MerQuR+Statistics Glossary

Statistics Glossary → Standardized regression coefficient (β)

Standardized regression coefficient (β)concept

The coefficient obtained when all variables are expressed in standard deviation units; it shows how many standard deviations the outcome changes for a one-standard-deviation increase in the predictor.

A raw regression coefficient (b) depends on the units of the variables: entering income in thousands rather than in single units changes the coefficient a thousandfold. The standardized coefficient β = b · sₓ / s_y removes the units; the same result is obtained by converting every variable to z scores and refitting the regression. Effects of predictors measured in different units can then be seen on one scale.

β depends on the standard deviations in the sample; the same true effect can give different β values in two samples where the predictor is spread differently. When predictors are strongly correlated, the size of the β values should not be read as 'relative importance'. For a binary (0/1) predictor, a one-standard-deviation change is not a meaningful unit, so the raw coefficient is easier to interpret.

An everyday example

House prices are predicted from floor area and distance to the city centre. The raw coefficients, such as '20 000 per square metre' and '−50 000 per kilometre', are in different units and cannot be compared directly. If the standardized coefficients are 0.55 and −0.30, a one-standard-deviation increase in size is more strongly related to price than a one-standard-deviation increase in distance.

In MerQur

'Standardized coefficients (β)' in Multiple Linear Regression; 'Standardized coefficients (β*)' with weighted standard deviations in Weighted Linear Regression; 'Standardized β + HDI' in Bayesian Linear Regression; and the standardized path coefficients of Path Analysis.

Analyses where this concept appears

Multiple Linear RegressionWeighted Linear Regression (Survey Regression)Bayesian Linear RegressionPath AnalysisTime-dependent Cox Regression

Related concepts

Regression coefficient (B) and parameter estimatesz-score (standardization)