Software

dagmv R · 0.1.0

Causally disciplined multiverse analysis. A dependency-free core parses candidate DAGs (dagitty-compatible syntax), tests d-separation, applies the generalized adjustment criterion, and classifies every candidate control as required, forbidden, or optional under each candidate graph. mv_run() fits each specification once and maps validity per graph; mv_decompose() returns per-graph robustness metrics and the within/between-graph variance decomposition with its structural share. Engines: linear models, negative binomial, fixed-effects panels.

github.com/sokubo/dagmv — install with remotes::install_github("sokubo/dagmv"). Bug reports and feature requests are welcome on the issue tracker.

panelcond R · 0.1.1

Refreshment-sample designs for panel conditioning. Estimates the causal effect of prior interviews on a wave-specific survey answer from a panel that has added a refreshment sample: the naive fresh-versus-continuing contrast, survival matching, its symmetric variant, the entry-wave correction (which uses the continuing cohort's own unconditioned first-wave answers to measure its attrition selection), attrition inverse-probability weighting, and the incremental contrast between two cohort-matched panels for doses longer than the available follow-up. Also returns the decomposition of the naive contrast into conditioning and attrition components, the two diagnostic tests that distinguish state-dependent from non-stationary attrition, worst-case bounds and breakdown values, and a simulation engine for design planning. Dependency-free.

github.com/sokubo/panelcond — install with remotes::install_github("sokubo/panelcond"). The accompanying paper is in preparation; a preprint will be linked here when posted.

Replication archives

Code and public data for each paper are released on GitHub when the paper is posted as a preprint. Analyses of licensed microdata (the Japanese Life Course Panel Surveys, government statistics) are released as code together with synthetic data that reproduces the variable structure but not the values.