Alternative codings for X

If the values of a numerical explanatory variable, X, that are used in an experiment are evenly spaced, the relationship can be equivalently modelled in terms of any recoding of the explanatory variable with evenly spaced values. Recoding may simplify data entry.

The simplest recoding of X replaces the actual values with 1, 2, ... .

If recoded variables are used, the fit of the model remains the same, but the numerical values of the parameter estimates change.

Coding for factors with two levels as ±1

If a numerical or categorical factor has two levels, the levels are often coded as numerical values 0 and 1. However there are advantages to coding them as '-1' and '+1' in some types of experiment that we will examine in later chapters.


Silicon wafer manufacture

The diagram below shows one of the data sets on the previous page.

Use the pop-up menu to recode the explanatory variable. Observe that the model has the same flexibility and can therefore describe explained variability in the data for all recodings.