Explain confounding with an Uber example

Question

Imagine you are sitting an interview for a Data Scientist position and you have been given access to Uber or Uber Eats data. Address the following points concerning confounding in causal inference:

1. In the setting where causal effects are estimated from observational data, define confounding. State what a confounder is, and explain the reason a confounder is capable of biasing an observed association between an exposure and an outcome.

2. Supply a specific example drawn from Uber (do not fall back on broad demographic variables such as age or sex). Within that example, make each of the following unambiguous:

   • the treatment or exposure, written as X,
   • the outcome, written as Y, and
   • the confounder, written as Z, which acts on both X and Y.

   Then give an intuitive explanation of the direction in which the bias operates — whether it could generate a spurious effect or conceal a genuine one.

3. Set out no fewer than two concrete methods you would apply to identify or lessen confounding in an analysis, whether during design or during modeling, and for each method specify the assumptions it depends upon.

Overview:

This Statistics and Math interview prompt for a Data Scientist gauges a candidate's grasp of confounding and causal inference when data are observational. A candidate is expected to define a confounder; to provide a concrete, non-demographic Uber example that identifies the exposure, the outcome, and the confounder along with the direction of the bias; and to describe at least two mitigation approaches together with the assumptions each one requires.