Analysis of the funding of social services from a spatial approach in Andalusia (Spain)

The Andalusian Social Service System guarantees social protection through Community Social Service Centres (CSSCs), which function as gatekeepers to specialized services and are managed by local and supra-local entities. Funding distribution across local entities is regulated, primarily by population size (90%), with additional weights for dependent population (2%), surface area (2%), scattered population (3%), large cities (3%), and a poverty index adjustment (33.34%) based on GDP-population ratios. The poverty index is limited, focusing solely on income and ignoring broader dimensions of deprivation (education, health, employment, assistance). Given Andalusia’s heterogeneous demographic, geographic, and deprivation profiles and the known spatial spillovers in public expenditure, the distribution of CSSC funding is unlikely to be spatially homogeneous. This study aims to analyze the spatial distribution of CSSC funding per capita by catchment area in 2019 and to explore associations with demographic and socioeconomic indicators using spatial statistics, thereby informing resource allocation and social policy.

Spatial data analysis has a long tradition in fields such as epidemiology, criminology, and economics, but is less common in public service financing. Prior applications include analyses of school funding in China, disability expenditures in Italy, social expenditures across EU countries, welfare and housing benefits in the UK, and various Spanish studies on public expenditure, local social services spending, and entrepreneurship promotion. In Andalusia, earlier work developed a multidimensional deprivation index (education, employment, income, housing, infrastructure, health) showing higher deprivation in rural mountain areas versus the more urbanized Guadalquivir valley and coastal zones. The current study builds on this literature by applying exploratory spatial data analysis and spatial econometrics to CSSC funding and by relating funding to contextual indicators, addressing potential spatial spillovers.

Study setting and units: Andalusia (southern Spain), a relatively disadvantaged region (2020 per capita income €17,747; unemployment 22.74%). CSSCs (n≈250) operate in Social Service Zones (SSZs) that can be intracity districts, single large municipalities, or groups of small municipalities. The 2019 analysis included 184 spatial units after aggregating intramunicipal districts to the municipal level where needed and excluding one SSZ with missing funding data. Data sources: Funding by SSZ (non-public; obtained from the Andalusian Equality, Social Policies and Reconciliation Department for 2019). Demographic and socioeconomic indicators at the municipal level from the Institute of Statistics and Cartography of Andalusia (2019); smallest municipalities aggregated to their SSZ. Variables: Dependent variable: CSSC funding per capita (€) by SSZ. Explanatory variables: sex ratio (women per 100 men), ageing index (older people per 100 children), dependency index (children+older per 100 adults), emigration rate (% emigrants over inhabitants), immigration rate (% nonnationals), unemployment rate (% unemployed aged 18–64), population density (inhabitants/km²), and employment in primary sector (% employees in primary sector aged 18–64). Transformations: Due to dispersion, population density and the dependent variable were log-transformed to meet linear model assumptions and improve normality of residuals. Spatial analysis: Global Moran’s I assessed overall spatial autocorrelation in funding. Local Indicators of Spatial Association (LISA) used: Local Moran’s I (clusters and outliers) and Local Getis–Ord G* (hot/cold spots). Spatial weights: second-order queen contiguity. Inference via permutation for Moran’s I. Modeling: Began with an OLS model including all variables (Model 1) to assess directions and significance, followed by a refined OLS (Model 2) retaining significant predictors and reducing collinearity. Diagnostics included normality (Jarque–Bera), heteroskedasticity (Breusch–Pagan), multicollinearity (Condition Index), and spatial dependence tests (Global Moran’s I of residuals; Lagrange Multipliers: LM-Lag, robust LM-Lag, LM-Error, robust LM-Error). Based on Anselin’s decision rule, spatial lag regressions were estimated corresponding to Model 1 (Model 3) and Model 2 (Model 4). Model fit compared via AIC and Schwarz (BIC) criteria. All analyses conducted in GeoDa 1.22.

