### Install and Load CausalImpact Package

Source: https://github.com/google/causalimpact/blob/master/README.md

Installs the CausalImpact package from CRAN and loads it into the R session. This is the first step before using any of its functionalities.

```r
install.packages("CausalImpact")
library(CausalImpact)
```

--------------------------------

### Perform Causal Impact Analysis with Date-Based Time Series

Source: https://context7.com/google/causalimpact/llms.txt

This example demonstrates using date-indexed zoo objects for causal impact analysis, which is useful for real-world data. The package automatically formats plots with dates and allows period specification using actual dates.

```r
library(CausalImpact)
library(zoo)

# Create example data with dates
set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 100)
y <- 1.2 * x1 + rnorm(100)
y[71:100] <- y[71:100] + 10

# Create date-indexed zoo object
time.points <- seq.Date(as.Date("2014-01-01"), by = 1, length.out = 100)
data <- zoo(cbind(y, x1), time.points)

# Specify periods using dates
pre.period <- as.Date(c("2014-01-01", "2014-03-11"))
post.period <- as.Date(c("2014-03-12", "2014-04-10"))

# Run analysis
impact <- CausalImpact(data, pre.period, post.period)

# The model stores period information in date format
print(impact$model$pre.period)
# [1] "2014-01-01" "2014-03-11"
print(impact$model$post.period)
# [1] "2014-03-12" "2014-04-10"
```

--------------------------------

### Integrate Custom bsts Model into CausalImpact

Source: https://context7.com/google/causalimpact/llms.txt

Pass a pre-fitted bsts model to CausalImpact for advanced state space configuration. Ensure the post-period response is set to NA in the input data.

```r
library(CausalImpact)
library(bsts)

# Create data
set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 52)
y <- 1.2 * x1 + rnorm(52)
y[41:52] <- y[41:52] + 10

pre.period <- c(1, 40)
post.period <- c(41, 52)

# Store actual post-period response before setting to NA
post.period.response <- y[post.period[1]:post.period[2]]

# Set post-period response to NA (required for bsts model)
y[post.period[1]:post.period[2]] <- NA

# Build custom bsts model with local level state component
ss <- AddLocalLevel(list(), y)

# Fit the model (use ping = 0 to suppress iteration output)
bsts.model <- bsts(y ~ x1, ss, niter = 1000, ping = 0)

# Pass the fitted model to CausalImpact
impact <- CausalImpact(bsts.model = bsts.model,
                       post.period.response = post.period.response)

# Results can be used normally
summary(impact)
plot(impact)
```

--------------------------------

### Generate Numerical Summary Statistics

Source: https://context7.com/google/causalimpact/llms.txt

Produce a table of actual vs. predicted values and effects, or access the raw summary data frame.

```r
library(CausalImpact)

set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 100)
y <- 1.2 * x1 + rnorm(100)
y[71:100] <- y[71:100] + 10
data <- cbind(y, x1)

impact <- CausalImpact(data, c(1, 70), c(71, 100))

# Print summary table
summary(impact)
# Output:
# Posterior inference {CausalImpact}
#
#                          Average        Cumulative
# Actual                   99             2982
# Prediction (s.d.)        89 (0.31)      2665 (9.3)
# 95% CI                   [88, 89]       [2647, 2683]
#
# Absolute effect (s.d.)   11 (0.31)      317 (9.3)
# 95% CI                   [9.9, 11]      [299, 335]
#
# Relative effect (s.d.)   12% (0.35%)    12% (0.35%)
# 95% CI                   [11%, 13%]     [11%, 13%]
#
# Posterior tail-area probability p: 0.001
# Posterior probability of an effect: 99.9%

# Access raw summary data for custom processing
impact$summary
# Returns data frame with columns: Actual, Pred, Pred.lower, Pred.upper, Pred.sd,
# AbsEffect, AbsEffect.lower, AbsEffect.upper, AbsEffect.sd,
# RelEffect, RelEffect.lower, RelEffect.upper, RelEffect.sd, p

# Control decimal precision
summary(impact, digits = 4)
```

