Production release

TSCSMethods.jl v2.0.0

TSCSMethods.jl v2.0.0, is a major milestone that transforms this package from an experimental implementation into a production-ready statistical software package with validation, testing, and better code organization.

Statistical Validation & Reliability

See documentation for details on unit testing.

Complete Code Reorganization

The entire codebase has been restructured into logical subsystems for improved maintainability:

src/
├── core/          # Core types and model construction (2 files)
├── matching/      # Matching algorithms and utilities (9 files)  
├── balancing/     # Balancing strategies and auto-balancing (7 files)
├── estimation/    # Treatment effect estimation and bootstrap (9 files)
├── advanced/      # Stratification, refinement, advanced features (3 files)
└── utilities/     # Helper functions, storage, inspection tools (7 files)

Testing Infrastructure

• Unit tests organized by subsystem in test/unit/
• Integration tests for complete workflows in test/integration/
• Statistical correctness tests in test/correctness/
• Performance benchmarks in test/benchmark/
• Automated validation gates for coverage and placebo testing

Complete Documentation

• Full API documentation with examples
• Statistical methodology explanations
• Tutorial notebooks and working examples
• Validation procedures and results
• Developer guidelines and contribution docs

Enhanced Statistical Features

Robust Matching System

• Mahalanobis distance matching with missing data handling
• Caliper constraints and refinement procedures
• Automated balancing with p-value optimization
• Support for time-varying and time-invariant covariates

Advanced Estimation

• Bootstrap-based inference with configurable iterations
• Stratified estimation for heterogeneous effects
• Multiple outcome support with simultaneous estimation
• Comprehensive result reporting with confidence intervals

Quality Assurance

• Input validation with informative error messages
• Type safety throughout the codebase
• Extensive parameter validation and bounds checking
• Memory-efficient algorithms for large datasets

Validation Results

This release includes comprehensive statistical validation, see documents.

(Breaking) Changes from v1.6

New Organized Structure

• File paths updated (old hardcoded paths will break)
• Test files reorganized by category
• Examples moved from vignette/ to examples/
• Validation scripts moved to validation/

Enhanced API Stability

• All exported functions remain the same
• Improved error messages and input validation
• More robust handling of edge cases
• Better memory management for large datasets

Dependency Updates

• Updated to support Julia 1.10+ with testing on 1.6, 1.10, 1.11 (earlier versions may work)

Quick Start

using TSCSMethods

# Load example data
data = example_data()

# Create model for causal inference
model = makemodel(
    data, :day, :fips, :gub, :death_rte,
    [:pop_dens], Dict(:pop_dens => false), 
    5:10,    # F: post-treatment periods  
    -15:-10  # L: pre-treatment periods
)

# Complete workflow
match!(model, data)      # Find matched control units
balance!(model, data)    # Balance covariates  
estimate!(model, data, dobayesfactor=false)  # Estimate treatment effects

# Results
println("ATT: ", model.overall.ATT)
println("95% CI: [", model.overall.p05, ", ", model.overall.p95, "]")

For Developers

New Development Commands

# Run all tests
julia --project=. -e "using Pkg; Pkg.test()"

# Run validation gates
julia --project=. validation/sim_bias_coverage.jl --seeds 20 --iterations 400
julia --project=. validation/placebo_permutation.jl --permutations 300 --iterations 100

# Build documentation
julia --project=docs/ docs/make.jl

Contributing

The new modular structure makes contributions easier:

• Clear separation of concerns across subsystems
• Comprehensive test coverage for new features
• Automated validation prevents regressions
• Detailed developer documentation

Performance & Reliability

• Memory optimized for large time-series cross-sectional datasets
• Numerically stable algorithms with extensive edge case handling
• Comprehensive error reporting with actionable error messages
• Statistical correctness verified through multiple validation frameworks

Documented version

This version extensively documents the prior version.

better version

This is the last version before the implementation of full-on handling of missing values

initial release

This is the initial full, publication-ready, version of the package.

Full Changelog: v0.2.0...v1.0.0

full sliding-window version

Full Changelog: v0.1.0...v0.2.0

v0.1.0

This is the initial version of the package, without sliding windows, used for the Aug. 2021 paper draft.