MerQuR+Statistics Glossary

Statistics Glossary → Coded factors (response surface)

Coded factors (response surface)concept

In response-surface and experimental-design analyses, factors converted from their original units to a common scale, usually running from -1 to +1.

Coding moves each factor's midpoint to zero and its extremes to -1 and +1: code = (x - M) / S, where M = (max + min) / 2 is the midpoint and S = (max - min) / 2 the half-range. For a temperature that varies between 150 and 200 °C, for instance, 175 °C corresponds to zero, 150 °C to -1 and 200 °C to +1. Factors measured in different units (temperature, time, pressure) are thus brought to the same scale and their coefficients can be compared directly: which factor affects the response most can be read from the size of the coefficients.

Another benefit of coding is that it reduces the correlation of quadratic and interaction terms with the main effects, which makes the coefficient estimates more stable and interpretable. In the coded model, the intercept is the predicted response at the midpoint of all factors, and the linear coefficients give the slopes at that point. Coding does not change the predictions themselves: the same model written in original units gives identical predictions, and only the numerical values and meaning of the coefficients differ. When reporting results and interpreting the optimum conditions (the stationary point), coded values must be converted back to original units.

An everyday example

A baker studies how a cake rises as a function of oven temperature (160-200 °C) and baking time (20-40 minutes). In original units the temperature coefficient is 'per degree' and the time coefficient 'per minute', so they cannot be compared directly. After coding, both factors run from -1 to +1, so if temperature's coefficient comes out at 0.8 and time's at 0.3, it is immediately clear that temperature has the larger effect on rising.

In MerQur

It is chosen with the 'Coding (SAS (midrange / half-range), No coding (original units))' parameter in Response Surface Regression (RSREG); with the default coding the factors are brought into the -1..1 range.

Analyses where this concept appears

Response Surface Regression (RSREG)

Related concepts

Ridge analysis (response surface)Stationary point and canonical analysis (response surface)