# What is the meaning of “trait” in biostatistics?

My question is related to the question "When and why do I have to use "trait" for multinomial multilevel models with MCMCglmm in R?" But in contrast to this question, I don't want to know, how trait has to be used in MCMCglmm, but what it means in (bio)statistics. The author of MCMCglmm, Jarrod Hadfield, is biologist. So what do biologists or biostatisticians commonly mean by this word?

(One example to make clear what I'm looking for: If I would be asking for the meaning of "cluster" in statistical papers about educational research, I would be looking for an explanation like "A cluster is a group, to which multiple observations belong to. For example, many students build a class and this class is then a cluster of students. Data with clusters are often analyzed by multilevel models.")

• I am not a biostatistician, but going from this, perhaps "trait" simply refers to a value taken by a categorical variable (in a particular individual)? – GeoMatt22 Oct 12 '16 at 16:16
• @GeoMatt22 it can also be a value of a continuous variable. The paper explaining MCMCglmm uses continuous measurements on tarsus and back as examples of "traits." – EdM Oct 12 '16 at 16:56
• OK, so essentially a "measurement on an individual"? @EdM perhaps you want to venture an answer? – GeoMatt22 Oct 12 '16 at 16:58

The paper explaining MCMCglmm indicates the sense in which Hadfield used the term "trait," which isn't quite the same as that recommended by the Wikipedia page. In my experience, I think that in common usage Hadfield's usage is generally accepted.

In biology, a trait is some characteristic of an organism. The distinction in usage is whether the "trait" is the name of the characteristic or the actual value of the characteristic. The Wikipedia page says:

For example, eye color is a character of an organism, while blue, brown and hazel are traits. (Original emphasis).

In the MCMCglmm paper, Hadfield uses names of characteristics (tarsus and back) as examples of "traits" rather than their actual values. In practice, I would also tend to use "trait" for the name of the characteristic rather than the value.

• OK. This makes sense, as my initial instinct prior to Wikipedia was "biostatistics trait = machine learning feature". – GeoMatt22 Oct 12 '16 at 17:08
• So if Hadfield writes "since the trait is partly sex limited" (p. 96), with "trait" he means the horn morph as a variable and not its possible values (or "levels") normal, polled or scurred, right? But what does he mean with the trait being "partly sex limited"? – Qaswed Nov 2 '16 at 11:09
• Even if I didn't ask about Hadfield's usage of "trait", I hope you can clearify the part of your answer where you cover this issue. In this example (p.99) I consider the trait to be the horn morph (variable horn in the SShorns data set). But now his model is "horn ~ trait + sex - 1", but why having trait as a covariate (which is not found in the data set) when the dependend variable horn is the trait? – Qaswed Nov 2 '16 at 12:48
• @Qaswed : I think that, in practice, use of the word "trait" often switches back and forth between the name of a characteristic and its value. That's what seems to be happening here. Note that trait is a reserved variable name in MCMCglmm; it automatically indexes the columns in a multi-response dependent variable. (The reserved variable name units indexes the rows.) See the MCMCglmm paper, page 6. So use of trait in MCMCglmm as a reserved variable name might be adding a further level of confusion. – EdM Nov 2 '16 at 14:11

Trait is endemic to genetics. The originating study here discusses genetic determinants of successful transition to work from school. The documentation in the TraMine package whence mvad originates is pretty paltry. To add to this problem trait seems to be pretty poorly defined in their work. According to them, trait can be a genotype or a phenotype. This is inclusive of trait as a categorical level referring to the inherited number or types of alleles in a gene that predicts disease, or mutations or SNPs thereof, it can be a concentration of mRNA or protein synthesized by such a gene, or even an overt phenotype in the sense of a person's height or weight. Basically, you will need to consult the documentation (There is none) to understand what it really is. Calling ?mvad should tell you.