Make CausalImpact compatible with ggplot2 version 4 and testthat edition 3:

* Use vdiffr for plot comparisons to avoid relying on internal representation of ggplot objects.
* Replace deprecated expectation statements.

Author: Alain Hauser

DESCRIPTION

@@ -1,22 +1,30 @@
 Package: CausalImpact
 Title: Inferring Causal Effects using Bayesian Structural Time-Series Models
-Date: 2022-11-07
-Author: Kay H. Brodersen <kbrodersen@google.com>,
-  Alain Hauser <alhauser@google.com>
+Date: 2025-09-24
+Authors@R: c(person(given = c("Kay", "H."),
+                    family = "Brodersen",
+                    role = "aut",
+                    email = "kbrodersen@google.com"),
+             person(given = "Alain",
+                    family = "Hauser",
+                    role = c("aut", "cre"),
+                    email = "alhauser@google.com"))
 Maintainer: Alain Hauser <alhauser@google.com>
 URL: https://google.github.io/CausalImpact/
 Description: Implements a Bayesian approach to causal impact estimation in time
   series, as described in Brodersen et al. (2015) <DOI:10.1214/14-AOAS788>.
   See the package documentation on GitHub
   <https://google.github.io/CausalImpact/> to get started.
-Copyright: Copyright (C) 2014-2022 Google, Inc.
-Version: 1.3.0
+Copyright: Copyright (C) 2014-2025 Google, Inc.
+Version: 1.4.1
 VignetteBuilder: knitr
 License: Apache License 2.0 | file LICENSE
 Imports: assertthat (>= 0.2.0), Boom, ggplot2, zoo
 Depends: bsts (>= 0.9.0)
-Suggests: 
+Config/testthat/edition: 3
+Suggests:
     covr,
     knitr,
     rmarkdown,
-    testthat
+    testthat,
+    vdiffr

R/impact_analysis.R

@@ -583,7 +583,7 @@ PrintSummary <- function(impact, digits = 2L) {
   # Print overall tail-area probability
   p <- summary$p[1]
   cat(paste0("Posterior tail-area probability p:   ", round(p, 5), "\n"))
-  cat(paste0("Posterior prob. of a causal effect:  ",
+  cat(paste0("Posterior probability of an effect:  ",
              round((1 - p) * 100, ifelse(p < 0.01, 5, ifelse(p < 0.05, 3, 0))),
              "%\n"))
   cat("\n")

R/impact_inference.R

@@ -356,6 +356,14 @@ InterpretSummaryTable <- function(summary, digits = 2L) {
                  "percentage is [", rel.effect.lower[1], ", ",
                  rel.effect.upper[1], "].")
 
+  # Comment on expected relative effect
+  if (abs(summary$RelEffect[1]-summary$AbsEffect[1]/summary$Pred[1])>=.05) {
+    stmt <- paste0(stmt, "\n\n(Note that the expected relative ",
+                 "effect isn't generally the same as the expected absolute ",
+                 "effect divided by the expected prediction since ",
+                 "distributions are often not symmetric.)")
+  }
+
   # Comment on significance
   if (sig && pos) {
     stmt <- paste0(stmt, "\n\nThis means that the positive effect observed ",
@@ -409,14 +417,18 @@ InterpretSummaryTable <- function(summary, digits = 2L) {
   if (p < summary$alpha[1]) {
     stmt <- paste0(stmt, "\n\nThe probability of obtaining this effect by ",
                    "chance is very small (Bayesian one-sided tail-area ",
-                   "probability p = ", round(p, 3), "). This means the causal ",
-                   "effect can be considered statistically significant.")
+                   "probability p = ", round(p, 3), "). This means the ",
+                   "effect is statistically significant. It can be considered ",
+                   "causal if the model assumptions are satisfied.")
   } else {
     stmt <- paste0(stmt, "\n\nThe probability of obtaining this ",
                    "effect by chance is p = ", round(p, 3), ". This ",
                    "means the effect may be spurious and would generally ",
                    "not be considered statistically significant.")
   }
+  stmt <- paste0(stmt, " For more details, including the model assumptions ",
+                 "behind the method, see ",
+                 "https://google.github.io/CausalImpact/.")
   return(stmt)
 }
 

