Plot longitudinal causal decomposition

Usage

# S3 method for class 'ineqx_causal_longit'
plot(
  x,
  type = "decomp",
  ci = FALSE,
  style = "line",
  stats = NULL,
  time = NULL,
  trim = 0.995,
  share = FALSE,
  ...
)

Arguments

x

An ineqx_causal_longit object (including Shapley-averaged results)

type

Character, the plot type:

"decomp": Decomposition of changes relative to reference. Use stats to select components (default: 3 aggregate + total).
"wibe": Cross-sectional treatment effects on within/between inequality at each time (levels, not changes).
"treat": Predicted treatment effect distributions. When time is specified, shows per-group distributions. When time = NULL, shows pi-weighted marginals across all times.
"treat.params": Treatment effect parameters (beta/lambda) over time (line chart).
"effect.prop": Proportional treatment effect on the mean by group over time: beta_g / mu0_g (identity fit) or exp(beta_g) - 1 (log fit). Overlapping lines indicate a proportional (group-invariant) effect.
"outcome": Predicted outcome distributions (control vs treated). When time is specified, shows per-group distributions. When time = NULL, shows pi-weighted marginals.
"outcome.params": Predicted means and SDs under control vs treatment over time (line chart). For DiD models, the "Control" line represents the DiD-implied counterfactual untreated level for the treated post-period subpopulation, so the gap between the lines equals the DiD ATT.
"pretrends": Pre-period predicted Treated and Control levels over time. Available only for DiD models. Under parallel trends the two lines should evolve with the same slope; their (constant) vertical gap reflects pre-existing selection. Diverging slopes indicate a violation of parallel trends.
"shapley": Shapley averages with ordering ranges (only when order = "shapley"). Use stats to select components.

ci

Whether to show confidence intervals. Accepts FALSE or "none" (no CIs), TRUE or "delta" (use stored SEs), "boot" (bootstrap with defaults), or boot_config() (bootstrap with custom settings). Default FALSE.

style

Character: "line" (default) for connected lines with ribbon CIs, or "point" for points with error bar CIs.

stats

Character vector of components to display. Meaning depends on type:

For "decomp" and "shapley":: Any combination of aggregate components ("behavioral", "compositional", "pretreatment", "total") and/or individual deltas ("delta_beta", "delta_lambda", "delta_pi", "delta_pi_b", "delta_pi_w", "delta_mu", "delta_sigma"). Default for "decomp": c("behavioral", "compositional", "pretreatment", "total"). Default for "shapley": c("behavioral", "compositional", "pretreatment").
For "wibe":: Components to display, default c("tau", "tau_b", "tau_w"). Canonical options: "tau" (total), "tau_b", "tau_w", "het_b", "cov_b", "rescale_b", "het_w", "cov_w", "rescale_w". Legend labels "cov_*" as “Sorting” and "rescale_*" as “Rescaling” (rescaling values are zero under ystat = "Var"). Shorthands: "het_cov" = the four het/cov sub-components (no rescaling), "het_cov_b" = c("het_b","cov_b"), "het_cov_w" = c("het_w","cov_w"), "subs" = all six sub-components, "subs_b" = c("het_b","cov_b","rescale_b"), "subs_w" = c("het_w","cov_w","rescale_w").

time

Numeric, time point for type = "treat" or type = "outcome". If NULL (default), pi-weighted marginal distributions are shown across all time points.

trim

Numeric between 0 and 1. For distribution plots ("treat", "outcome"): quantile at which to trim tails. Default 0.995.

share

Logical. For type = "decomp", plot each component as a share of the total effect at that time (%) instead of in absolute units. Periods where the total is \(\approx 0\) are blanked, since shares are undefined there. Default FALSE.

...

Additional arguments (currently unused)

Value

A ggplot2 object