
  sb2 WITH sb2 (ra);
  sb3 WITH sb3 (ra);
  sb4 WITH sb4 (ra);
  sb5 WITH sb5 (ra);
  sb6 WITH sb6 (ra);
  dsb0 WITH dsb1@0 dsb2@0 dsb3@0 dsb4@0 dsb5@0;
  dsb1 WITH dsb2@0 dsb3@0 dsb4@0 dsb5@0;
  dsb2 WITH dsb3@0 dsb4@0 dsb5@0;
  dsb3 WITH dsb4@0 dsb5@0;
  dsb4 WITH dsb5@0;
  ipcls0  WITH ipcls0;
  s_pcls  WITH s_pcls;
  ipcls0  WITH s_pcls;
  lpcls0 WITH lpcls0@0;
  lpcls1 WITH lpcls1@0;
  lpcls2 WITH lpcls2@0;
  lpcls3 WITH lpcls3@0;
  lpcls4 WITH lpcls4@0;
  lpcls5 WITH lpcls5@0;
  lpcls6 WITH lpcls6@0;
  dpcls0 WITH dpcls0@0;
  dpcls1 WITH dpcls1@0;
  dpcls2 WITH dpcls2@0;
  dpcls3 WITH dpcls3@0;
  dpcls4 WITH dpcls4@0;
  dpcls5 WITH dpcls5@0;
  pcls0 WITH pcls0 (rb);
  pcls1 WITH pcls1 (rb);
  pcls2 WITH pcls2 (rb);
  pcls3 WITH pcls3 (rb);
  pcls4 WITH pcls4 (rb);
  pcls5 WITH pcls5 (rb);
  pcls6 WITH pcls6 (rb);
  dpcls0 WITH dpcls1@0 dpcls2@0 dpcls3@0 dpcls4@0 dpcls5@0;
  dpcls1 WITH dpcls2@0 dpcls3@0 dpcls4@0 dpcls5@0;
  dpcls2 WITH dpcls3@0 dpcls4@0 dpcls5@0;
  dpcls3 WITH dpcls4@0 dpcls5@0;
  dpcls4 WITH dpcls5@0;
  isb0  WITH ipcls0;
  s_sb  WITH s_pcls;
  s_sb  WITH ipcls0;
  isb0  WITH s_pcls;
  sb0 WITH pcls0 (wc);
  sb1 WITH pcls1 (wc);
  sb2 WITH pcls2 (wc);
  sb3 WITH pcls3 (wc);
  sb4 WITH pcls4 (wc);
  sb5 WITH pcls5 (wc);
  sb6 WITH pcls6 (wc);