R/impact_plot.R

@@ -137,20 +137,20 @@ CreateImpactPlot <- function(impact, metrics = c("original", "pointwise",
                                     impact$model$post.period,
                                     time(impact$series))
   q <- q + geom_vline(xintercept = xintercept,
-                      colour = "darkgrey", size = 0.8, linetype = "dashed")
+                      colour = "darkgrey", linewidth = 0.8, linetype = "dashed")
 
   # Add zero line to pointwise and cumulative plot
   q <- q + geom_line(aes(y = baseline),
-                     colour = "darkgrey", size = 0.8, linetype = "solid",
+                     colour = "darkgrey", linewidth = 0.8, linetype = "solid",
                      na.rm = TRUE)
 
   # Add point predictions
   q <- q + geom_line(aes(y = mean), data,
-                     size = 0.6, colour = "darkblue", linetype = "dashed",
+                     linewidth = 0.6, colour = "darkblue", linetype = "dashed",
                      na.rm = TRUE)
 
   # Add observed data
-  q <- q + geom_line(aes(y = response), size = 0.6, na.rm = TRUE)
+  q <- q + geom_line(aes(y = response), linewidth = 0.6, na.rm = TRUE)
   return(q)
 }
 

README.md

@@ -24,6 +24,13 @@ to be stable during the post-intervention period. Understanding and checking
 these assumptions for any given application is critical for obtaining valid
 conclusions.
 
