Synthetic control

As explained in Causal Inference for Data Science

Synthetic control is a causal estimation method that constructs a predictive baseline from correlated time-series data to estimate what would have happened to a treated unit without the intervention.

Synthetic control estimates the causal effect of a time-related event when an A/B test is impractical. It builds a predictive baseline for the treated outcome from correlated time series that remain unaffected by the event. section 11.0 section 11.3

A new movie release

To estimate whether releasing Star Wars VII increased views of Star Wars III, train a model on pre-release views of correlated Lord of the Rings movies. Forecasting after the release estimates what Star Wars III views might have been without the new film; the gap from actual views estimates the effect. section 11.3

Apply the method in four stages:

• Choose a pretreatment interval before event time T for training and a post-treatment interval afterward for evaluation. section 11.3.2 section 11.3.5
• Select control series that predict outcome Y, need not measure the same thing, and are not affected by the event. section 11.3.2 section 11.3.3
• Train a forecasting model using only pretreatment observations, then predict the outcome throughout the post-treatment interval. section 11.3 section 11.3.5
• Average the differences between predicted and factual outcomes to estimate the average treatment effect on the treated. section 11.3.4

Interpretation depends on three assumptions: controls predict the outcome before T, the event does not affect controls, and no nearby event changes the outcome–control relationship. Violating these assumptions can bias the counterfactual estimate. section 11.3.3

Figure 11.13 We can train a model with the data before the new release and predict after the release. These new predictions tell us what would have happened if there were no release. The effect of the release can be estimated by the difference between the real values and the predicted ones.

Follow the pre-event training and post-event forecasting diagram to see how predicted outcomes represent the no-event counterfactual and how their difference from actual outcomes estimates impact. section 11.3

How Causal Inference for Data Science covers it

How deep it goes

Explains terminology, assumptions, four procedural steps, forecasting, and ATET calculation, while deferring detailed period selection.

Examples

Uses Star Wars and Lord of the Rings search trends plus simulated outcome and correlated control series.

What it assumes

Assumes time-series data and predictive modeling familiarity while defining outcomes, controls, treatment periods, and assumptions explicitly.

What you can do after

Apply it via a forecasting model to estimate an event’s post-treatment ATET from observed outcomes and unaffected controls.

How it connects

Connects causal inference with counterfactuals, time series, supervised learning, forecasting, and average treatment effect on the treated.

liveBookmanning.com

Read Causal Inference for Data Science on liveBook Get the book at manning.com

Related concepts

• language modeluses · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
• treatment effectmeasures · A treatment effect is the difference in an outcome attributable to receiving an intervention or exposure rather than a comparison condition.
• time seriesuses · A time series is a sequence of observations indexed or ordered by time.
• effect of treatment on the treatedproduces · The effect of treatment on the treated is the average causal difference between treated outcomes and their untreated potential outcomes among units that received treatment.
• regression discontinuity designcontrasts with · Regression discontinuity design is a causal inference method that estimates treatment effects by comparing observations near a cutoff where treatment assignment changes.
• experimentationalternative to · Experimentation is the deliberate testing of an idea, intervention, or alternative by observing outcomes to gather evidence or compare effects.

This explainer was generated by Manning's knowledge graph from the text of Causal Inference for Data Science. The section links point to the parts of the book it draws on.