Descriptive statistics before the fit

A variance component is an answer about a set of data; whoever has not looked at the set does not know what they are estimating. This decides nothing: it measures and shows.

Usage

describe(
  data,
  trait,
  classes = NULL,
  pedigree = NULL,
  covariates = NULL,
  missing_code = NULL,
  id = NULL
)

Arguments

data

data.frame

trait

trait column

classes

class columns to tabulate

pedigree

data.frame or NULL

covariates

covariate columns to check against the missing-value code and for values that look like one (see Value)

missing_code

missing-value code for the trait observations: the records holding it are missing, as in every fitter, and stay out of the descriptives. When given together with covariates, each covariate is checked: the code only applies to the trait, and a -999 in a covariate enters the regression LITERALLY, in real data this once went unnoticed past two programs at the same time

id

the animal column of the data. With pedigree, the records with an observed trait (missing_code counts as missing, as in the fit) are split into BLOCKS, the connected components of the relationship graph (the animals linked through their sires and dams; an animal absent from the pedigree is a founder, as in the fit; an id written another way than in the pedigree stops here with the message of the fit), and the classes of classes that give each block levels of its own are named. Two blocks with no relative in common whose records sit in different contemporary groups cannot be compared: any genetic difference between them is absorbed by the groups (Fernando, Gianola and Grossman 1983; Kennedy and Trus 1993). Every fitter warns about it from its own terms; describe() sees only the pedigree and the classes given, so blocks that genotypes link are counted here and not warned about by a single-step fit

Value

an object of class br_describe: n (the observed records: not NA, NaN, Inf, -Inf or missing_code) and n_na (the missing ones), mean, sd, quartiles, the Tukey fences and outside_fences on the observed records, and, when asked, classes, pedigree and leak (per covariate, the records with the trait observed and the covariate at the missing-value code). suspect lists the findings the fitters warn about before a fit (see model()), on the observed records: a common missing-value code (-9, -99, -999, -9999, -99999, 99, 999, 9999, 99999, or -123456789) held by two records or more, outside the range of the other values and at least 3 robust SDs (the median absolute deviation, scaled to the SD) from their median; and the other records more than 10 robust SDs from the median. suspect_cov does the same per covariate, on the records with the trait observed. The print shows each finding as a WARNING line. With id and pedigree, blocks: the number of blocks of records with no relationship between them, the records of each and the classes that give each block levels of its own (absorbed_by); the print shows them as a WARNING line when such a class exists

References

Fernando, R.L., Gianola, D. & Grossman, M. (1983). Identifying all connected subsets in a two-way classification without interaction. Journal of Dairy Science 66:1399-1402.

Kennedy, B.W. & Trus, D. (1993). Considerations on genetic connectedness between management units under an animal model. Journal of Animal Science 71:2341-2352.