Migration from R CausalImpact

This guide helps R CausalImpact users transition to the Python implementation.

Installation

pip install bsts-causalimpact

API Mapping

R

library(CausalImpact)

impact <- CausalImpact(data, pre.period, post.period,
                       model.args = list(niter = 1000))
summary(impact)
summary(impact, "report")
plot(impact)

Python

from causal_impact import CausalImpact

ci = CausalImpact(data, pre_period, post_period,
                  model_args={"niter": 1000})
print(ci.summary())
print(ci.report())
fig = ci.plot()

Parameter Mapping

R (model.args)	Python (model_args / ModelOptions)	Default
`niter`	`niter`	1000
`nseasons`	`nseasons`	`None`
`season.duration`	`season_duration` or `model_args["season.duration"]`	`1` when `nseasons` is set
`prior.level.sd`	`prior_level_sd`	0.01
`standardize.data`	`standardize_data`	True
`expected.model.size`	`expected_model_size`	2

Data Format

R

data <- zoo(cbind(y, x1, x2), dates)
pre.period <- c(as.Date("2020-01-01"), as.Date("2020-03-14"))
post.period <- c(as.Date("2020-03-15"), as.Date("2020-04-14"))

Python

data = pd.DataFrame({"y": y, "x1": x1, "x2": x2}, index=dates)
pre_period = ["2020-01-01", "2020-03-14"]
post_period = ["2020-03-15", "2020-04-14"]

Key differences:

Use pandas DataFrame instead of zoo object
Periods are string lists instead of Date vectors
Integer indices are also supported

Output Mapping

R	Python	Notes
`summary(impact)`	`ci.summary()`	Same tabular format
`summary(impact, "report")`	`ci.report()`	Same narrative output
`plot(impact)`	`ci.plot()`	Returns matplotlib Figure
`impact$summary`	`ci.summary_stats`	Dict with aggregate stats
`impact$series`	`ci.inferences`	DataFrame with per-timestep data

Numerical Equivalence

This library verifies ±3% agreement with R CausalImpact on point estimates and cumulative effects across multiple test scenarios. Tests run on every PR.

The seasonal model uses the same state-space algorithm as R bsts AddSeasonal(), achieving ±1% CI parity on the seasonal fixture.

Differences arise from independent RNG implementations (R's set.seed vs Rust's ChaCha8Rng), not from algorithmic differences.

Use model_args={"state_model": "local_linear_trend"} when you want the optional local linear trend state instead of the default local level model.

What Is Not Supported

Custom bsts model objects

These features may be added in future versions.