causalfrag 0.2.1

Bug fixes

crosswalk() now computes the inference Robustness Value with the full definition used by sensemakr::robustness_value(): zero when f <= g, the ordinary formula applied to f - g when g < f <= 1/g, and the extreme robustness value (f^2 - g^2) / (1 + f^2) when f > 1/g, where f = |t| / sqrt(df) and g = t*_{df-1} / sqrt(df - 1). Version 0.2.0 used the ordinary formula in the last case as well, which overstated the inference RV for very large |t| relative to df (for example, 0.798 instead of 0.788 at t = 8, df = 10). Results in the ordinary branch, which covers typical applications, are unchanged. The new element rv_alpha_branch records which branch applies.

For a bare t statistic, crosswalk() no longer assumes n = df + 2. The RIR is reported as a percentage, and as a number of cases only when n is supplied. The lm method still uses nobs().

New tests compare the point and inference Robustness Values with sensemakr::robustness_value() over a grid of t statistics and degrees of freedom that covers all three branches, and check continuity at the switch between the ordinary and extreme branches.

crosswalk() reports the point and inference Robustness Values, the conditional ITCV and the RIR percentage computed from the focal |t| and residual degrees of freedom, states that their agreement is largely fixed by that shared input, and flags the boundary band in which ITCV and RIR are positive while the inference Robustness Value is still zero. Methods for lm fits, sens_results objects and a bare t statistic.
run_sensitivity() attaches the crosswalk to sens_results objects built from lm fits; print() and generate_report() display it.

The Causal Fragility Index (compute_cfi(), compute_cfi_weighted(), compute_cfi_benchmarked(), compare_cfi(), print_cfi()) is deprecated and will be removed in a future release. Under OLS, most of its components are functions of the same |t| and degrees of freedom, so the composite counted the same test information several times. flag_fragility() no longer computes it automatically.

sens_report() is renamed fragility_report() so that it no longer masks confoundvis::sens_report() when both packages are attached.

The konfound call in run_sensitivity() now passes nobs(model) and the number of model-matrix columns, so the residual degrees of freedom match the fitted model when factors or missing values are present.

Initial CRAN submission.
Introduces the Causal Fragility Index (CFI): a 0-100 composite score integrating sensitivity evidence across frameworks.
run_sensitivity() – unified dispatch to sensemakr, EValue, konfound.
flag_fragility() – five-level fragility classification, auto-computes CFI.
compute_cfi() / print_cfi() – CFI scoring and display.
interpret_sensitivity() – concise template-based narrative interpretation.
generate_report() – structured markdown/text report with CFI section.
visualize_sensitivity() – sensitivity plot integration.
sens_results S3 class for unified output across all functions.