!Fixed Zero Constraints
  sb0 ON 1@0;
  sb1 ON 1@0;
  sb2 ON 1@0;
  sb3 ON 1@0;
  sb4 ON 1@0;
  sb5 ON 1@0;
  sb6 ON 1@0;
  lsb0 ON 1@0;
  lsb1 ON 1@0;
  lsb2 ON 1@0;
  lsb3 ON 1@0;
  lsb4 ON 1@0;
  lsb5 ON 1@0;
  lsb6 ON 1@0;
  dsb0 ON 1@0;
  dsb1 ON 1@0;
  dsb2 ON 1@0;
  dsb3 ON 1@0;
  dsb4 ON 1@0;
  dsb5 ON 1@0;
  lsb0 WITH lsb0@0;
  lsb1 WITH lsb1@0;
  lsb2 WITH lsb2@0;
  lsb3 WITH lsb3@0;
  lsb4 WITH lsb4@0;
  lsb5 WITH lsb5@0;
  lsb6 WITH lsb6@0;
  dsb0 WITH dsb0@0;
  dsb1 WITH dsb1@0;
  dsb2 WITH dsb2@0;
  dsb3 WITH dsb3@0;
  dsb4 WITH dsb4@0;
  dsb5 WITH dsb5@0;
  dsb0 WITH dsb1@0 dsb2@0 dsb3@0 dsb4@0 dsb5@0;
  dsb1 WITH dsb2@0 dsb3@0 dsb4@0 dsb5@0;
  dsb2 WITH dsb3@0 dsb4@0 dsb5@0;
  dsb3 WITH dsb4@0 dsb5@0;
  dsb4 WITH dsb5@0;
  pcls0 ON 1@0;
  pcls1 ON 1@0;
  pcls2 ON 1@0;
  pcls3 ON 1@0;
  pcls4 ON 1@0;
  pcls5 ON 1@0;
  pcls6 ON 1@0;
  lpcls0 ON 1@0;
  lpcls1 ON 1@0;
  lpcls2 ON 1@0;
  lpcls3 ON 1@0;
  lpcls4 ON 1@0;
  lpcls5 ON 1@0;
  lpcls6 ON 1@0;
  dpcls0 ON 1@0;
  dpcls1 ON 1@0;
  dpcls2 ON 1@0;
  dpcls3 ON 1@0;
  dpcls4 ON 1@0;
  dpcls5 ON 1@0;
  lpcls0 WITH lpcls0@0;
  lpcls1 WITH lpcls1@0;
  lpcls2 WITH lpcls2@0;
  lpcls3 WITH lpcls3@0;
  lpcls4 WITH lpcls4@0;
  lpcls5 WITH lpcls5@0;
  lpcls6 WITH lpcls6@0;
  dpcls0 WITH dpcls0@0;
  dpcls1 WITH dpcls1@0;
  dpcls2 WITH dpcls2@0;
  dpcls3 WITH dpcls3@0;
  dpcls4 WITH dpcls4@0;
  dpcls5 WITH dpcls5@0;
  dpcls0 WITH dpcls1@0 dpcls2@0 dpcls3@0 dpcls4@0 dpcls5@0;
  dpcls1 WITH dpcls2@0 dpcls3@0 dpcls4@0 dpcls5@0;
  dpcls2 WITH dpcls3@0 dpcls4@0 dpcls5@0;
  dpcls3 WITH dpcls4@0 dpcls5@0;
  dpcls4 WITH dpcls5@0;

!Time-Varying Covariates (TVCs)-ones with labels identical to blocks above are duplicative and can be deleted if so desired
  dsb0 ON lsb0 (l2c_sb);
  dsb1 ON lsb1 (l2c_sb);
  dsb2 ON lsb2 (l2c_sb);
  dsb3 ON lsb3 (l2c_sb);
  dsb4 ON lsb4 (l2c_sb);
  dsb5 ON lsb5 (l2c_sb);
  dsb1 ON dsb0 (c2c_sb);
  dsb2 ON dsb1 (c2c_sb);
  dsb3 ON dsb2 (c2c_sb);
  dsb4 ON dsb3 (c2c_sb);
  dsb5 ON dsb4 (c2c_sb);
  dsb0 ON lpcls0 (l2c_cross);
  dsb1 ON lpcls1 (l2c_cross);
  dsb2 ON lpcls2 (l2c_cross);
  dsb3 ON lpcls3 (l2c_cross);
  dsb4 ON lpcls4 (l2c_cross);
  dsb5 ON lpcls5 (l2c_cross);
  dpcls0 ON lsb0 (l2c_cross);
  dpcls1 ON lsb1 (l2c_cross);
  dpcls2 ON lsb2 (l2c_cross);
  dpcls3 ON lsb3 (l2c_cross);
  dpcls4 ON lsb4 (l2c_cross);
  dpcls5 ON lsb5 (l2c_cross);
  dsb1 ON dpcls0 (c2c_cross);
  dsb2 ON dpcls1 (c2c_cross);
  dsb3 ON dpcls2 (c2c_cross);
  dsb4 ON dpcls3 (c2c_cross);
  dsb5 ON dpcls4 (c2c_cross);
  dpcls1 ON dsb0 (c2c_cross);
  dpcls2 ON dsb1 (c2c_cross);
  dpcls3 ON dsb2 (c2c_cross);
  dpcls4 ON dsb3 (c2c_cross);
  dpcls5 ON dsb4 (c2c_cross);

Modification Indices

What are these? These values suggest specific changes you can make to your model to improve model fit. Modification Index (MI) tells you how much the model fit would improve if a path were added; values over 3.84 will result in a significant chi-square difference test at 1 degree of freedom. Expected Parameter Change (EPC) estimates the size of that path if it were added. Only consider changes that make theoretical sense. Many of these parameters are not estimated purposefully because of how the latent change score model is defined, so proceed with caution. Specific recommendations about the most statistically defensible model alterations can be found on the Residual Diagnostics tab. It is highly recommended that you attempt to make those changes before any of these.