The standard deviation of the residuals (s) measures the typical distance that observed y-values fall from the regression line. In AP Stats, it estimates σ (the spread around the population regression line) and plugs into the standard error of the slope, SE = s/(sx√(n-1)), for Unit 9 inference.
The standard deviation of the residuals, written as s, tells you how far data points typically land from the regression line. Every point has a residual, which is the actual y-value minus the predicted one (yi − ŷi). The value s is essentially the "average size" of those residuals, measured in the same units as y. If you're predicting exam scores and s = 4.2, your predictions are typically off by about 4.2 points. Smaller s means tighter clustering around the line and a model that predicts more precisely.
In Unit 9, s gets a bigger job. The CED frames it this way (per AP Stats 9.2.A): each residual from your sample regression line estimates the deviation of y from the population regression line, and s estimates σ, the standard deviation of all those deviations in the population. That makes s the bridge from describing your sample to doing inference about the true slope β. You'll almost always read s off computer output (it's the "S" value in regression printouts) rather than compute it by hand.
This term lives in Unit 9: Inference for Quantitative Data: Slopes, specifically Topic 9.2 (Confidence Intervals for the Slope of a Regression Model). It directly supports learning objective AP Stats 9.2.C, where the margin of error for the slope is t* times the standard error, and that standard error is built from s through the formula SE = s/(sx√(n-1)). It also underpins AP Stats 9.2.A, since s is your sample estimate of σ, the population-level spread around the true regression line. Without s, you can't construct the confidence interval b ± t*(SEb) from AP Stats 9.2.D. On the exam, s is one of the numbers you have to correctly identify on a regression output table, and confusing it with sx or R² is one of the fastest ways to lose points on a slope-inference problem.
Keep studying AP Statistics Unit 9
Residuals (Unit 2)
Residuals are the raw ingredients of s. You first meet residuals in Unit 2 as a way to judge fit visually with residual plots. Unit 9 takes those same residuals and condenses their spread into one number, s, so you can do inference instead of just description.
Margin of Error and Confidence Intervals (Units 6 and 9)
Every confidence interval follows the pattern estimate ± (critical value)(standard error). For slopes, s feeds the standard error through SE = s/(sx√(n-1)), so a noisier scatterplot (bigger s) means a wider interval for the true slope β.
Coefficient of Determination R² (Unit 2)
R² and s answer related but different questions. R² gives the proportion of variation in y explained by the model (a unitless percent), while s gives the typical prediction error in actual y-units. A model can have a high R² and still have an s too large to be useful for your purposes.
Degrees of Freedom (Units 7 and 9)
Slope inference uses a t-distribution with n − 2 degrees of freedom, not n − 1 like a one-sample t procedure. You lose two degrees of freedom because the regression line estimates two parameters, the slope and the intercept.
Expect to work with s when reading computer output. A typical multiple-choice stem shows a regression printout and asks you to identify s (often labeled simply "S"), interpret it in context, or use it to find the standard error of the slope. The classic interpretation template is worth memorizing in your own words. Something like "the actual [y-variable] is typically about s [units] away from the value predicted by the regression line" earns credit. On free-response questions about confidence intervals or significance tests for slope, you'll usually pull SEb straight from the output rather than rebuilding it from s, but knowing that SE = s/(sx√(n-1)) helps you explain why more scatter or a smaller sample widens your interval. No released FRQ requires computing s by hand; the exam tests whether you can find it, interpret it, and connect it to the precision of your slope estimate.
These two share a printout and a formula, so they get mixed up constantly. The standard deviation of the residuals (s) describes scatter of data points around the regression line, in y-units. The standard error of the slope (SEb) describes how much the slope b itself would vary from sample to sample. They're linked by SE = s/(sx√(n-1)), so s is an input to SEb, not the same thing. When you build the interval b ± t*(SEb), the number multiplied by t* is SEb, never s.
The standard deviation of the residuals (s) measures the typical distance that actual y-values fall from the regression line, in the same units as y.
In Unit 9, s serves as the sample estimate of σ, the standard deviation of deviations from the population regression line.
The standard error of the slope is SE = s/(sx√(n-1)), so larger scatter (bigger s) or a smaller sample makes the slope estimate less precise.
On regression computer output, s usually appears simply as "S," and you need to distinguish it from sx, R², and SEb.
A smaller s means the points hug the regression line more tightly, which translates into narrower confidence intervals for the true slope β.
Interpret s in context as "predictions from this line are typically off by about s [units of y]."
It's the number s that measures how far observed y-values typically fall from the regression line, in y-units. In Unit 9, it estimates σ, the spread of the population around the true regression line, and feeds into the standard error of the slope.
No. The value s describes scatter of data points around the line, while SEb describes sample-to-sample variability of the slope b. They're connected by SE = s/(sx√(n-1)), so s is an ingredient in SEb, not a synonym for it.
R² is a unitless proportion telling you what fraction of variation in y the model explains, while s tells you the typical prediction error in actual units of y. For example, R² = 0.85 says the model explains 85% of the variation, but only s tells you whether predictions are off by 2 points or 200.
Look for the value labeled "S" near the bottom of the printout, usually next to "R-Sq." Don't confuse it with the numbers in the coefficient table, where the "SE Coef" column gives the standard error of the slope instead.
No. Exam questions give you s on computer output and ask you to identify it, interpret it in context, or use it within SE = s/(sx√(n-1)). Your job is interpretation and application, not hand computation.