--------------------------------

### Customize Causal Impact Model with model.args

Source: https://context7.com/google/causalimpact/llms.txt

Fine-tune the underlying Bayesian structural time-series model using the `model.args` parameter. This allows control over MCMC iterations, seasonality, prior specifications, and regression dynamics. Adjust `nseasons` and `season.duration` for different seasonal patterns.

```r
library(CausalImpact)

# Create data
set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 100)
y <- 1.2 * x1 + rnorm(100)
y[71:100] <- y[71:100] + 10
data <- cbind(y, x1)

pre.period <- c(1, 70)
post.period <- c(71, 100)

# Run analysis with custom model parameters
impact <- CausalImpact(data, pre.period, post.period,
  model.args = list(
    niter = 5000,                  # Number of MCMC samples (default: 1000)
    standardize.data = TRUE,       # Standardize data before fitting (default: TRUE)
    prior.level.sd = 0.01,         # Prior SD for local level (default: 0.01)
    nseasons = 1,                  # Seasonal period (default: 1, no seasonality)
    season.duration = 1,           # Duration of each season (default: 1)
    dynamic.regression = FALSE     # Time-varying coefficients (default: FALSE)
  )
)

# For weekly seasonality with daily data:
# model.args = list(nseasons = 7, season.duration = 1)

# For weekly seasonality with hourly data:
# model.args = list(nseasons = 7, season.duration = 24)

# For volatile data, use larger prior.level.sd:
# model.args = list(prior.level.sd = 0.1)
```

--------------------------------

### Generate Verbal Interpretation Report

Source: https://context7.com/google/causalimpact/llms.txt

Produce a plain-language summary of the analysis findings or access the full report text directly.

```r
library(CausalImpact)

set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 100)
y <- 1.2 * x1 + rnorm(100)
y[71:100] <- y[71:100] + 10
data <- cbind(y, x1)

impact <- CausalImpact(data, c(1, 70), c(71, 100))

# Print verbal report
summary(impact, "report")
# Output:
# Analysis report {CausalImpact}
#
# During the post-intervention period, the response variable had an average
# value of approx. 99. By contrast, in the absence of an intervention, we
# would have expected an average response of 89. The 95% interval of this
# counterfactual prediction is [88, 89]. Subtracting this prediction from
# the observed response yields an estimate of the causal effect the
# intervention had on the response variable. This effect is 11 with a 95%
# interval of [9.9, 11]. For a discussion of the significance of this
# effect, see below.
#
# Summing up the individual data points during the post-intervention period
# ... [continues with cumulative effects and statistical significance]

# The full report text is also stored in the object
cat(impact$report)
```

--------------------------------

### Examine Predictor Inclusion Probabilities

Source: https://context7.com/google/causalimpact/llms.txt

Analyze variable influence by inspecting posterior inclusion probabilities from the underlying bsts model. Access the bsts model directly for advanced diagnostics.

```r
library(CausalImpact)
library(bsts)

# Create data with multiple predictors
set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 100)
x2 <- 50 + arima.sim(model = list(ar = 0.9), n = 100)
x3 <- rnorm(100)  # Noise predictor
y <- 1.2 * x1 + 0.5 * x2 + rnorm(100)
y[71:100] <- y[71:100] + 10

data <- cbind(y, x1, x2, x3)

impact <- CausalImpact(data, c(1, 70), c(71, 100))

# Plot predictor inclusion probabilities
# Bar height shows probability of inclusion in model
# Shading indicates direction (white = negative, black = positive)
plot(impact$model$bsts.model, "coefficients")

# Access the bsts model directly for more diagnostics
bsts.model <- impact$model$bsts.model
print(names(bsts.model))

# View posterior samples
posterior.samples <- impact$model$posterior.samples
dim(posterior.samples)
# Returns [n_samples x n_observations] matrix
```

--------------------------------

### Configure CausalImpact with Custom Confidence Level

Source: https://context7.com/google/causalimpact/llms.txt

Adjust the alpha parameter to change the credible interval width from the default 95% to a custom level.

```r
library(CausalImpact)

set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 100)
y <- 1.2 * x1 + rnorm(100)
y[71:100] <- y[71:100] + 10
data <- cbind(y, x1)

pre.period <- c(1, 70)
post.period <- c(71, 100)

# Run analysis with 90% credible intervals
impact <- CausalImpact(data, pre.period, post.period, alpha = 0.1)

# The alpha value is stored in the model
print(impact$model$alpha)
# [1] 0.1

# Summary will show 90% CI instead of 95% CI
summary(impact)
```

