How to interpret a regression coefficient

Example resource
This page demonstrates the resource format. Replace or adapt it before treating it as a complete teaching note.

Regression output is only useful when a numerical coefficient is translated back into the language of the question. A strong interpretation identifies the variables, uses their units, states what is held constant, and avoids implying causation unless the study design supports it.

Begin with the model

For a simple linear regression,

$$\widehat{y} = \beta_0 + \beta_1x.$$

The slope $\beta_1$ describes the expected change in the response variable $y$ associated with a one-unit increase in the explanatory variable $x$.

In a multiple regression, the same idea applies after accounting for the other variables in the model.

A worked example

Suppose a model relates weekly study time to an examination score:

$$\widehat{\text{score}} = 58.4 + 2.1(\text{hours studied}).$$

The estimated slope is 2.1 marks per hour.

A careful interpretation is:

For each additional hour studied per week, the model predicts an average increase of 2.1 marks in examination score.

If this were one coefficient in a multiple regression, add the relevant condition:

After accounting for the other variables in the model, each additional hour studied per week is associated with an estimated 2.1-mark increase in examination score, on average.

Four checks before you write

Check	Question to ask	Why it matters
Direction	Is the coefficient positive or negative?	This determines whether the response is expected to rise or fall.
Magnitude	What change does one unit represent?	A coefficient without units is difficult to interpret.
Context	Which population and variables are being studied?	Generic wording can make a correct calculation scientifically vague.
Claim strength	Does the design justify causal language?	Association alone does not establish causation.

A reusable sentence structure

Use the structure below as a starting point, then edit it so it sounds natural:

Holding the other variables in the model constant, a one-[unit] increase in [explanatory variable] is associated with an estimated [coefficient and response unit] [increase/decrease] in [response variable], on average.

The phrase “holding the other variables constant” describes the model comparison. It does not mean that a real person or system can always change one variable while everything else remains fixed.

Common errors

• Reporting only the sign or p-value and never explaining the size of the estimated effect.
• Reversing the response and explanatory variables.
• Forgetting that a transformed predictor changes the interpretation of a one-unit increase.
• Treating an observational association as proof that changing $x$ will cause $y$ to change.
• Writing “increases by” when the coefficient is an estimate with uncertainty.

Check the coefficient in R

model <- lm(score ~ hours_studied, data = students)

coef(model)
confint(model)

The first command returns the fitted coefficients. The confidence interval helps communicate the precision of the slope estimate; it should not be replaced by a binary statement about statistical significance.

Final checklist

Before submitting an interpretation, check that it contains:

1. the explanatory variable and its unit;
2. the response variable and its unit;
3. the direction and magnitude of the estimate;
4. the appropriate conditional language for multiple regression; and
5. wording that matches the study design.