What do A, C and E stand for?

A represents additive genetic variance, C shared environmental variance, and E nonshared environmental variance including measurement error in the basic model.

Does ACE directly measure specific genes?

No. The classical model infers broad components from patterns of family resemblance. Identifying particular genetic variants is a different research task.

Can all genetic and environmental components be estimated from any twin dataset?

No. Identifiability depends on the design and assumptions. Additional components may require more informative family relationships, measurements or other data.

AI Glossary / Statistics and probability / Intermediate

ACE model

Definition

The ACE model is a statistical model that partitions variation in an observed trait into additive genetic variance (A), shared environmental variance (C) and nonshared environmental variance (E). In the basic model, E also includes measurement error. It is commonly fitted to twin or related-family data under explicit assumptions about genetic and environmental similarity.

What it means

ACE describes variation across people in a studied population, rather than dividing one person's trait into genetic and environmental percentages. For example, an estimated additive genetic proportion concerns how much of the observed variation is attributed to that component under the model.

The letters identify latent variance components, not a list of measured genes or particular experiences. A shared influence is one that contributes to similarity between relatives in the design; a nonshared influence contributes to their differences. The estimates depend on the population, measurement and model assumptions and should not be treated as immutable properties of a trait.

How it works

In the classical twin design, researchers compare monozygotic twins and dizygotic twins. The basic model represents their additive genetic correlations as one and one-half, respectively. Shared environmental components are modeled as correlated across twins in both groups, while nonshared components are uncorrelated across twins.

Writing the variance components as VA, VC and VE gives total variance VP = VA + VC + VE. Expected within-pair covariance is VA + VC for monozygotic twins and 0.5VA + VC for dizygotic twins under the usual assumptions. These patterns help identify the components.

Researchers often fit the model using structural equation methods and report uncertainty intervals and fit diagnostics. Assumptions include appropriate genetic relatedness and comparable trait-relevant environmental similarity across twin types. Nonadditive genetic effects, assortative mating, gene-environment relationships and measurement problems can complicate interpretation or require richer models and designs.

Three practical examples

1. A research team uses twin data to examine the sources of variation in a measured physical trait, reporting model assumptions and uncertainty rather than individual predictions.

2. A methods researcher compares estimation procedures for ACE variance components using simulated data with known population parameters.

3. A study designer considers repeated measurements to help distinguish measurement error from other nonshared variation, which the simplest ACE model combines.

Think of it this way

Think of studying why measurements vary across a collection of related pairs. The model uses patterns of similarity to allocate variance to broad statistical components. It does not open each person's measurement and label separate pieces as genes, family environment and personal experience.

Worked example

Consider an original simplified example with standardized trait variance equal to one. Suppose the monozygotic twin correlation is 0.70 and the dizygotic correlation is 0.45. Under the basic ACE equations, rMZ = A + C and rDZ = 0.5A + C, where A, C and E now denote proportions of total variance.

Subtracting gives 0.25 = 0.5A, so A = 0.50. Then C = 0.70 - 0.50 = 0.20, and E = 1 - 0.50 - 0.20 = 0.30. Substitution checks the second correlation: 0.5(0.50) + 0.20 = 0.45.

These are illustrative population-style calculations, not a full statistical fit to real observations. Sampling uncertainty and assumption checks remain necessary. The value 0.50 does not mean that half of any individual's trait is genetic, and E = 0.30 does not identify which experiences or measurement errors produced that variance.

Illustrative case study

Fictional case study: Pairwise Research analyses a continuous measurement from a twin registry. Its first report describes E as entirely attributable to personal experiences and presents the A estimate as a fixed property of the trait.

An internal review corrects both statements. The basic E component includes measurement error, and the estimated proportions refer to the studied population and measurement conditions. The team adds uncertainty intervals, documents recruitment and measurement procedures, and examines whether the model adequately represents the observed twin similarities.

This fictional case illustrates interpretation and reporting. It does not assert that the hypothetical data satisfy the ACE assumptions or establish a causal explanation for any named environmental factor.

Common misconceptions and mistakes

Heritability is not destiny: a variance proportion does not establish whether an intervention can change a trait. It also does not explain differences in population means or justify conclusions about individuals.

Shared environment does not mean every experience occurring in the same household, because experiences can affect siblings differently. Nonshared environment in the basic model is not a pure measurement of unique experiences, because measurement error contributes too. Finally, a good numerical fit does not prove that every causal assumption is correct or identify the specific genes and environmental mechanisms involved.

Business use

For most businesses, ACE is primarily relevant when interpreting research or building tools that support family-based statistical studies. It is not a general-purpose customer-segmentation model or a method for assigning individual ability from ancestry.

Analysts should report the study population, phenotype definition, design, modeling assumptions and uncertainty. Keep variance decomposition separate from prediction and intervention claims. Before reusing a published estimate, consider whether the population and measurement conditions match the new question. Changes in environmental variation or measurement reliability can alter component proportions without implying that biology has fundamentally changed.

Check your understanding

An ACE analysis estimates A = 0.4 for a particular measured trait in one population. Does that mean 40% of each person's trait is genetically caused?

Answer: No. The estimate concerns a proportion of variation across people under that model and study setting. It is not an individual-level partition or a complete causal account.

Related terms

Variance components
heritability
variance component
twin study
monozygotic twins
dizygotic twins
structural equation model
measurement error and equal-environments assumption

Learn next

derive expected twin covariances from selected A, C and E values, then examine how uncertainty in observed correlations affects the estimates. Compare the basic model with designs that explicitly represent repeated-measurement error or additional family relationships.

Sources

Primary sources: OpenMx, Genetic Epidemiology, Path Specification, classical ACE twin model: https://openmx.ssri.psu.edu/docs/OpenMx/2.0.0-4004/GeneticEpi_Path.html

Arbet and colleagues, A Robust and Unified Framework for Estimating Heritability in Twin Studies using Generalized Estimating Equations: https://arxiv.org/abs/1710.09326

National Research Council, Education, Fertility, and Heritability, discussion of ACE components and measurement error: https://www.ncbi.nlm.nih.gov/books/NBK97282/

Previous: Accuracy barrier Next: ACID transactions