Statistical Methods

Construct and Interpret Orthonormal Helmert Contrasts

Define the task

Helmert contrasts provide coordinates for differences among an ordered list of group summaries. They are useful for checking a contrast basis before fitting a model or interpreting a transformation. Fix the category order first: changing that order changes which comparison each row represents. The tool constructs an orthonormal basis and multiplies it by supplied summaries. It does not estimate sampling uncertainty, adjust for unequal group sizes or perform an omnibus test.

Carry out the check

For row j starting at one, assign 1 divided by square root of j(j+1) to the first j positions, −j divided by that denominator to the next position, and zero afterward. Each row sums to zero and has unit length; different rows are orthogonal. This convention compares earlier categories with the next one. Other software can reverse signs or use different scaling without changing the contrast subspace. Preserve coefficients rather than relying only on a contrast name.

Worked hypothetical example

For hypothetical summaries (2,4,8), the rows are (1,−1,0)/sqrt(2) and (1,1,−2)/sqrt(6). Coordinates are −sqrt(2) and −10/sqrt(6). Adding the same constant to all three summaries leaves both coordinates unchanged. Equal numerical weights do not imply equal precision when group sizes differ. To obtain a standard error, a separate covariance matrix and a design-appropriate model are needed. Retain category labels next to coefficient columns when exporting results.

Method checked 19 November 2026 against Primary method documentation: helmert.

Open the helmert contrast coordinate audit.

Download the helmert coefficient mapping worksheet.

Related reading

Reading path: Analyze data. Guide 175 of 182. Choose any guide to suit your project.

Statistical Methods

Check an Analytic Gradient Before Trusting an Optimizer

Practical steps and a hypothetical example: Audit a supplied analytic gradient at several parameter values.

Statistical Methods

Distinguish an Expectile from a Percentile

Practical steps and a hypothetical example: Distinguish asymmetric squared-loss location from a quantile.

Browse this reading pathTools for this stageGuides and downloads for this stage