ECOLOGICAL DATA STUDY · ENGLAND
COVID-19
Mortality &
Social Inequality
A statistical investigation of how deprivation, health, housing, and age structure relate to COVID-19 mortality across English Local Authorities.
RtidyversepsychPCARegressionrandomForest
RESEARCH OVERVIEW
Do structural inequalities shape local mortality?
The study combined COVID-19 deaths, population estimates, and 2021 Census variables from ONS and NOMIS. Rates were standardised per 1,000 residents or households so authorities with different population sizes could be compared fairly.
RQ
Research questionHow do health status, household deprivation, housing conditions, and age composition relate to COVID-19 mortality across England?
01 · DATASET
Multiple public sources, one comparable dataset.
Data preparation
Designed for fair area-level comparison.
LA_code linked the source tables. Identification fields were excluded from statistical analysis, while raw counts were converted to standardised rates.02 · EXPLORATORY ANALYSIS
Mortality followed a socially uneven pattern.
Mortality was non-normal
W .958Shapiro–Wilk testing returned p < .001, supporting rank-based and non-parametric analysis.
Deprivation led
ρ .421Two-dimension household deprivation had the strongest positive bivariate relationship.
Bad health mattered
ρ .392Poor baseline health showed a strong positive association with mortality.
Age alone was weak
ρ .03The 80+ share had only a small, non-significant direct relationship with area mortality.
SPEARMAN ASSOCIATIONS
Structural disadvantage showed the clearest signal.
42.05%Dep 2
39.15%Bad health
35.58%Health*
23.66%Dep 4*
8%Age 60
3%Age 80
Partial correlations
Bad health | severe deprivationρ = .3558
Severe deprivation | age 80+ρ = .2366
Both testsp < .001
Association tests
Deprivation × mortality groupsp < .001
High vs low deprivationp = 1.6 × 10⁻⁷
Age 60–79 vs age 80+p < .001
03 · METHODOLOGY
A complete, reproducible statistical workflow.
01Merge & standardiseONS, NOMIS, mortality, and population data
→02Explore & testHistograms, Q–Q plots, Shapiro and KS
→03AssociateSpearman, partial, Chi-square, Fisher
→04Reduce dimensionsKMO, scree plot, varimax PCA
→05Model & diagnoseRegression, BIC selection, Random Forest
Multicollinearity controlled
Age, deprivation, health, and dwelling categories form balances. Correlation matrices, VIF diagnostics, variable omission, and PCA reduced unstable overlap.
KMO improved to 0.906
The initial KMO was 0.51. Removing two weak, redundant age variables produced excellent revised sampling adequacy.
Non-parametric validation
Spearman, Fisher, Wilcoxon, and Mann–Whitney tests complemented regression where normality or expected-count assumptions were unsuitable.
04 · MODEL PERFORMANCE
Model fit, stability, and interpretation.
ModelR²Adjusted R²PredictorsStrengthLimitationDecisionRole
BIC Stepwise BEST FIT.2529.24264ParsimoniousCompositional effectsRetainedPrimary
PCA Regression.1626.15104 RCsStable factorsLower fitSupportedInterpretation
Random Forest.186*—All numericNon-linear benchmarkOverfit riskBenchmarkImportance
Equivalent reduced model
ANOVA comparison with the full model was not significant (p = 0.1141), indicating no meaningful loss of fit.
Four components explained 87%
RC1 captured deprivation and poor health; RC2 age structure; RC3 shared housing; and RC4 the younger population profile.
500-tree benchmark
Age bands, deprivation, and poor health ranked highly, but the large in-sample R² suggested potential overfitting.
05 · INTERPRETATION
Deprivation and health formed the central dimension.
HH deprivation · 295
Health · bad89
HH deprivation · 385
Health · fair76
HH deprivation · 175
Health · very good74
Age · 40–5965
KEY FINDING · RC1 LOADINGS
Social conditions explained more than age alone.
Deprivation and poor health remained the most consistent area-level predictors across rank correlations, partial correlations, direct regression, PCA regression, and Random Forest importance.
Age increases biological vulnerability, but the local mortality pattern was more strongly structured by existing social and health inequality.
06 · CONCLUSION
COVID-19 exposed and amplified existing inequalities.
Local Authorities with greater household deprivation and poorer baseline health experienced higher mortality rates. The reduced regression model explained approximately one quarter of area-level variation without a significant loss of fit.
This ecological study identifies area-level associations rather than individual causation. Even so, consistent evidence across several methods supports policies that address deprivation, population health, and unequal vulnerability before future public-health emergencies.