When the Predictor Has No Variation

Introduction

The least-squares slope describes how a fitted response changes as the predictor changes. That description requires the observed predictor values to provide at least some change. If every observation has exactly the same predictor value, the slope formula divides by zero. The result is not a slope of zero; the slope is undefined from these data.

This distinction matters when calculating or interpreting a regression line. A zero slope is a calculable value that describes a horizontal fitted line. An undefined slope signals that the data contain no predictor variation from which to calculate a slope at all.

The Slope Formula and Its Condition

Suppose we have $n$ observed pairs $(x_i,y_i)$, where $x_i$ is the predictor value and $y_i$ is the response value for observation $i$. A least-squares line has fitted response $\hat{y}_i=b_0+b_1x_i$; least squares chooses $b_0$ and $b_1$ to minimize the sum of squared residuals. Here, $b_1$ denotes the calculated slope, and $\bar{x}$ and $\bar{y}$ are the sample means of the observed predictor and response values.

Definition: When the predictor values are not all identical, the least-squares slope is $$ b_1=\frac{\sum_{i=1}^{n}(x_i-\bar{x})(y_i-\bar{y})}{\sum_{i=1}^{n}(x_i-\bar{x})^2}. $$ The denominator is the sum of squared deviations of the predictor values from their mean. It must be greater than zero for this formula to calculate a slope.

The formula is a descriptive calculation from the observed pairs. No probability distribution assumption is needed to check whether its denominator is zero. In particular, if all predictor values equal the same number $c$, then their mean is also $c$, so each deviation $x_i-\bar{x}$ equals zero. Consequently, every term in the denominator is zero. The numerator is also zero because each of its terms contains the factor $x_i-\bar{x}=0$. The formula becomes $0/0$, which has no defined numerical value.

The denominator is a sum of squares, so it cannot be negative. It equals zero exactly when every squared deviation is zero; that is, exactly when every $x_i=\bar{x}$. Thus, checking for a zero denominator is equivalent to checking whether all observed predictor values are identical.

Worked Examples

Worked Example: All Predictor Values Are Identical

Consider hypothetical observations with predictor values $2,2,2,2$ and response values $5,7,6,10$. Calculate the pieces of the slope formula and decide whether a slope can be calculated.

The predictor mean is $$ \bar{x}=\frac{2+2+2+2}{4}=2, $$ and the response mean is $$ \bar{y}=\frac{5+7+6+10}{4}=7. $$ For every observation, $x_i-\bar{x}=2-2=0$. Therefore, the numerator is $$ \sum_{i=1}^{4}(x_i-\bar{x})(y_i-\bar{y}) =0(-2)+0(0)+0(-1)+0(3)=0, $$ and the denominator is $$ \sum_{i=1}^{4}(x_i-\bar{x})^2=0^2+0^2+0^2+0^2=0. $$

Substitution gives $b_1=0/0$, which is undefined. The response values vary, but that variation does not supply any variation in $x$ with which to calculate a slope. It would be incorrect to report $b_1=0$.

Worked Example: A Constant Predictor With a Changing Response

A hypothetical data set records a predictor value of $15$ for each of three observations, with responses $31,29,35$. Find the numerator and denominator in the slope formula.

The means are $$ \bar{x}=\frac{15+15+15}{3}=15,\qquad \bar{y}=\frac{31+29+35}{3}=\frac{95}{3}\approx31.67. $$ Each predictor deviation is $15-15=0$. Hence, $$ \sum_{i=1}^{3}(x_i-\bar{x})(y_i-\bar{y})=0, \qquad \sum_{i=1}^{3}(x_i-\bar{x})^2=0. $$ The slope expression is again $0/0$, so the slope cannot be calculated. In particular, the fact that the responses $31,29,35$ differ does not change the denominator. Slope calculation requires variation in the predictor, not merely variation in the response.

Worked Example: A Predictor With Small but Nonzero Variation

Consider hypothetical pairs with predictor values $1.00,1.01,1.02$ and response values $4,5,7$. Determine whether the slope is defined and calculate it.

The means are $\bar{x}=1.01$ and $\bar{y}=16/3\approx5.333$. The predictor deviations are $-0.01,0,0.01$, while the response deviations are $-4/3,-1/3,5/3$. Thus the numerator is $$ (-0.01)(-4/3)+(0)(-1/3)+(0.01)(5/3) =0.01333+0+0.01667=0.03000, $$ and the denominator is $$ (-0.01)^2+0^2+(0.01)^2=0.0001+0+0.0001=0.0002. $$ Therefore, $$ b_1=\frac{0.03000}{0.0002}=150. $$ The slope is defined because the denominator is positive. It has units of response units per predictor unit. The predictor values are close together, however, so a small range of predictor values accompanies this calculated slope. This example contrasts with a constant predictor: small variation is not the same as no variation.

