Why Combine Genotypes Into a Carrier Group?
A person with one copy of a variant allele and a person with two copies can be grouped together when the question concerns the mean trait difference between people who carry the variant and people who do not. This is called dominant coding: the heterozygote and variant-allele homozygote receive the same code, while the reference homozygote receives a different code. The coding is useful when that shared carrier contrast is the effect of interest, but it also imposes a specific pattern on the genotype means.
Here, the outcome is a quantitative trait, such as a measured concentration or length. Let \(G_i\) denote the number of copies of the variant allele for individual \(i\): \(G_i=0\) for a reference homozygote, \(G_i=1\) for a heterozygote, and \(G_i=2\) for a variant-allele homozygote. Dominant coding turns these three dosage values into a binary carrier indicator.
The model’s key feature is not merely that there are two codes. It says that the two carrier genotypes share one expected trait mean. In terms of genotype-specific means, it imposes
The coding therefore targets a particular comparison: carriers as a combined group versus reference homozygotes. It is not a model of a one-copy effect and a separate two-copy effect. The word “dominant” names this coding pattern; by itself it does not establish a biological mechanism.
Building and Reading the Model
For a quantitative trait, the simple dominant-coded mean model is
Here \(Y_i\) is the trait value for individual \(i\), \(\beta_0\) is the intercept, \(\beta_D\) is the carrier coefficient, and \(\varepsilon_i\) is the individual’s deviation from the mean specified by the model. The zero-mean condition ensures that the model’s conditional means have the stated interpretation. The outcome’s units determine the units of both coefficients.
For estimating the two group means descriptively, no normality assumption is needed: calculate the mean for \(D=0\) and the mean for \(D=1\). To use conventional small-sample regression inference for this model, additional assumptions are needed, including independent observations and an appropriate error distribution; a normal, constant-variance error model gives the usual exact small-sample regression inference. Those inferential assumptions are separate from the definition of the coding and the meaning of its coefficient.
In a sample, let \(\bar{Y}_0\) be the mean trait value among reference homozygotes and \(\bar{Y}_1\) the mean among carriers. When both groups are observed, fitting this two-group model by least squares gives
Thus the fitted carrier mean is \(\hat{\beta}_0+\hat{\beta}_D=\bar{Y}_1\). A positive coefficient means the observed carrier mean is higher; a negative coefficient means it is lower. The coefficient is a difference in trait units, not a percentage or a per-allele effect.
Worked Example: Assigning the Carrier Indicator
Suppose a hypothetical sample has genotypes recorded by variant-allele dosage. The trait is a laboratory measurement in milligrams per liter. Assign the dominant code to each genotype.
| Variant-allele dosage \(G_i\) | Genotype group | Carrier indicator \(D_i\) |
|---|---|---|
| 0 | Reference homozygote | 0 |
| 1 | Heterozygote | 1 |
| 2 | Variant-allele homozygote | 1 |
The rule is \(D_i=I(G_i\geq1)\). Applying it gives \(I(0\geq1)=0\), \(I(1\geq1)=1\), and \(I(2\geq1)=1\). For example, the dosage sequence \(0,2,1,0,1\) becomes the carrier-indicator sequence \(0,1,1,0,1\).
The original dosage distinguishes one from two variant copies; the dominant indicator deliberately does not. The resulting coefficient will compare the combined \(G=1\) and \(G=2\) group with the \(G=0\) group. It cannot, on its own, report separate one-copy and two-copy contrasts.
Why the Coefficient Is a Difference in Group Means
The group-mean interpretation follows directly from least squares. Write the residual sum of squares as the sum of squared deviations in the reference group and the carrier group. The fitted mean for the reference group is \(\beta_0\); the fitted mean for the carrier group is \(\beta_0+\beta_D\). For any group of observations, the sum of squared deviations is minimized when its fitted value equals that group’s sample mean. Consequently, the two fitted group means must equal \(\bar{Y}_0\) and \(\bar{Y}_1\), respectively.
Equating fitted and observed group means gives \(\hat{\beta}_0=\bar{Y}_0\). For carriers, it gives \(\hat{\beta}_0+\hat{\beta}_D=\bar{Y}_1\). Subtracting the first equation from the second yields the coefficient formula:
This justification also explains why each carrier contributes to one combined mean: the model gives every carrier the same fitted mean, regardless of whether \(G_i=1\) or \(G_i=2\). When other predictors are added to a model, the coefficient’s interpretation depends on that model’s specification; the simple difference-in-means identity applies to the two-group model above.
