Diagnostic Plots for ANOVA
You'll learn how to visually check whether your ANOVA assumptions hold by interpreting three key diagnostic plots.
Once you've run an ANOVA, you need to verify that your data meets the assumptions of independence, normality of residuals, and homogeneity of variance. Diagnostic plots are your visual toolkit for this health check—they reveal patterns that numbers alone might miss.
The Three Essential Plots
Residual Plot
This scatter plot shows fitted values (group means) on the x-axis and residuals (observed minus predicted) on the y-axis. You want to see random scatter around zero with no fan shapes or curves. If points form patterns—like a megaphone shape widening to the right—you've violated homogeneity of variance. If they curve, your model may be misspecified.
Q-Q Plot (Quantile-Quantile Plot)
This compares your residuals' distribution to a theoretical normal distribution. Points should fall roughly along a straight diagonal line. Deviations at the tails (S-curves or points peeling away) suggest non-normality. Think of it as a "normalcy report card"—if dots stray far from the line, your residuals aren't normally distributed.
Scale-Location Plot
Also called a "spread-location plot," this displays the square root of standardized residuals against fitted values. You're looking for a horizontal line with equal spread top-to-bottom. If the red trend line slopes or the scatter widens/narrows, you have unequal variances across groups.
Why It Matters
These plots catch violations that summary statistics might overlook. A significant Levene's test tells you variances differ, but residual plots show you how and where—guiding your next steps (transform data, use Welch's correction, or switch to non-parametric tests).
Key Takeaway: Diagnostic plots for ANOVA—residual, Q-Q, and scale-location—visually confirm that your residuals are randomly scattered, normally distributed, and have constant variance, ensuring your F-test conclusions are trustworthy.