|
Abstract
|
Background
Data can be missing when a survey fails to collect information from certain regions due to feasibility issues, which can impose problems while performing spatial analysis.
Objective
The present study aims to estimate missing aggregate level public health spatial data by utilizing the information from neighbouring regions and accounting for spatial autocorrelation inherently present in the data.
Methodology
Data was simulated for fixed values of various parameters in spatial regression models under low and high autocorrelation scenarios in dependent and independent variables. In dependent variable, 5%-25% of values were assumed to be missing. Stochastic regression imputation using spatial regression models namely spatial lag model, spatial error model, spatial Durbin model, spatial Durbin error model and spatial lag of X model was performed. The performance of these models were also compared using data from Annual Health Survey 2012-13.
Results
The simulation analysis revealed that for any amount of missing values in the data, irrespective of whether the other variables in the regression model are spatially autocorrelated or not, if autocorrelation in the variable with missing values is high, stochastic regression imputation performed using spatial lag model, spatial Durbin model and spatial Durbin error model gives accurate estimates of missing values. If the autocorrelation is low, in addition to these three models, spatial lag X model was also found to be effective in estimating the missing values.
Conclusion
The proposed mechanism results in optimal imputation of missing values in spatial data, which can yield complete data useful for public health professionals for effective interventions.
Language: en |