Independence of Observations
What you'll learn: Why each data point in your regression must stand alone and what happens when they don't.
The Core Idea
When you fit a linear regression model, each observation (row of data) should be independent of every other observation. This means that knowing the value or residual of one observation shouldn't give you information about another.
Think of it like flipping a fair coin: each flip doesn't "remember" the previous flip. Your regression observations should work the same way—each one provides fresh, independent information.
Why Independence Matters
When observations are not independent, your standard errors become unreliable. You might think your coefficients are statistically significant when they're really not, leading to false confidence in your model.
Common Violations
Autocorrelation (also called serial correlation) is the most common violation. It occurs when:
- Time series data: Today's stock price is influenced by yesterday's
- Clustered data: Multiple measurements from the same person or location
- Spatial data: Neighboring geographic areas influence each other
For example, if you're predicting daily ice cream sales, a hot day often follows another hot day, and high sales follow high sales—the observations are correlated through time.
Detecting and Addressing Violations
If you suspect autocorrelation, you can:
- Check residual plots for patterns over time
- Use the Durbin-Watson test (a future lesson!)
- Adjust your model with time-lagged variables
- Use methods designed for dependent data (like time series models)
Key Takeaway: Independence means each observation contributes unique information—when violated through autocorrelation or clustering, your model's inference becomes unreliable even if predictions look reasonable.