Spatial patterning: CSSC funding per capita exhibited significant positive spatial autocorrelation: Global Moran’s I = 0.261 (pseudo p<0.01). LISA identified hot spots in northern Huelva, central Córdoba, northwestern Almería, and northern Granada; cold spots appeared around major cities (Seville, Málaga, Granada, Cádiz). Regression results (see Table 2): - Model 2 (refined OLS): Significant predictors were ageing index (+), % employees in primary sector (+), % foreign population (–), and log(population density) (–). Adjusted R² ≈ 0.58; improved collinearity (Condition Index 15.08). Residuals showed spatial autocorrelation, prompting spatial regression. - Spatial dependence tests recommended spatial lag specification. Spatial lag models (Models 3 and 4) retained the same significant variables as their OLS counterparts. Spatial autoregressive parameter (ρ) was significant (Model 4: 0.3061***), and residual spatial autocorrelation became non-significant. - Model fit: Spatial lag models had lower (better) information criteria than OLS (e.g., Model 4 AIC −275.837; Schwarz −256.547) indicating improved fit. Effect sizes (Model 4): Interpreting coefficients on log funding per capita: each one-unit increase in the ageing index and in % primary-sector employees was associated with approximately +0.14% and +0.05% higher funding per capita, respectively; each one-unit increase in % foreign population and in population density was associated with about −0.31% and −0.06% lower funding per capita, respectively. Non-significant variables (in full models): unemployment rate, emigration rate, dependency index, and sex ratio (directions: unemployment, emigration, dependency positive; sex ratio negative) but not statistically significant. Descriptive statistics (n=184): mean funding €120.36 per capita (SD 49.57); mean ageing index 119.55; mean % primary sector employees 44.79; mean % foreign population 5.6; mean population density 568.4/km² (highly skewed).

Findings confirm that CSSC funding is spatially clustered, with hot spots concentrated in rural and generally more deprived mountainous areas (Sierra Morena and Subbaetic systems) and cold spots around major urban centers, despite the funding formula’s strong emphasis on population size. The lack of hot spots in large cities may reflect aggregation to the municipal level masking within-city disparities; some of Spain’s lowest-income neighborhoods occur in Seville and Málaga but did not appear as hot spots at this scale. Spatial lag results and significant spatial autoregression suggest spillover effects in funding across neighboring SSZs, consistent with prior evidence on spatial dependence in public spending. Importantly, not all regulatory criteria align with observed associations: while ageing (linked to dependency) and density (proxy for population size) were significant, the dependency index itself was not. Two additional factors emerged: higher employment in the primary sector (rurality) associated with higher funding, and higher shares of foreign population associated with lower funding. Potential mechanisms include lower utilization among affluent foreign residents in coastal/urban areas and limited access among irregular migrants; meanwhile, rurality proxies service needs and delivery costs. The spatial distribution of funding clusters broadly aligns with prior multidimensional deprivation patterns in Andalusia, particularly unemployment, income, and housing domains, suggesting some suitability of current allocations. Nonetheless, identified cold spots in disadvantaged areas (e.g., parts of Cádiz) indicate places warranting further assessment and potentially adjusted allocations or tailored programs. Overall, spatial analytics provide actionable insights to support evidence-informed social services planning and evaluation.

The study demonstrates non-random, spatially clustered distributions of CSSC funding per capita across Andalusia, with rural hot spots and urban cold spots. Spatial regression identifies ageing and primary-sector employment as positive associates of funding, and foreign population share and population density as negative associates; the dependency index, despite its role in the allocation formula, was not significant. These insights can guide assessments of funding adequacy against social need and inform adjustments to allocation criteria. The work illustrates the utility of spatial methods for social services financing and is suitable for replication in other regions. Future research should include longitudinal analyses to examine temporal stability and impacts, incorporate submunicipal data to capture intra-urban disparities, and extend models with planning and fiscal variables (e.g., professional ratios, beneficiary rates, tax revenues, transfers) as data become available.

- Data limited to a single year (2019); inability to assess temporal stability or causal impacts. Economic context has since shifted (COVID-19 pandemic, Ukraine war), though recent policies avoided sharp social spending cuts. - Lack of submunicipal sociodemographic data precluded analysis within large cities, potentially masking intra-urban disparities. - Unavailability of certain financial data (e.g., for municipalities <50,000 inhabitants) prevented inclusion of fiscal covariates such as tax revenues and current transfers. - Potential omitted planning-related variables (e.g., professional ratios, beneficiary rates) not included here could improve explanatory power. - One SSZ lacked funding data and was excluded; smallest SSZs (district-level) were aggregated to municipalities, possibly inducing measurement error in contextual variables.