Unmixing aggregate data: estimating the social composition of enumeration districts

Area of Science:

• Data Science
• Computational Social Science
• Geographic Information Systems

Background:

• Interpreting and classifying aggregate data sources is challenging due to inherent limitations.
• Standard classification tools in image processing and census analysis are often hindered by data aggregation.
• Aggregate data masks underlying subunit variations, complicating accurate analysis.

Purpose of the Study:

• To introduce and adapt an 'unmixing' approach for analyzing aggregate data.
• To reveal subunit variations masked by aggregation in census data.
• To apply this technique to Southampton's small area statistics for social composition analysis.

Main Methods:

• The study adapts an 'unmixing' technique, previously successful in Earth Observation.
• An artificial neural network is utilized to perform the unmixing of aggregate data.
• The method is applied to Census small area statistics for Southampton, UK.

Main Results:

• The adapted unmixing technique successfully revealed variations in the social composition of Southampton's enumeration districts.
• The artificial neural network effectively processed aggregate census data to infer subunit characteristics.
• The study demonstrates the utility of Earth Observation unmixing methods in social science applications.

Conclusions:

• The 'unmixing' approach is effective for interpreting and classifying aggregate census data.
• Artificial neural networks can overcome limitations posed by aggregated data in social science.
• This method offers a novel way to understand the social composition of geographic areas.