LimitLine

Learn · updated 2026-10-28

Regression Analysis: How One Variable Drives Another

By Nicholas Williford, Lean Six Sigma Black Belt · About LimitLine

Regression analysis fits a line (or surface) through your data to estimate how much an output changes for each unit change in an input, and how much of its variation that input explains.

In improvement work, regression turns "hotter seems harder" into a number you can set: how much the output moves for each unit of the input, and how sure you can be. It is the tool for the Analyze phase once you suspect a cause.

Liney = a + b·xSlope bΣ(x − x̄)(y − ȳ) ÷ Σ(x − x̄)²Intercept aȳ − b·x̄R²share of the variation in y the line explains

Worked Example

Ten parts cured at temperatures from 150 to 195 °C, hardness measured on each (an illustrative example).

50556065150160170180190200150 °C: 52.1155 °C: 53.8160 °C: 55.2165 °C: 56.1170 °C: 58175 °C: 59.4180 °C: 60.2185 °C: 62.5190 °C: 63.1195 °C: 65cure temperature, °C
Hardness rises almost in a straight line with cure temperature.
ResultValueWhat it says
Slope b0.2805Each extra °C adds about 0.28 hardness units
Intercept a10.16The line's value at 0 °C: not meaningful here, far outside the data
R²0.995The line explains 99.5% of the variation in hardness
p-value of the slope< 0.0001Very unlikely to be chance alone
S (residual σ)0.313Typical scatter of a part around the line

So to reach hardness 60 you need about (60 − 10.16) ÷ 0.2805 = 177.7 °C, and individual parts will land within about ±0.76 hardness units of 60 (the 95% prediction interval there).

Before You Trust the Line

  1. Look at the residuals. Plot each part's distance from the line against x and in time order. A curve or a funnel means a straight line is the wrong model.
  2. Don't extrapolate. The data run from 150 to 195 °C; the line says nothing about 220 °C.
  3. Correlation is not cause. If temperature was raised on the same day the resin lot changed, the regression can't tell them apart. A designed experiment can.
Run a regression in LimitLineSimple and multiple regression with R², p-values, residual plots and VIF (a Pro tool; Free users can try it on the example).

Sources

  1. NIST/SEMATECH e-Handbook, §4.1.4.1 Linear Least Squares Regression
  2. D. C. Montgomery, E. A. Peck and G. G. Vining (2012), Introduction to Linear Regression Analysis, 5th edition, Wiley

The LimitLine Newsletter

About one email a month: new guides, tool updates and one practical SPC tip. Unsubscribe in one click.

We'll email you once to confirm before anything else is sent. Privacy