Question 1

What is Synthetic Control and when should I use it?

Accepted Answer

Synthetic control is a causal inference method for estimating the effect of an intervention when you only have aggregated data (like cities, states, or stores) and can't use traditional difference-in-differences. Instead of relying on a single control unit, it creates a 'synthetic' control by combining multiple untreated units to better match the treated unit's pre-intervention behavior.

Question 2

How does synthetic control work?

Accepted Answer

The algorithm finds optimal weights for donor units to minimize the difference between the treated unit and the synthetic control in the pre-intervention period. These weights are typically constrained to be non-negative and sum to 1 (a convex combination). Once fitted, the same weights are applied to the post-intervention period to estimate what would have happened without the treatment.

Question 3

What does 'Pre RMSE' tell me?

Accepted Answer

Pre-treatment RMSE (Root Mean Squared Error) measures how well the synthetic control matches the actual treated unit before the intervention. A lower RMSE indicates a better fit. If the pre-treatment fit is poor, the post-treatment effect estimates are less reliable. It's common to set a threshold (e.g., RMSE < 80) and exclude poorly-fitted units from analysis.

Question 4

How are confidence intervals computed?

Accepted Answer

We run placebo (permutation) tests by iteratively treating each donor unit as if it were treated and refitting the model using the remaining donors. The distribution of placebo effects defines pointwise bands and aggregate confidence intervals (e.g., 95%).

Question 5

What is Ridge regularization (λ)?

Accepted Answer

Ridge adds L2 regularization (λ||w||²) to prevent overfitting and stabilize weights. A small value (0.001) prevents weights from becoming extreme while allowing good fit. Higher values make weights more uniform but may worsen fit. Lower values allow more extreme weights but risk overfitting. The default 0.001 works well for most cases.

Question 6

How do I interpret the results?

Accepted Answer

The 'effect' is the difference between actual and synthetic outcomes in each post-intervention period. 'Post Mean Effect' is the average effect across all post-treatment periods. 'Post Cumulative Effect' sums up all effects over time, showing the total impact. The vertical line in charts marks T0 (intervention time). Before T0, actual and synthetic should track closely. After T0, divergence indicates the treatment effect.

Synthetic Control Analysis

1. Data Upload

2. Configuration

Frequently Asked Questions

Explore More Tools