About This Project¶
Fertility rates in the United States and other developed countries have been declining for several decades. Understanding these trends and projecting future fertility patterns is crucial for policy planning in areas such as education, healthcare, and social security. This project uses Bayesian statistical methods to model historical fertility patterns and project future cohort fertility rates using data from the US Census Current Population Survey (CPS).
Motivation¶
Cohort fertility rates measure the average number of children born to women in a specific birth cohort over their reproductive lifetime. Unlike period fertility rates (which measure births in a given year), cohort fertility rates better capture long-term trends and can reveal generational shifts in reproductive behavior.
Recent trends suggest that cohorts born after 1980 may experience substantially lower lifetime fertility than previous generations. This has important implications for:
Demographic projections: Population growth and aging patterns
Economic planning: Labor force size and dependency ratios
Social policy: Healthcare, education, and retirement systems
Key Questions¶
This project addresses several key questions:
What are the long-term fertility trends for recent birth cohorts?
How low might fertility rates go for cohorts currently in their reproductive years?
Can we reliably project future fertility using historical patterns?
How well do model predictions validate against historical Census data?
Contents¶
Technical Report - Full analysis: Comprehensive documentation of data sources, methodology, Bayesian model specification, results, validation, and projections.
Methodology Overview¶
Our approach uses a Bayesian hierarchical log-linear model to capture age-specific fertility patterns across birth cohorts:
Data: US Census Current Population Survey (CPS) data from 1976-2024
Model structure: Log-linear model with cohort and age effects
Priors: Gaussian random walk priors to capture smooth temporal changes
Validation: Backtesting against historical Census data
Projection: Posterior predictive distributions for future cohort fertility
The model accounts for survey weights through resampling, ensuring that results reflect the actual US population rather than just the survey respondents.
Key Findings¶
Preliminary results suggest:
Declining fertility for cohorts born after 1980
Projected CFR could drop below 1.0 child per woman for cohorts born in the 2000s
Model validation shows good agreement with historical Census data, increasing confidence in projections
Uncertainty quantification provides credible intervals for all projections
For detailed findings, see the Technical Report.
Data Sources¶
IPUMS CPS: Historical CPS data (June 1976 - June 2022)
US Census Bureau: June 2024 CPS data
Census Historical Tables: Validation data for model checking
About the Model¶
The Bayesian approach provides several advantages:
Uncertainty quantification: All estimates include credible intervals
Flexibility: Can incorporate prior knowledge and handle missing data
Validation: Enables rigorous backtesting and out-of-sample prediction
Interpretability: Parameters have clear demographic interpretations
This project is a work in progress. Results are preliminary and subject to revision.