Social determinants

It adds a model of the possible mechanism: which factors and pathways are compatible with the worse health observed and where an intervention could act, rather than just the finding that it exists. It applies when an association has already been established, for example that income, education or living environment are related to health, and you need to turn that starting point into something actionable. An association between a determinant and health does not tell you whether it acts directly or through intermediate behaviours, or whether its effect changes with context. Social determinants of health analysis models that chain: it separates the weight of each factor, distinguishes direct pathways from those that run through intermediate links and helps identify points in the chain where an intervention could have more room to act, given the model and its assumptions. It is the difference between knowing that a group is worse off and having a model of the chain that a policy can act on, which is what makes the evidence useful for a decision.

That is exactly the question that separates a compositional component from a contextual one, and it is addressed with multilevel models that distinguish the two. It applies whenever individuals are grouped within geographical units, such as neighbourhoods, municipalities or regions, and you want to know whether an area's worse health relates more to who lives there or to the area itself. A compositional component means that the place concentrates people with a particular risk profile; a contextual one means that there is variation associated with the context over and above the individual characteristics observed. Hierarchical models treat individuals as nested within their context and partition the variation between the individual and contextual levels, quantifying how much belongs to each. That variance partitioning helps distinguish how much of the observed pattern relates to individual characteristics and how much to context, and each leads to a different intervention: acting on individuals or acting on environments. Confusing the two levels leads to the wrong conclusions about where to put the effort.

Modelling pathways means estimating whether the observed relationships are compatible with direct and indirect pathways between a determinant and health: whether it acts directly or whether its relationship with health runs through intermediate factors, such as behaviours or material conditions. It applies when you want to know not only how strongly a factor is associated with health, but how that relationship might be transmitted. For example, education may be related to better health directly and also through habits or health literacy, and telling those pathways apart changes where it makes sense to intervene. The pathways are estimated with mediation analysis, which is the tool of our cross-sector service, causal mediation and moderation analysis: here it is applied to estimate direct and indirect components under the proposed model, while the method itself belongs to that service. The result points to the link in the chain where an intervention could break the transmission of risk, always read as pathways compatible with the data under the proposed model, not as a proven causal chain.

Yes: a conceptual model of determinants, the typical diagram chaining structural, intermediate and proximal factors, can be formalised and assessed empirically with structural equation models (SEM). It applies when there is a theoretical framework for how the determinants relate to each other and you want to move from the diagram to an empirical assessment. SEM estimate direct and indirect relationships between several determinants and health at the same time. They also incorporate latent variables, constructs that are not observed directly, such as socioeconomic status or health literacy, measured through their indicators. That turns a qualitative sketch into a model whose consistency with the data can be assessed and whose relationships and parameters can be estimated. The scope needs to be clear: SEM test whether the data are compatible with the proposed structure and estimate its parameters under the model's assumptions. They do not prove that this structure is the only one or the true one, and their causal reading depends on the design and the assumptions made.

For an analysis of social determinants of health, we need individual-level health data, information on the determinants you want to study, and a variable linking each person to their context (neighbourhood, municipality or region). Determinants may be individual, such as education, employment or income, and contextual, such as characteristics of the local area. The dataset does not need to arrive fully structured. Before modelling, we review how each determinant has been measured and define the hierarchical structure and the conceptual pathway model. Measurement is a sensitive point: a construct such as socioeconomic status or health literacy is rarely captured well by a single variable, and what the model can estimate depends on how it is measured. We also check whether there are enough geographical units and enough variation between them: if there are very few, or they are almost identical, the contextual part of the multilevel model will be estimated with great uncertainty or may be out of reach.

It adds the explanation: social determinants of health analysis models the possible mechanisms behind the gap that health inequalities analysis measures. Health inequalities and equity analysis quantifies how much inequality there is and between which groups, using indices such as the SII, the RII or the concentration index. Social determinants of health analysis models which factors and pathways could contribute to that gap: how structural, intermediate and proximal factors link up to the health outcome, using multilevel and structural equation models. The two do not overlap in method: this service does not redefine the inequality indices, it models the chain that could lie behind them. They follow a natural sequence: measuring the gap comes first, and modelling its possible mechanisms comes next. If the question is how much inequality exists and how it is changing, it is health inequalities analysis; if it is which factors and pathways could contribute to the gap and where to intervene in that chain, it is social determinants of health analysis.

Not with the strength of causal proof: a social determinants of health analysis of observational data estimates associations and pathways compatible with the data under a model, but does not demonstrate causality with the certainty of an experiment. This needs saying because mediation and structural equation models produce estimates of "direct and indirect effects", a name that invites a stronger causal reading than the design supports. The causal interpretation is limited to what the design can sustain. With observational data, the estimated components depend on assumptions that cannot be fully verified, such as no unmeasured confounding, or relationships that run in the assumed direction, and those assumptions are made explicit. What the service provides is a rigorous quantification of associations, pathways and effects estimated under a defensible model, with their uncertainty and limits stated, not a claim of causality that observational data cannot establish. That honesty is what makes the evidence usable to defend a policy without overselling it.

We deliver a model of the data-compatible pathways linking determinants and health under an explicit conceptual framework: the variance partitioning between the individual and contextual levels, the estimated direct and indirect components with their decomposition by pathway, and the assessment of the conceptual model when it is formalised with structural equations, all with the causal interpretation limited to what the design allows. It is for those who need to build the case for an intersectoral policy, whether in education, employment or urban planning, with evidence that a factor outside the health system is associated with health and an estimate of the size of that relationship under the model. The deliverable helps you argue to decision-makers where an intervention lever could sit: at which link in the chain and at which level, individual or contextual, an intervention would have most scope, with the assumptions and uncertainty made explicit. The client designs and implements the intersectoral policy; the service models the pathways, estimates the components and sets out what the evidence allows you to claim.