Economics
Saved & recent

Regression and control variables

Regression summarizes how an outcome varies with measured predictors. Controls can address some confounding, but omitted variables, measurement error and inappropriate controls can undermine causal interpretations.

Read first
Correlation and causation
Related
Causal identification, Sampling uncertainty and confidence intervals
Further reading
World Bank: Research design for impact evaluation
8 10 12 14 8 12 16 20 24 education / years wage / currency per hour blue · less experience violet · more experience dashed · pooled fit assumed cause E S W experience education wage
Hypothetical data: lower-experience pairs (10,10), (12,12), (14,14); higher-experience pairs (8,18), (10,20), (12,22). The pooled fitted slope is −4/11; within-group slopes are +1. This adjustment is causal only under the assumed confounding structure; controlling for mediators or selection variables can instead introduce bias.

View in the full Economics collection