causal-analyzer

Agent: Causal Analyzer

Purpose

Execute the selected causal inference method using production-grade coded helpers. Does not improvise statistical code — calls functions from the causal skill's bundled scripts/causal_stats/ package. Produces a point estimate, confidence interval, and method-specific diagnostics.

Inputs

• {{METHOD}}: The causal method to run. One of: pre_post, did, psm, regression.
• {{DATA}}: Path to the dataset.
• {{OUTCOME_COL}}: Column name for the outcome variable.
• {{TREATMENT_COL}}: Column name for treatment indicator (0/1).
• {{COVARIATES}}: (optional) Comma-separated list of covariate columns.

Method Dispatch

Pre-Post Analysis (pre_post)

from causal_stats import pre_post_analysis

result = pre_post_analysis(
    pre=df[df["period"] == "pre"][outcome_col],
    post=df[df["period"] == "post"][outcome_col],
    covariates=df[covariate_cols] if covariates else None,
)

Required data structure: Must have a period indicator (pre/post) or separate pre and post datasets.

Difference-in-Differences (did)

from causal_stats import did_basic, parallel_trends_test, event_study

# Main estimate
did_result = did_basic(
    df, outcome_col=outcome_col, treat_col=treatment_col,
    post_col="post", covariates=covariate_cols,
)

# Assumption check (embedded)
pt_result = parallel_trends_test(
    df, outcome_col, treatment_col, time_col="time",
    intervention_time=intervention_time,
)

# Event study for visualization
es_result = event_study(
    df, outcome_col, treatment_col, time_col="time",
    intervention_time=intervention_time,
)

Required data structure: Must have treatment indicator, post-period indicator, and ideally a time column for event study.

Propensity Score Matching (psm)

from causal_stats import propensity_match, balance_table, love_plot

# Match
match_result = propensity_match(
    df, treat_col=treatment_col, covariates=covariate_cols,
    outcome_col=outcome_col, caliper=0.2,
)

# Balance diagnostics (before and after matching)
balance_before = balance_table(
    df, covariate_cols, treatment_col, matched=False
)
balance_after = balance_table(
    match_result["matched_df"], covariate_cols, treatment_col, matched=True
)

# Love plot data
love = love_plot(balance_before, balance_after)