Methodology

Lifesight combines two advanced Synthetic Control Methods (SCM) techniques: Augmented Synthetic Control Methods (ASCM) and Generalized Synthetic Controls (GSC).

Lifesight uses ASCM to estimate and correct for bias in the synthetic control. It then applies GSC, which is robust to small sample sizes and differences across units, to deliver reliable inference.

This two-step approach helps manage issues caused by the curse of dimensionality, where increasing the number of units or historical data makes exact matching harder and increases bias. GSC also uses parametric bootstrapping to provide solid estimates of uncertainty and inference.

By integrating ASCM and GSC, GeoLift offers a robust solution to bias in synthetic control estimation. This does require more processing power but delivers more reliable and accurate measurement for your marketing experiments.

Comparison

METHODOLOGY	ASSUMPTIONS	DRAWBACKS	OTHER IMPORTANT POINTS
Matched Market Tests (MMT)	Test and control would have followed parallel trends without treatment; no hidden differences.	Low power; high risk of bias if the control is not a true twin; vulnerable to anomalies in either market.	Simple and cost-effective for a few geos; historical precursor to synthetic-control methods
Augmented Synthetic Control (ASCM)	Reduces bias via ASCM, delivers higher statistical power and tighter confidence intervals, supports multiple test markets and built-in power analysis for design.	More complex and compute-intensive; needs sufficient historical data and technical expertise to implement and interpret.	Delivers tighter confidence intervals and higher sensitivity than single-market tests