+For Python users we recommend the implementation in
+[TFP CausalImpact](https://github.com/google/tfp-causalimpact), contributed by
+Colin Carroll, David Moore, Jacob Burnim, Kyle Loveless, and Susanna Makela. The
+Python implementation uses TensorFlow Probability and has been designed to
+produce results close to those of the R package. Some differences are expected
+given the nature of randomized algorithms.
+
 ## Installation
 
 ```r
@@ -40,12 +47,15 @@ library(CausalImpact)
 ## Further resources
 
 *   Manuscript:
-    [Brodersen et al., Annals of Applied Statistics (2015)](https://research.google/pubs/pub41854/)
+    [Brodersen et al., Annals of Applied Statistics (2015)](https://research.google/pubs/inferring-causal-impact-using-bayesian-structural-time-series-models/)
 
 *   For questions on the statistics behind CausalImpact:
     [Cross Validated](https://stats.stackexchange.com/questions/tagged/causalimpact)
 
 *   For questions on how to use the CausalImpact R package:
-    [Stack Overflow](https://stackoverflow.com/questions/tagged/causalimpact)
+    [Stack Overflow](https://stackoverflow.com/questions/tagged/causal-inference)
 
 *   [Bug reports](https://github.com/google/CausalImpact/issues)
+
+*   Python version:
+    [TFP CausalImpact](https://github.com/google/tfp-causalimpact)

inst/CITATION

@@ -8,7 +8,7 @@ bibentry(
 
     title = paste0("Inferring causal impact using {B}ayesian structural ",
                    "time-series models"),
-    author = personList(
+    author = c(
         as.person("Kay H. Brodersen"),
         as.person("Fabian Gallusser"),
         as.person("Jim Koehler"),
@@ -18,5 +18,5 @@ bibentry(
     year = "2014",
     volume = "9",
     pages = "247--274",
-    url = "https://research.google/pubs/pub41854/"
+    url = "https://research.google/pubs/inferring-causal-impact-using-bayesian-structural-time-series-models/"
 )

tests/testthat/test-impact-analysis.R

@@ -1,4 +1,4 @@
-# Copyright 2014-2022 Google Inc. All rights reserved.
+# Copyright 2014-2025 Google Inc. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -12,8 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-testthat::context("Unit tests for impact_analysis.R")
-
 # Authors: kbrodersen@google.com (Kay H. Brodersen)
 #          gallusser@google.com (Fabian Gallusser)
 
@@ -48,7 +46,6 @@ CallAllS3Methods <- function(impact) {
   expect_error(print(impact, "foo"), "summary")
   expect_error(plot(impact), NA)
   expect_error(q <- plot(impact), NA)
-  expect_error(plot(q), NA)
 }
 
 test_that("FormatInputForCausalImpact", {
@@ -736,6 +733,7 @@ test_that("PrintSummary", {
                                           model.args))
   expect_output(PrintSummary(impact), "\\([0-9.]{3}[0-9]*\\)")
   expect_output(PrintSummary(impact, digits = 10), "\\([0-9.]{11}[0-9]*\\)")
+  expect_output(PrintSummary(impact), "Posterior probability of an effect")
   expect_error(PrintSummary(impact, digits = 0), "positive")
   expect_error(PrintSummary(impact, digits = "test"), "positive")
 })

tests/testthat/test-impact-inference.R

@@ -12,8 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-testthat::context("Unit tests for impact_inference.R")
-
 # Author: kbrodersen@google.com (Kay Brodersen)
 
 test_that("GetPosteriorStateSamples", {
@@ -293,7 +291,7 @@ test_that("CompilePosteriorInferences", {
                                            alpha, UnStandardize)
   expect_equal(names(inferences), c("series", "summary", "report",
                                     "posterior.samples"))
-  expect_is(inferences$posterior.samples, "matrix")
+  expect_true(is.matrix(inferences$posterior.samples))
   expect_equal(ncol(inferences$posterior.samples), 200)
   expect_gte(nrow(inferences$posterior.samples), 80)
   expect_lte(nrow(inferences$posterior.samples), 100)

tests/testthat/test-impact-misc.R

@@ -12,8 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-testthat::context("Unit tests for impact_misc.R")
-
 # Authors: kbrodersen@google.com (Kay Brodersen)
 #          gallusser@google.com (Fabian Gallusser)
 #          alhauser@google.com (Alain Hauser)
@@ -188,7 +186,7 @@ test_that("StandardizeAllVariables", {
     expect_equal(sd(result$data[, column]), 1, tolerance = 0.0001)
   })
   expect_equal(result$UnStandardize, Standardize(data[, 1])$UnStandardize,
-               check.environment = FALSE)
+               ignore_function_env = TRUE)
 
   # Test that several columns are standardized correctly when fitting mean and
   # SD only over part of the rows.
@@ -198,7 +196,7 @@ test_that("StandardizeAllVariables", {
     expect_equal(sd(result$data[11 : 90, column]), 1, tolerance = 0.0001)
   })
   expect_equal(result$UnStandardize, Standardize(data[, 1])$UnStandardize,
-               check.environment = FALSE)
+               ignore_function_env = TRUE)
 
   # Test healthy input: single series only
   set.seed(1)
@@ -209,15 +207,15 @@ test_that("StandardizeAllVariables", {
   expect_equal(mean(result$data), 0, tolerance = 0.0001)
   expect_equal(sd(result$data), 1, tolerance = 0.0001)
   expect_equal(result$UnStandardize, Standardize(data)$UnStandardize,
-               check.environment = FALSE)
+               ignore_function_env = TRUE)
 
   # Test that a single series is standardized correctly when fitting mean and SD
   # only over part of the data range.
   result <- StandardizeAllVariables(data, c(11, 90))
   expect_equal(mean(result$data[11: 90]), 0, tolerance = 0.0001)
   expect_equal(sd(result$data[11 : 90]), 1, tolerance = 0.0001)
   expect_equal(result$UnStandardize, Standardize(data)$UnStandardize,
-               check.environment = FALSE)
+               ignore_function_env = TRUE)
 })
 
 test_that("GetPeriodIndices.InvalidInput", {

tests/testthat/test-impact-model.R

@@ -12,8 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-testthat::context("Unit tests for impact_model.R")
-
 # Author: kbrodersen@google.com (Kay Brodersen)
 
 test_that("ObservationsAreIllConditioned", {
@@ -175,9 +173,12 @@ test_that("ConstructModel", {
     expect_false(is.null(bsts.model))
     expect_equal(class(bsts.model), "bsts")
     expect_equal(as.numeric(bsts.model$original.series), as.numeric(data[, 1]))
-    expect_equivalent(bsts.model$predictors[, 1], rep(1, nrow(data)))
-    expect_equivalent(bsts.model$predictors[, 2], as.numeric(data[, 2]))
-    expect_equivalent(bsts.model$predictors[, 3], as.numeric(data[, 3]))
+    expect_equal(bsts.model$predictors[, 1], rep(1, nrow(data)),
+                 ignore_attr = TRUE)
+    expect_equal(bsts.model$predictors[, 2], as.numeric(data[, 2]),
+                 ignore_attr = TRUE)
+    expect_equal(bsts.model$predictors[, 3], as.numeric(data[, 3]),
+                 ignore_attr = TRUE)
     expect_equal(bsts.model$state.contributions,
                  expected.model$state.contributions)
   })