Spatial and spatio-temporal variability in social, emotional and behavioural development of children

Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2023
Thesis identifier
  • T16677
Person Identifier (Local)
  • 201992647
Qualification Level
Qualification Name
Department, School or Faculty
  • Neighbourhood differences in early development can be explored by incorporating spatial and spatio-temporal information with population data. Spatial refers to the relationship between neighbouring areas, while spatio-temporal refers to the relationship between neighbouring areas over time. At the time of writing, most population studies have focused on spatial variation in early development over a single year or short time period. This project identifies spatial and spatio-temporal referenced data that can be linked with population data on child social, emotional and behavioural difficulties in Glasgow, United Kingdom (UK). The Child Mental Health in Education (ChiME) study is a unique resource that can be used to model long term trends in a preschool population. In the ChiME study, Strengths and Difficulties Questionnaire (SDQ) forms were analysed for 35,171 children aged 4–5 years old across 180 preschools in Glasgow, UK, between 2010 and 2017 as part of routine monitoring. Using ChiME data, this work examined how early development varies over space and time, how the neighbourhood is defined, how important the neighbourhood is and what neighbourhood characteristics are related to early development. A literature review of 71 studies (from 2012-2022) in Chapter 2 discusses the neighbourhood constructs that are associated with variation in early development for children in Scotland. These constructs included the physical environment (e.g. greenspace) and social environment (e.g. social networks). The availability of data and the strength of evidence to support each construct varied. For many constructs, there was limited understanding of their relevance to younger children as opposed to adolescent or adult populations. There are gaps in the literature in the extent that neighbourhood constructs relate to developmental outcomes at an individual level or how this may change over time. To address these gaps, much more multilevel research, using population data is required. Chapter 3 provides a methodological review of the multilevel spatio-temporal approaches used to date. There is limited methodological guidance on how to model spatio-temporal variation for multilevel data. There is a risk of over complicating the model when attempting to account for spatial, temporal and/or spatio-temporal effects. Choosing the appropriate spatio-temporal multilevel model depends on the structure of the data, the degree of correlation, the goal of the analysis and overall model fit. Using a Bayesian workflow, each component of the model is reviewed in an iterative process to provide the best model for the data in Chapter 4. This includes evaluation of the outcome (total difficulties scores vs high scores) and comparing discrete distributions (Poisson, Negative Binomial and Zero-Inflated Negative Binomial models). Workflow analysis supported the use of Zero Inflated Negative Binomial distribution for total difficulties scores and the use of approximation methods for estimation. The total difficulties score for an individual child nested in their preschool, electoral ward and ward:year was modelled using a multilevel model with the components selected in Chapter 4. In Chapter 5, models were built incrementally, considering the value of each context. Boys, those of increasing deprivation and children outside the average age, had more difficulties on average. Preschool and ward variation, although minimal, highlight potential priority areas for local service provision. After consideration of demographics (sex, age, and deprivation), the overall spatial effect found the electoral wards of Anderston, Craigton, North East and Pollokshields were worse than expected (Relative Risk > 1) from 2010 to 2017. There were 72 preschools that were worse than expected based on their demographics. Approximately half of the children who lived in a ward that was worse than expected also attended a preschool that was worse than expected. There were independent spatio-temporal patterns in total difficulties, that exist in addition to the overall spatial effect. Spatial effects were not solely due to consistently poor performing areas. Instead, there is evidence of yearly variations in performance. Spatial analysis using only a single or few years may lead to misleading conclusions about area level variability. For example, once considering the spatio-temporal effect, Pollokshields was no longer considered worse than expected. There were differences in spatial and spatio-temporal variation depending on the neighbourhood definition (electoral ward, locality, Intermediate Zone (2001 and 2011) and Consistent Areas Through Time (CATTs)) found in Chapter 6. Looking at the different spatial scales together, can help support diffuse or more concentrated intervention delivery. Localities and 2011 Intermediate Zones had a similar spatial distribution to the ward. The relative importance of the neighbourhood compared to other contexts can be quantified through the Variance Partition Coefficient (VPC). Estimated VPC of the neighbourhood on early development was expected to be between 0 and 9% according to recent literature. Though the typical VPC equation does not apply to discrete distributions, recent approximations have been developed. Using these approximations, it was found that proportionally, the neighbourhood context alone does not make a considerable contribution to variation in difficulties scores. VPC values ranged from 0.39-1.1% depending on the neighbourhood definition. From the perspective of decision-making, the partitioned variance suggests that considering the neighbourhood along with other contexts would be more meaningful than the neighbourhood alone. Preschool and neighbourhood characteristics are thought to provide a more feasible target for intervention compared to individual level characteristics. Cross-level effects (which describe the association between a higher level covariate and a lower level outcome) are investigated in Chapter 7. Preschool and neighbourhood indicators were derived from openly available administrative data. The quality of these indicators and their relevance to this project varied. Preschools were classified as small/medium/large local authority, private business or voluntary. Most children were in local authority preschools. Total difficulties scores were lower in private business compared to small local authority preschools. Spatial variation was in part explained by a child’s prosocial behaviour and its interaction with their preschool provider. The mechanisms underlying these differences are at present unknown. There were ecological correlations between total difficulties and the neighbourhood indicators (participation, child poverty, domestic abuse, free time places, vandalism and proximity to greenspace (at 400 m and 800m)). These correlations did not translate to a cross-level association with individual level total difficulties. In conclusion, there are multiple contexts that account for variation in total difficulties. The preschool and spatio-temporal context and their composition could provide additional information about how the neighbourhood relates to early development. There is a need for more spatio-temporal data, that can be linked to population data, to understand how the neighbourhood is associated with development at an individual level, beyond deprivation. Multi-level spatio-temporal models can be used to understand early development and support the selection of delivery areas for place-based interventions.
Advisor / supervisor
  • Thompson, Lucy
  • Barry, Sarah J. E.
Resource Type