Least Squares Regression

Fitting a Line Through Scattered Data Points

Least squares regression finds the single straight line that best fits a scatter of data points. Enter a list of x, y pairs — one per line — and this calculator finds the line’s slope and intercept, plus how strongly the points actually follow that line.

The Formula

$$\vC{m} = \frac{n\sum{xy} - \sum{x}\sum{y}}{n\sum{x^2} - \left(\sum{x}\right)^2}$$ $$\vC{b} = \frac{\sum{y} - m\sum{x}}{n}$$

where $m$ is the slope, $b$ is the intercept, and $n$ is the number of data points — together they describe the line $y = mx + b$ that minimizes the total squared distance from every point to the line.

Worked Example

Fitting a line through (1, 1), (2, 2), and (3, 4):

1. $m = \frac{3(17) - 6(7)}{3(14) - 6^2} = \frac{51 - 42}{42 - 36} = \frac{9}{6} = \vC{1.5}$

2. $b = \frac{7 - 1.5(6)}{3} = \frac{7 - 9}{3} = \vC{-0.667}$ — so the fitted line is $y = 1.5x - 0.667$, with a correlation coefficient of about $0.98$.

Key Factors to Consider

• A strong correlation coefficient doesn’t necessarily mean x actually causes y. Two variables can be strongly correlated because one genuinely influences the other, because both are driven by a third factor, or purely by coincidence in a small dataset — correlation describes the strength of a linear relationship, not the reason behind it.
• This method assumes the true relationship is linear, which isn’t always the right assumption. If a scatter of points actually follows a curve, forcing a straight line through it can produce a technically “best fit” line that still describes the data poorly — plotting the points visually before trusting the fit is good practice.
• A single unusual point (an outlier) can disproportionately pull the fitted line toward it. Because least squares minimizes SQUARED distances, a point far from the general trend contributes much more to the fit than a point close to it — a small dataset with one extreme outlier is especially sensitive to this.
• More data points generally produce a more reliable fit, all else equal. A line fit through only 3-4 points (as in the worked example) can look convincing but carries much more uncertainty than the same relationship measured across dozens of points — correlation strength alone doesn’t capture how much data actually backs up the fit.

Common Mistakes

• Treating a high correlation coefficient as proof of causation. A strong r-value only shows the points cluster tightly around a line — it says nothing about whether one variable actually causes changes in the other.
• Fitting a straight line to data that’s actually curved. Least squares always returns a slope and intercept, even when a linear fit badly describes the underlying pattern — plot the points first, or check whether the correlation coefficient is surprisingly low, before trusting the result.
• Using too few data points and treating the fit as reliable. A line through just 3-4 points can look convincing but carries far more uncertainty than the same relationship measured across dozens of points.
• Letting a single outlier dominate the fit without noticing. Because least squares minimizes squared distances, one far-off point can pull the whole line toward it — worth checking whether removing an unusual point changes the result substantially.

Useful to Know

• Need to estimate a value between two known points instead of fitting a trend line through many? The Linear Interpolation Calculator calculator covers that more direct case.
• Want to see how one specific value compares to the overall spread of a dataset? The Z-Score Calculator calculator measures exactly that.
• Working with the same list of numbers for broader summary statistics? The Statistics Calculator calculator covers mean, median, and standard deviation.