Compute DID estimate and pretrend flag
Company: Roblox
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: hard
Interview Round: Online Assessment
You are given three equal-length arrays describing observations from a panel-like dataset:
- `period[i]` ∈ {0,1}: 0 = pre period, 1 = post period
- `treat[i]` ∈ {0,1}: 1 = treated group, 0 = control group
- `y[i]` (float): outcome
Tasks:
A) Compute the **difference-in-differences (DID)** estimate of the treatment effect:
\[
\text{DID} = (\bar y_{post,treated} - \bar y_{pre,treated}) - (\bar y_{post,control} - \bar y_{pre,control})
\]
B) Compute a simple **pretrend flag**:
- Let \(d_{pre} = \bar y_{pre,treated} - \bar y_{pre,control}\).
- Given a scalar `threshold` ≥ 0, return `pretrend=True` if \(|d_{pre}| > threshold\), else `False`.
Assumptions/requirements:
- Use simple means over the subsets implied by (`period`, `treat`).
- Describe how you would handle missing cells (e.g., no treated observations in pre period).
Overview: This Analytics & Experimentation question evaluates basic causal-inference and data-manipulation skills by requiring computation of a difference-in-differences (DID) estimate and a binary pretrend flag from panel-like arrays, at an implementation-level focused on summary statistics and edge-case handling.
Read the full Roblox Data Scientist interview experience this question came from