What is a common shock design?

A common shock design has a single event at time $t_0$ that affects all units simultaneously: a regulatory change, platform update, pandemic, or a national campaign. Treatment is

$$ D_{it}=\mathbf{1}\{t\ge t_0\}, $$

so there is no cross-sectional variation in treatment timing.

Event studies are the workhorse

With no variation in timing, identification relies on time-series variation. The standard tool is an event study that estimates separate coefficients for each lead and lag around $t_0$.

  • Leads diagnose anticipation and help assess pre-shock stability.
  • Lags trace the dynamic response and accumulation over time.

The core identification risk

In pure common-shock settings, event studies are primarily descriptive unless you can defend an exclusion-style argument: nothing else changes at $t_0$ that could drive the observed break.

Without cross-sectional variation, the entire identification hinges on the claim that the shock is exogenous and unique. This is a strong assumption in marketing settings where multiple changes often co-occur (platform tweaks, pricing adjustments, macro conditions).

How to strengthen identification

When additional variation exists, you can move from descriptive to causal:

  • Exposure gradients: if units vary in exposure intensity, use a continuous-treatment DiD to exploit heterogeneity.
  • Comparison groups: if some units are plausibly less exposed, use them as a control.
  • Timing checks: confirm no other policy or operational changes align with $t_0$.

Diagnostics that matter

  • Pre-shock trend stability.
  • Sensitivity to alternative windows around $t_0$.
  • Robustness to adding controls for concurrent changes.

Takeaway

Common shock designs can be informative, but they are fragile. Event studies are the right first step, yet credible causal interpretation requires either strong exclusion arguments or additional cross-sectional variation in exposure.

References

  • Shaw, C. (2025). Causal Inference in Marketing: Panel Data and Machine Learning Methods (Community Review Edition), Section 3.2.4.
  • Sun, L., and Abraham, S. (2021). Estimating dynamic treatment effects in event studies. Journal of Econometrics.