Projections of Models
proj returns a matrix or list of matrices giving the projections
of the data onto the terms of a linear model. It is most frequently
## S3 method for class 'aov': proj(object, onedf = FALSE, unweighted.scale = FALSE, \dots)
## S3 method for class 'aovlist': proj(object, onedf = FALSE, unweighted.scale = FALSE, \dots)
## S3 method for class 'default': proj(object, onedf = TRUE, \dots)
## S3 method for class 'lm': proj(object, onedf = FALSE, unweighted.scale = FALSE, \dots)
- An object of class
"lm"or a class inheriting from it, or an object with a similar structure including in particular components
- A logical flag. If
TRUE, a projection is returned for all the columns of the model matrix. If
FALSE, the single-column projections are collapsed by terms of the model (as represented in the analysis of variance table).
- If the fit producing
objectused weights, this determines if the projections correspond to weighted or unweighted observations.
- Swallow and ignore any other arguments.
A projection is given for each stratum of the object, so for
models with an
Error term the result is a list of projections.
- A projection matrix or (for multi-stratum objects) a list of
Each projection is a matrix with a row for each observations and either a column for each term (
onedf = FALSE) or for each coefficient (
onedf = TRUE). Projection matrices from the default method have orthogonal columns representing the projection of the response onto the column space of the Q matrix from the QR decomposition. The fitted values are the sum of the projections, and the sum of squares for each column is the reduction in sum of squares from fitting that column (after those to the left of it).
The methods for
aovmodels add a column to the projection matrix giving the residuals (the projection of the data onto the orthogonal complement of the model space).
onedf = FALSEthe result is not a projection, but the columns represent sums of projections onto the columns of the model matrix corresponding to that term. In this case the matrix does not depend on the coding used.
Chambers, J. M., Freeny, A and Heiberger, R. M. (1992) Analysis of variance; designed experiments. Chapter 5 of Statistical Models in S eds J. M. Chambers and T. J. Hastie, Wadsworth & Brooks/Cole.
N <- c(0,1,0,1,1,1,0,0,0,1,1,0,1,1,0,0,1,0,1,0,1,1,0,0) P <- c(1,1,0,0,0,1,0,1,1,1,0,0,0,1,0,1,1,0,0,1,0,1,1,0) K <- c(1,0,0,1,0,1,1,0,0,1,0,1,0,1,1,0,0,0,1,1,1,0,1,0) yield <- c(49.5,62.8,46.8,57.0,59.8,58.5,55.5,56.0,62.8,55.8,69.5, 55.0, 62.0,48.8,45.5,44.2,52.0,51.5,49.8,48.8,57.2,59.0,53.2,56.0) npk <- data.frame(block = gl(6,4), N = factor(N), P = factor(P), K = factor(K), yield = yield) npk.aov <- aov(yield ~ block + N*P*K, npk) proj(npk.aov) ## as a test, not particularly sensible options(contrasts = c("contr.helmert", "contr.treatment")) npk.aovE <- aov(yield ~ N*P*K + Error(block), npk) proj(npk.aovE)