Spatial Microsimulation — Telework Estimates
Census Block Group telework estimates via IPF reweighting across ACS, HPS, and CPS surveys
Rutgers RUCI Lab
Problem
Problem
National surveys (American Community Survey, Household Pulse Survey, Current Population Survey) collect telework data at too coarse a geographic resolution to be useful for local planning. Census Block Groups — the finest standard geographic unit — are not directly sampled at scale. Spatial microsimulation via Iterative Proportional Fitting (IPF) reweights survey microdata to match Block Group marginal distributions, generating synthetic small-area estimates.
Approach
Approach
The project implements IPF reweighting: survey microdata from ACS, HPS, and CPS are reweighted so that their joint distribution matches the known marginal distributions at the Census Block Group level. The original R-based IPF implementation was reimplemented in vectorized Python (NumPy matrix operations), restructuring the convergence loop for a 6× computational speedup. Moran's I spatial autocorrelation tests validate that the synthetic estimates exhibit the expected spatial clustering patterns.
Architecture
Architecture
Spatial Microsimulation — Telework Estimates — system diagram
Key Technical Decisions
Key Technical Decisions
Vectorized Python over R for IPF
The original R implementation iterated row-by-row through the contingency table. Reimplementing as NumPy matrix operations — restructuring the convergence loop to operate on the full matrix per iteration — achieved 6× speedup on the same hardware with identical numerical results.
Results
Results
- ✓Census Block Group telework estimates synthesized across ACS, HPS, and CPS surveys
- ✓6× computational speedup from R-to-vectorized-Python IPF reimplementation
- ✓Moran's I validation confirms spatial clustering patterns are preserved
- ✓Research conducted at Rutgers Urban and Civic Informatics Laboratory
Tech Stack
Tech Stack
Links
Links