Read a coefficient in its stated units

THE IDEA

Read the example.

For a linear model with other inputs fixed, coefficient w gives the change in prediction per unit change in x. Its numerical size depends on the feature units. The same relationship can have different coefficients after rescaling.

Change in prediction = w × change in x

Use it well

A coefficient is not automatically causal, and a large magnitude does not by itself establish feature importance. Correlated inputs and uncertainty affect interpretation.