Checking for Independence: Residual Sequence Plots
What you'll learn: How to use residual sequence plots to verify that your regression errors are independent, not correlated over time or observation order.
The Independence Assumption
In regression, we assume residuals (errors) are independent—each observation's error shouldn't depend on previous errors. When this assumption fails, your model's standard errors and hypothesis tests become unreliable. This commonly happens with time-series data or any ordered observations where consecutive errors might be related (called autocorrelation).
What is a Residual Sequence Plot?
A residual sequence plot graphs residuals against their observation order (or time). Simply put:
- X-axis: Observation number (1, 2, 3, ...) or time points
- Y-axis: Residuals
You're looking for patterns that suggest dependence between neighboring residuals.
What to Look For
Good (independence):
- Random scatter with no discernible pattern
- Residuals jumping up and down unpredictably
- Like flipping a coin—each flip independent of the last
Bad (dependence):
- Cycles or waves: Residuals oscillate in regular patterns
- Runs: Long stretches of consecutive positive residuals followed by negative ones (or vice versa)
- Trends: Residuals gradually increasing or decreasing over time
Think of it like tracking mood swings. Independent residuals are like random daily moods. Autocorrelated residuals are like moods influenced by yesterday's mood—if you were happy yesterday, you're likely happy today.
Why This Matters
When residuals are correlated, your regression underestimates uncertainty. You might report overly narrow confidence intervals or reject null hypotheses too readily. Time-series data, repeated measurements, or spatially ordered data are especially vulnerable.
Key Takeaway: Plot residuals in observation order to detect non-independence; random scatter indicates independence, while patterns, cycles, or runs suggest autocorrelation that violates regression assumptions.