We know that….

• Mental health outcomes are socially stratified across the life course (Kirkbride et al. 2024; Marmot and Bell 2016) but causal pathways and best routes to prevention remain contested (Alegría et al. 2018)

• Policy interest in childhood experiences, particularly childhood adversity (e.g. Scotland’s ACE-Aware Nation campaign) Davidson, Critchley, and Wright (2020)

• Limited distinction between stable inequalities (e.g., poverty, social class), and events/experiences (e.g. abuse, trauma) c.f. Yang, Hernandez, and Plastino (2023)

• Related to wider issue of under-exploration of change in childhood circumstances and studies (Adam 2004) and confounded by the correlation between “stable” inequalities and adverse experiences (see Camacho and Clark Henderson 2022; Marryat and Frank 2019)

“Structural” framework

• Poverty (Reiss 2013) and food insecurity and area-level deprivation (Abraham and Walker-Harding 2022), maternal education (Mensch et al. 2019; Meyrose et al. 2018); racism (Yang, Hernandez, and Plastino 2023; Abraham and Walker-Harding 2022; Simons et al. 2002); housing (Rolfe et al. 2020); and social class (Dougall et al. 2024)

Theorised as accumulation of risk factors for poor social and emotional development, i.e. elevated stress, e.g. stress process theory (Leonard I. Pearlin et al. 2005), which interact with the child and family through shaping environments and exposures (Bronfenbrenner 1988, 1977)

• Emerging interest in a a “Whole Family Approach” to support the family unit (Government 2023), requiring more detailed understanding of household effects and their covariates over time.

• Design interventions targeted at the root of mental health inequalities across the life course.

“Experiences” framework

• Adverse Childhood Experiences (ACEs) (Thurston et al. 2025; Felitti et al. 1998; Aces Aware, n.d.)

Emotional, physical and sexual abuse, physical and emotional neglect, domestic violence, parental separation, family incarceration, mental illness and substance abuse

• Benevolent Childhood Experiences (BCES/ PACES) (Narayan et al. 2018; Bethell et al. 2019; Burstein et al. 2021; Kallapiran et al. 2025)

Time-stable “protective” factors: family relationships, trusted adults, home stability, community organisations and support, friendships. Intertwined with ACEs (Han et al. 2023)

Unpacking these mechanisms matters because…

• Mental health issues are worsening across Scotland (The Scottish Government 2019), especially for the youngest generation (Deakin et al. 2023).

• Policy focus on adverse childhood experiences (ACEs) in Scotland (The Scottish Government 2024), despite ample critique on the concept (S. White et al. 2019) and usage (Lacey and Minnis 2020).

• Significant focus on adversity and trauma can distract stable structural inequalities and blame parents over upstream causes (Treanor 2019; Macvarish and Lee 2019), but structural frameworks alone cannot account for variance of outcomes given similar “starting” contexts.

Opportunities

• Rich, longitudinal data allows disentanglement of timing, contextual change, and short-to-medium term effects
• Repeated psychometric measures provide stable estimation of change and routes to modelling development
• Repeated measures allows short-to-medium term effects, providing greater control for potential confounders
• Prospective evidence; majority of existing evidence on positive childhood experiences relies on retrospective reporting (Kallapiran et al. 2025)

Theoretical framework

• Socio-ecological model for psychological development

Children develop through interacting systems of the household, neighbourhood, institutions, wider social network, and overarching cultural and political systems (Bronfenbrenner 1977)

• Social determinents of mental health and Stress Process Theory

Structural inequalities shape exposure to household stressors, impacting parenting behaviours and support (L. I. Pearlin et al. 1981; Nomaguchi and Milkie 2017; Alegría et al. 2018)

• Turning point theory

Fateful moment (Giddens 1991) and turning points (Rutter 1996) theories offer a theoretical underpinning unpacking change across the life course

Can be a single shock “event”, or a more gradual process of significant change.

Operationalising the critical event

Goodman’s (1997) Strengths and Difficulties Questionnaire (SDQ) is a 25 item psychometric screener.

Widely used validated (Bergström and Baviskar 2021; Vugteveen et al. 2021; R. Goodman 2001), although not without criticism (Kersten et al. 2016; J. White et al. 2013).

Measures of 4 dimensions of “difficulties”:

• emotional symptoms (e.g. “often seems worried”)
• peer problems (e.g. “rather solitary”)
• conduct issues (e.g. “fights or bullies”)
• hyperactivity/inattention (e.g. “easily distracted”)

And 1 dimension of “strengths”:

• pro-social behaviours (e.g. “shared with other children”)

Goodman’s SDQ measurement (details)

• 5 items per dimension, e.g. “Has at least one good friend”, “Sees tasks through to the end, good attention span”
• Sum responses from from 0 “Not true” to 2 “Certainly true” (ergo, ordinal)
• Greater values indicate greater difficulties (or strengths, for pro-social dimension). Some items are reverse-scored to achieve this.

Each dimension treated as a separate outcome

• Common to sum into “total difficulties” + strengths scales
• But different dimensions have divergent relationships with development and trends over time.
• Differential patterns in report bias, evident from inter-respondent discrepancies

How do critical events in childhood impact social and emotional development?

Easy to associate at one outcome point, e.g. \({Y} = \beta_0 + \beta_{x1} + \beta_{x2} +... \epsilon\), but this ignores baseline differences in \(Y\), difficult to temporally isolate cause-effect.

