Methods for estimating population density in data-limited areas: evaluating regression and tree-based models in Peru.

Area of Science:

• Demography
• Statistical Modeling
• Geographic Information Systems

Background:

• Accurate small area population estimates are vital for policy and health planning, especially in data-limited countries.
• Traditional methods often struggle with limited direct sampling data for small areas.
• Existing small area estimation models rely on temporal or spatial information transfer.

Purpose of the Study:

• To evaluate the effectiveness of model-based methods for population estimation in areas lacking direct sample data.
• To compare the performance of non-parametric tree-based models against conventional regression techniques for population density estimation.

Main Methods:

• Focus on model-based approaches for population estimation without direct samples.
• Comparison of six tree-based model structures, including Random Forest and Bayesian Additive Regression Trees.
• Evaluation of predictive accuracy in small, non-sampled areas.

Main Results:

• Non-parametric tree-based models demonstrated higher prediction accuracy than conventional regression methods.
• This improved accuracy was observed specifically when data from prior time periods were not utilized.
• Random Forest and Bayesian Additive Regression Trees showed particular efficacy.

Conclusions:

• Tree-based models are effective for improving population density estimates in non-sampled areas.
• These methods are crucial for regions with incomplete census data, enhancing policy and planning.
• Findings have significant implications for economic, health, and development policies in data-scarce environments.