Which Point Is Farthest From The Line Of Best Fit

Understanding which point is farthest from the line of best fit is essential for anyone working with bivariate data, because that single observation can dramatically influence the slope, intercept, and overall interpretation of a regression model. In this article we will walk through the statistical reasoning behind the question, show step‑by‑step calculations, illustrate the concept with a concrete example, and address the most frequent misunderstandings that arise when dealing with outliers in scatter plots.

What Is a Line of Best Fit?

Definition and Purpose

The line of best fit, often called the regression line or trend line, is a straight line that summarizes the relationship between two quantitative variables. It is chosen so that the sum of the squared vertical distances between each data point and the line is minimized. This criterion is known as the least squares method.

The Equation In its simplest form, the line can be expressed as

[ \hat{y}=b_0+b_1x ]

where (b_0) is the intercept, (b_1) is the slope, (x) is the independent variable, and (\hat{y}) is the predicted dependent variable. The slope (b_1) is calculated as

[ b_1=\frac{\sum{(x_i-\bar{x})(y_i-\bar{y})}}{\sum{(x_i-\bar{x})^2}} ]

and the intercept (b_0) follows from

[ b_0=\bar{y}-b_1\bar{x} ]

where (\bar{x}) and (\bar{y}) are the sample means of the (x) and (y) variables, respectively.

Visual Interpretation

When plotted, the line of best fit divides the cloud of points into a pattern that suggests a linear trend. Still, not every point will sit exactly on the line; some will lie above, others below, and a few may be positioned at a noticeable distance from it. Identifying which point is farthest from the line of best fit helps analysts spot potential anomalies or influential observations.

How to Determine the Farthest Point

Step‑by‑Step Procedure 1. Compute the regression coefficients (b_0) and (b_1) using the formulas above.

Calculate the fitted value (\hat{y}_i) for each observation (x_i) using (\hat{y}_i = b_0 + b_1x_i).
Find the residuals (e_i = y_i - \hat{y}_i). These represent the vertical distances from each point to the line. 4. Square each residual to obtain (e_i^2).
Determine the absolute distance from the line by taking the square root of the squared residual, or simply compare the absolute residual values (|e_i|).
Identify the maximum absolute residual; the corresponding observation is which point is farthest from the line of best fit.

Why Use Absolute Residuals? Because the regression line minimizes the sum of squared residuals, the point with the largest absolute residual is the one that deviates most strongly in the vertical direction. While the squared residuals are used for fitting, the raw residuals are convenient for locating the outlier.

Example Calculation

Suppose we have the following five data points:

(x)	(y)
1	2
2	3
3	5
4	4
5	6

Compute (\bar{x}=3) and (\bar{y}=4).
Calculate (b_1 = \frac{(1-3)(2-4)+(2-3)(3-4)+(3-3)(5-4)+(4-3)(4-4)+(5-3)(6-4)}{(1-3)^2+(2-3)^2+(3-3)^2+(4-3)^2+(5-3)^2

A Deeper Look: take advantage of and Influence

While the absolute residual tells us which observation is vertically farthest from the fitted line, it does not capture the whole picture of an observation’s impact on the regression model. Two additional concepts are worth mentioning:

Concept	What it measures	Why it matters
put to work	How far an (x)‑value lies from the mean (\bar{x}). That's why formally, the use of observation i is (h_i = \frac{1}{n} + \frac{(x_i-\bar{x})^2}{\sum_{j=1}^{n}(x_j-\bar{x})^2}).	Points with high take advantage of can pull the regression line toward themselves, even if their residuals are modest. Which means
Influence	The combined effect of a large residual and high put to work. On top of that, a common metric is Cook’s distance (D_i = \frac{e_i^2}{p,\hat{\sigma}^2} \times \frac{h_i}{(1-h_i)^2}), where (p) is the number of parameters (2 for simple linear regression) and (\hat{\sigma}^2) is the mean squared error.	An observation with a large Cook’s distance can substantially change the estimated coefficients if it were removed.

In practice, the “farthest point” identified by the largest (|e_i|) is often also the most influential, but not always. A point with a moderate residual but extreme (x) value can have a higher Cook’s distance than the point with the largest vertical distance.

Want to learn more? We recommend x 2 x 2 y 2 and who sank the boat activities for further reading.

Quick diagnostic checklist

Plot residuals vs. fitted values – look for patterns or outliers.
Compute apply values – flag any (h_i) greater than (2\frac{p}{n}) (a common rule of thumb).
Calculate Cook’s distance – observations with (D_i > 4/(n-p)) deserve closer scrutiny.
Re‑fit the model without the suspect point(s) – compare coefficients, (R^2), and predictive performance to see how sensitive the model is.

Extending to Multiple Linear Regression

When more than one predictor is present, the same principles apply, but the geometry changes:

Residuals are still the vertical distances from observed (y_i) to the hyperplane defined by (\hat{y}i = b_0 + b_1x{i1} + \dots + b_px_{ip}).
apply becomes a function of the entire row vector (\mathbf{x}_i) and is derived from the hat matrix (H = X(X^\top X)^{-1}X^\top). The diagonal elements (h_i) again quantify how “far” each observation is in the multidimensional predictor space.
Cook’s distance retains its form, now using the multivariate residuals and the same make use of values.

Thus, the procedure for locating the farthest point in a multiple‑regression context is identical: compute residuals, take absolute values, and locate the maximum. That said, always complement this with use and Cook’s distance diagnostics to ensure you are not missing a high‑take advantage of, moderate‑residual point that could be more damaging to model integrity.

Practical Tips for Implementation

Situation	Recommended Tool	Reason
Small data sets (≤ 30 points)	Hand calculations or a spreadsheet	Transparent learning of the mechanics
Medium‑size data (30–10,000 points)	R (`lm()`, `rstandard()`, `cooks.In practice, distance()`) or Python (`statsmodels. api.

Example in Python (statsmodels)

import pandas as pd
import statsmodels.api as sm

# Assume df contains columns 'x' and 'y'
X = sm.add_constant(df['x'])          # adds intercept term
model = sm.OLS(df['y'], X).fit()

# Residuals and use
df['residual'] = model.resid
df['abs_resid'] = df['residual'].abs()
df['use'] = model.get_influence().hat_matrix_diag
df['cooks_d']   = model.get_influence().cooks_distance[0]

# Identify the farthest point
farthest_idx = df['abs_resid'].idxmax()
print("Farthest point (by absolute residual):")
print(df.loc[farthest_idx])

The output will show the observation with the largest vertical deviation, along with its make use of and Cook’s distance, giving you a complete diagnostic snapshot.

Concluding Thoughts

Identifying the point farthest from the line of best fit is a straightforward yet powerful first step in regression diagnostics. By:

Computing the regression coefficients,
Deriving fitted values and residuals,
Comparing absolute residuals to locate the maximum,

you quickly surface observations that may be data entry errors, measurement anomalies, or genuine outliers worth investigating further. Even so, remember that vertical distance alone does not capture an observation’s full influence. Complement the residual analysis with put to work and Cook’s distance to guard against points that, because of their extreme predictor values, could disproportionately sway the estimated relationship.

In practice, a disciplined workflow—visual inspection, residual‑apply plots, and influence metrics—ensures that the final model rests on a solid, trustworthy foundation. On the flip side, whether you are fitting a simple two‑variable line or a high‑dimensional regression surface, the same principles apply: detect the outliers, assess their impact, and decide whether to keep, transform, or exclude them. By doing so, you preserve the integrity of your inference and improve the predictive performance of your model.