Option 1: Linear or quadratic growth model (GCM) (i.e. Burant (2016))

• Difficult to isolate timing of event and timing of gradient of change
• Requires development to either be linear, or follow another polynomial

Option 2: Person or group-centred approach (e.g. group-based trajectory model, Nagin et al. (2016))

• Requires qualitatively interpreting group shapes, high uncertainty
• Can be problematic when many “shapes” are flat lines

Option 3: Categorised change (e.g. move from “below threshold” to “above threshold”)

• Requires threshold effects (not clear for SDQ) and disregards dimensional quality of scale (A. Goodman and Goodman 2009)

Estimating dimensional change over time

• Outcome \(Y\) is the \(\Delta\) within a developmental time period
• Gradient is therefore based on a dimensional change in behaviour
• Evaluates the short-to-medium term associations

Goal: Estimate the effect of experiencing a critical event (\(X\)) on the change (\(\Delta\)) in a construct (\(Y\)) over a developmental time period.

We want to do this: \[\Delta_{Y} = Y_{t1} - Y_{t2}\] \[\Delta_{Y} = \beta_0 + \beta_{x1} + \beta_{x2} +... \epsilon \]

However, directly measuring the coefficient of \(\beta_{x}\) concerns estimation issues (see detailed discussion in Edwards (1994) and Laird and De Los Reyes (2013))

Two Y estimates with equal variance and covariance with an X var cannot produce a significant x –> \(y_{diff}\) relationship.

Same error for estimating \(X\) predictors of inter-respondent discrepancies

McArdle (2001; 2009) proposed estimating the change/difference as a latent factor in an SEM framework.

• \(Y\) is estimated from the common variance of indicator items + error \(\epsilon\)

• Fix intercept of \(Y_{t1}\) and \(Y_{t2}\) to 0, and the residual variance of \(Y_t{1}\) to 0

• Estimate \(Y_{t2}\) as the constrained \(Y_{t1}\) + \(\Delta\) by loading \(Y_{t2}\) and \(Y{t1}\) on \(\Delta\).

\[Y_{t2} = Y_{t1} + \Delta\]

\(\Delta\) gets its own mean and variance that are separated from the observed \(Y\) at both time periods and therefore not susceptible to the same issues as straight \(Y_{t2}-Y_{t1}\).

Measurement model

• The \(\Delta\) becomes a second order latent factor when the \(Y_{t1}\) and \(Y_{t2}\) are estimated as latent factors composed of their manifest questionnaire items

• Constraining measurement invariance between time points (item configuration, loadings and intercepts constrained) isolates the residual error. Strict scalar invariance for longitudinal parent-reported SDQ in GUS.

\[Y_{t1} = Y_{true} + \epsilon_{t1}\]

• Can claim \(\Delta\) composed entirely of the “true” difference between measurement occasions.

Source Vecchione and Zuffianò (2024)

• The \(\Delta\), as a second order latent variable, is estimated as another parameter with its own mean and variance.

• These can tell us about the properties of average difference in estimates between two time periods, inter-individual variability in this change (McArdle and Hamagami 2001; Vecchione and Zuffianò 2024)

• The resulting \(\Delta\) can be treated as a variable in the SEM and we can use this to study attributes and predictors of change.

Specifications

• Mplus8 via MplusAutomation in R v. 4.5.0, using Type=COMPLEX for sampling design adjustments.

• Ordinal parametrisation for manifest items

• Loadings derived from probability of endorsing a category, not a coefficient.

  • Robust Weighted Least Squared using diagonal weight matrix (WLSMV estimator). Robust to non-normality. Uses polychoric correlation for categorical items.

  • Theta parametrisation (frees residual variances of latent variables). Required for ordinal threshold estimation.

Reference works: Grimm, Ram, and Estabrook (2016), Crayen (2016), Xu et al. (2017)

Model permutations

Separated by:

• behavioural component (emotional difficulties, peer problems, hyperactivity/intention, conduct problems, pro-social strengths)

• Event grouping (1 through 7), developmental period (early or middle childhood).

• Result: 100+ model runs

Model C: Event

Without controlling for context or confounding relationships……

• Adversity predicts elevated acquisition of behavioural difficulties.

• Much more muddled effect for “positive” household changes

Relationships are largely but not entirely robust to covariate contribution

• Some instability around mean change estimate

• Participation in social groups predicts a greater-than-average accumulation of pro-social strengths in adolescence, but is not significantly associated with tempering difficulties.

• Changes in family composition predict elevated accumulation of emotional difficulties and externalising problems in early and middle childhood.

• Significant health problems in the family are associated with elevated accumulation of all difficulties.

• Encounters with crime/policing in middle childhood predict both elevated accumulation of difficulties and reduced accumulation of pro-social strengths.

• Moving home “without agency” predicted elevated conduct problems and emotional difficulties.

• Moving into relative poverty predicts elevated internalising problems

• Moving home appears to have a stronger impact on development in the early years (could be evidence of a timing effect)

• Other evidence for timing/developmental period effects are more muted

• Increased adversities have a clear and negative impact on social and emotional development, but the protective factor of “positive” changes is less evident at this level of population averages

• Expected to see a significant positive impact of moving out of relative poverty, but results are fairly muted.

Limitations

• only estimates short-to-medium term effects

• Many theories of childhood adversity and life-course effects hang on lasting/cumulative effects over longer period of time

• Some estimate instability, may come with the SEM territory

• Relies on parent-reports (which systemically differ from child self-reports)

• Relies on researcher assignment of the qualitatively positive/negative meaning of a critical event/turning point, which is crudge, generalising and imperfect.

• Difficult to construct “positive” changes (due to study design/interests of the time)

• Overall, while most events have a statistically significant impact on development, they contribute poorly to explaining the variance in behavioural change over time. short-term impact on social and emotional behaviours

• However, at a population level, this explains for very little of the between-individual variance in behavioural change over either developmental period

• Much more variance is explained by baseline covariates

Implications: Critical events “matter” at an individual level, and this impact can generally be detected at the population level, but are a less convincing explainer of the emergence of the social gradient of mental health