Synthetic Population Generation and Validation Proposal

Problem Statement

The Problem: We currently lack a real population dataset to test the synthetic population (synth-pop) method effectively.

Axioms

A synthetic population should be as close to a twin of the target population as possible.
The synthetic population generation method should be applicable to all populations.
We can create a precise artificial population using deterministic methods.

Hypothesis

The synthetic population generation method should be able to create a population that is a twin of an artificial population

Methodology

Deterministic Rules for Artificial Population Generation:

Create sample LOAs with a mixed makeup.
Take a subsection of the artificial population for survey purposes.
Conduct a census on the full artificial population.

Comparison Methods:

Compare the results between:
- IPF + Monte Carlo
- Just Synthetic Annealing
- GAN + Simulated Annealing (offer this to Daniel)

Proposed Steps

Step 1: Create a National Artifical Population

Decide on:

Size of the population
Granularity of the population

Create rules for population generation

Have population groupings
Have diferent proportions of grouping within an area
Add randomness

Expected Outcomes

A validated synthetic population generation method that can be applied to various populations.
Insights into the effectiveness of different comparison methods (IPF+ Monte Carlo with GAN + SA, just SA, and GAN+ R version of SA).
Identification of any selection biases in the survey and census data.

Conclusion

This proposal outlines a comprehensive approach to generating and validating a synthetic population using deterministic methods and various comparison techniques. The results will provide valuable insights into the accuracy and applicability of the synthetic population generation method.