--------------------------------

### Visualize CausalImpact Results

Source: https://context7.com/google/causalimpact/llms.txt

Generate multi-panel plots for observed data, counterfactual predictions, and causal effects. The plot function returns a ggplot2 object, allowing for further customization.

```r
library(CausalImpact)
library(ggplot2)

set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 100)
y <- 1.2 * x1 + rnorm(100)
y[71:100] <- y[71:100] + 10
data <- cbind(y, x1)

impact <- CausalImpact(data, c(1, 70), c(71, 100))

# Create default 3-panel plot
plot(impact)
# Panel 1 (original): Shows observed data (solid) vs counterfactual prediction (dashed)
# Panel 2 (pointwise): Shows point-by-point causal effect
# Panel 3 (cumulative): Shows cumulative causal effect over time

# Select specific panels
plot(impact, "original")                    # Only counterfactual comparison
plot(impact, "pointwise")                   # Only pointwise effects
plot(impact, "cumulative")                  # Only cumulative effects
plot(impact, c("original", "pointwise"))    # Two panels

# Customize the plot (returns ggplot2 object)
impact.plot <- plot(impact) +
  theme_bw(base_size = 20) +
  labs(title = "Causal Impact of Marketing Campaign")
print(impact.plot)

# Add custom y-axis label
impact.plot <- plot(impact) + ylab("Daily Sales ($)")
print(impact.plot)

# Access the underlying series data for custom plotting
series.data <- as.data.frame(impact$series)
head(series.data)
```

--------------------------------

### Access Time Series Results

Source: https://context7.com/google/causalimpact/llms.txt

Extract detailed time series data including response, predictions, and effects with credible intervals. Data is returned as a zoo object.

```r
library(CausalImpact)
library(zoo)

set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 100)
y <- 1.2 * x1 + rnorm(100)
y[71:100] <- y[71:100] + 10
data <- cbind(y, x1)

impact <- CausalImpact(data, c(1, 70), c(71, 100))

# The series field contains all time series data as a zoo object
series <- impact$series

# Available columns:
print(names(series))
# [1] "response"           "cum.response"       "point.pred"
# [4] "point.pred.lower"   "point.pred.upper"   "cum.pred"
# [7] "cum.pred.lower"     "cum.pred.upper"     "point.effect"
# [10] "point.effect.lower" "point.effect.upper" "cum.effect"
# [13] "cum.effect.lower"   "cum.effect.upper"

# Extract specific columns
observed <- series$response
predicted <- series$point.pred
effect <- series$point.effect
cumulative_effect <- series$cum.effect

# Get post-period effects only
post.start <- 71
post.end <- 100
post.effects <- window(series, start = post.start, end = post.end)

# Convert to data frame for further analysis
df <- as.data.frame(series)
df$time <- index(series)

# Calculate average effect in post-period
mean(series$point.effect[post.start:post.end], na.rm = TRUE)
```

--------------------------------

### Perform Causal Impact Analysis with Numeric Time Series

Source: https://context7.com/google/causalimpact/llms.txt

Use this function to perform causal impact analysis on numeric time series data. Ensure your data is in a matrix format with the response variable in the first column and covariates in subsequent columns. Define pre- and post-intervention periods using numeric indices.

```r
install.packages("CausalImpact")
library(CausalImpact)

# Create example data with a response variable (y) and covariate (x1)
set.seed(1)
x1 <- 100 + arima.sim(model = list(ar = 0.999), n = 100)
y <- 1.2 * x1 + rnorm(100)
y[71:100] <- y[71:100] + 10  # Add intervention effect of 10 units

# Combine into data matrix (response in first column, covariates in subsequent columns)
data <- cbind(y, x1)

# Define pre-intervention and post-intervention periods
pre.period <- c(1, 70)
post.period <- c(71, 100)

# Run the causal impact analysis
impact <- CausalImpact(data, pre.period, post.period)

# Access results
print(names(impact))
# [1] "series"  "summary" "report"  "model"

# View the time series with predictions
head(impact$series)
# Returns zoo object with: response, cum.response, point.pred, point.pred.lower, 
# point.pred.upper, cum.pred, cum.pred.lower, cum.pred.upper, point.effect, 
# point.effect.lower, point.effect.upper, cum.effect, cum.effect.lower, cum.effect.upper
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.