Worked Example: Estimating a Carrier Mean Difference
In a hypothetical sample, four reference homozygotes have trait measurements of 10, 12, 11, and 13 mg/L. Five carriers, combining heterozygotes and variant-allele homozygotes, have measurements of 14, 15, 13, 16, and 14 mg/L. Calculate and interpret the dominant-coded coefficients.
The reference mean is
The combined carrier mean is
Therefore, the fitted intercept and carrier coefficient are
The fitted mean is 11.5 mg/L for reference homozygotes and \(11.5+2.9=14.4\) mg/L for carriers. In this hypothetical sample, carriers have a mean measurement 2.9 mg/L higher than reference homozygotes. This is a sample mean contrast; it does not by itself establish a population difference or a causal effect.
What Combining the Carrier Genotypes Assumes
Dominant coding is appropriate when the target contrast is “at least one variant allele” versus “no variant allele,” and when a common carrier mean is a reasonable model for the question. Because the model equates the expected values for \(G=1\) and \(G=2\), it may be a poor description if those two means differ substantially. A combined coefficient can still be calculated, but it then summarizes the selected carrier group rather than proving that the two genotype-specific means are equal.
The combined carrier mean is weighted by the numbers of individuals in its two genotype groups. It is not generally the unweighted average of the heterozygote mean and the variant-homozygote mean. A larger genotype group contributes more observations to the combined mean. This matters especially when the variant-allele homozygotes are uncommon.
Worked Example: A Weighted Mean for the Combined Carriers
Consider another hypothetical sample. Eight heterozygotes have a mean trait value of 106 units, and four variant-allele homozygotes have a mean of 105 units. Ten reference homozygotes have a mean of 100 units. Calculate the combined carrier mean and dominant-coded contrast.
There are \(8+4=12\) carriers. Their total trait value is \(8(106)+4(105)=848+420=1{,}268\) units. Thus their combined mean is
The reference mean is \(\bar{Y}_0=100\) units, so the dominant-coded contrast is
The unweighted average of 106 and 105 would be 105.5 units, which is not the combined sample mean because the two genotype groups have different sizes. The dominant-coded estimate compares the observed combined-carrier mean, 105.7 units, with the reference mean, 100 units.
Interpreting Results Carefully
A dominant-coded coefficient is tied to the chosen reference genotype, the grouping rule, and the trait scale. State all three when reporting it. For instance, “The estimated mean trait value was 5.7 units higher among individuals with at least one variant allele than among reference homozygotes” names the groups, direction, and units. If reporting uncertainty or a test from a fitted model, describe the interval or test in the context of this same carrier-versus-reference contrast.
The coefficient describes an association in the data under the specified model. Genotype assignment alone does not make an observed contrast causal: population structure, selection, relatedness, measurement, or other design features can affect interpretation. Nor does a positive coefficient mean that every carrier has a higher trait value than every reference homozygote; it compares group means, while individual values may overlap.
Common Mistake
- Wrong reasoning: “The coefficient is the effect of each additional variant allele, because the genotype is coded numerically.”
- Correction: Dominant coding maps both \(G=1\) and \(G=2\) to \(D=1\). The coefficient is the carrier-versus-reference mean difference, not a per-allele increment.
- Wrong reasoning: “A dominant-coded result proves that heterozygotes and variant homozygotes have identical trait means.”
- Correction: The model imposes one fitted mean for the combined carrier group. Whether that restriction is a good description requires considering the genotype-specific pattern and the scientific question.
- Wrong reasoning: “Dominant means the variant biologically masks the reference allele.”
- Correction: Here “dominant” names a statistical grouping of genotypes. A coding choice alone does not demonstrate a molecular mechanism.
Check Your Understanding
Answer each question using the dominant carrier coding defined above.
- What values of \(D\) correspond to genotypes with \(G=0\), \(G=1\), and \(G=2\) variant alleles?
- A study reports \(\hat{\beta}_D=-1.8\) cm. State the mean comparison represented by this coefficient.
- Why is the combined carrier mean not necessarily the unweighted average of the heterozygote and variant-homozygote means?
- What shared-mean restriction does the model impose on \(E(Y\mid G=1)\) and \(E(Y\mid G=2)\)?
- Why does a dominant-coded association not, by itself, establish a causal or molecularly dominant effect?