Why a Slope Cannot Be Identified From a Constant Predictor

The zero denominator is the immediate obstacle in the formula. The least-squares criterion also shows why the data cannot choose one slope in this situation. Suppose $x_i=c$ for every observation. For any proposed coefficients $b_0$ and $b_1$, the fitted value for every observation is the same number $m=b_0+b_1c$. The sum of squared residuals is therefore $$ \sum_{i=1}^{n}(y_i-m)^2 =\sum_{i=1}^{n}(y_i-\bar{y})^2+n(m-\bar{y})^2. $$ To verify this equality, write $y_i-m=(y_i-\bar{y})+(\bar{y}-m)$, square, and sum. The cross-product term is zero because $\sum_{i=1}^{n}(y_i-\bar{y})=0$; the squared constant term contributes $n(\bar{y}-m)^2$.

The second term on the right is nonnegative and is smallest when $m=\bar{y}$. Any coefficients satisfying $b_0+b_1c=\bar{y}$ attain that minimum. For every choice of $b_1$, one can choose $b_0=\bar{y}-b_1c$ to satisfy this condition. So the least-squares criterion does not select a unique slope: many coefficient pairs give the same minimum sum of squared residuals. This is a mathematical reason, beyond the division by zero, that the data do not determine a slope.

Worked Example: Many Slopes Give the Same Minimum

Return to the four observations with $x_i=2$ and responses $5,7,6,10$. Their response mean is $\bar{y}=7$. For any chosen slope $b_1$, set $$ b_0=7-2b_1. $$ Then, at the only observed predictor value, the fitted value is $b_0+2b_1=7$. For example, $b_1=0$ gives $b_0=7$, while $b_1=3$ gives $b_0=1$. Both give fitted values of $7$ at every observed predictor.

For either choice, the residuals are $5-7=-2$, $7-7=0$, $6-7=-1$, and $10-7=3$. The sum of their squares is $$ (-2)^2+0^2+(-1)^2+3^2=4+0+1+9=14. $$ Thus these different slopes give the same least-squares value. This does not make the slope zero; it shows that the observations cannot distinguish among slopes.

Undefined Slope Is Not Zero Slope

When a least-squares slope is zero and is calculable, the denominator in the centered-data formula is positive while the numerator is zero. A zero slope is therefore a specific numerical result. When all predictor values are identical, both numerator and denominator are zero, so the formula does not yield any numerical slope. Confusing these cases turns “not calculable” into a false numerical claim.

Situation	Denominator	Conclusion about the slope
Predictor values vary and numerator is zero	Positive	The calculated slope is zero
All predictor values are identical	Zero	The slope formula is undefined
Predictor values vary and numerator is nonzero	Positive	The calculated slope is nonzero

A constant predictor also does not show that the response is unrelated to the predictor in every possible setting. It shows a more basic limitation: within these observed data, there are no differing predictor values to compare. The slope is not estimable by the simple least-squares formula from this data set.

Common Mistakes

Calling $0/0$ a zero slope. Zero divided by a positive number equals zero; $0/0$ is undefined. Check the denominator before interpreting the numerator.
Assuming response variation is enough. A range of response values does not prevent a zero denominator if every predictor value is the same.
Using “no predictor variation” to claim there is no association. The data contain no predictor contrast with which to calculate a slope. That is not evidence that a calculable slope equals zero.
Thinking a unique least-squares line must always exist. With a constant predictor, multiple intercept-slope pairs attain the same minimum squared error. The observed data do not determine a unique slope.
Treating a small denominator as zero without checking. A positive denominator means the formula is defined, even if the predictor values are close. The calculation may merit caution, but it is not the same as division by zero.

Key takeaway: The least-squares slope formula requires $\sum_{i=1}^{n}(x_i-\bar{x})^2>0$. If every predictor value is identical, this sum is zero and the formula becomes $0/0$. The slope is undefined, not zero.

Check Your Understanding

Use the centered-data slope formula to reason through each question.

If all predictor values equal $8$, what are $\bar{x}$ and the denominator in the slope formula?
Can a data set have varying response values and still have an undefined least-squares slope? Explain why.
What is the difference between a zero slope and a slope that is undefined because the predictor is constant?
If the predictor values are $3,3,3.1$, is the slope formula defined? State the condition that determines your answer.
When every predictor value equals $c$, why can different slopes be paired with intercepts to give the same fitted value at the observed predictor?