Predicted Means of a Linear Model

Share:

Description

This function obtains predicted means, SE of means, SED of means, LSDs and plots of means with Stder bar or LSD bar for parametric models such as aov, lm, glm, gls, lme, and lmer. The function also perfomrs pairwise comparisons and permutation tests.

Usage

1
2
3
4
5
predictmeans(model, modelterm, pairwise=FALSE, atvar=NULL, adj="none", 
  Df=NULL, level=0.05, covariate=NULL, trans = NULL, responsen=NULL, 
  count=FALSE, plotord=NULL, plottitle=NULL, mplot=TRUE, barplot=FALSE, 
  pplot=TRUE, bkplot=TRUE, plot=TRUE, jitterv=0, basesz=12, prtnum=TRUE, 
  newwd=TRUE, permlist=NULL)

Arguments

model

Model object returned by aov, lm, glm, gls, lme, and lmer.

modelterm

Name (in "quotes") for indicating which factor term's predicted mean to be calculated. The modelterm must be given exactly as it appears in the printed model, e.g. "A" or "A:B".

pairwise

An option for showing pair-wise LSDs and p-values, or not. The default is FALSE.

atvar

When pairwise = TRUE, a quoted name indicating within levels of which variable in modelterm the multiple comparison will be performed.

adj

Name (in "quote") for indicating a method for adjusting p-values of pairwise comparisons. The choices are "none", "tukey", "holm", "hochberg", "hommel", "bonferroni", "BH", "BY" and "fdr". The default method is "none".

Df

A degree of freedom for calculating LSD. For the above models, Df is obtained from the function automatically.

level

A significant level for calculating LSD. The default value is 0.05.

covariate

A numerical vector to specify values of covariates for calculating predicted means. The default values are the means of the associated covariates.

trans

A function object for calculating the back transformed means, e.g. trans=exp.

responsen

Name (in "quotes") of the back transformed response variable in the model.

count

An option for indicating the back transformed mean values are counts or not. The default is FALSE.

plotord

A numeric vector specifying the order of plotting for two or three way interaction (e.g. plotord = c(2, 1, 3) will put the second variable in modelterm on the X axis, the first variable as the grouping variable, and the third one as the panel variable). The defaults are c(1, 2) and c(1, 2, 3) for two and three way interactions.

plottitle

A character vector specifying the main title for plot(s). The default is NULL.

mplot

An option for drawing a means plot, or not. The default is TRUE.

barplot

An option for drawing a bar chart, or not. The default is FALSE.

pplot

An option for drawing a p-values plot, or not when there are more than six p-values. The default is TRUE.

bkplot

An option for drawing back transformed plot, or not. The default is TRUE.

plot

An option for drawing plots, or not. The default is TRUE.

jitterv

A degree of jitter in x and y direction in the back transformed means graph. The default is zero.

basesz

The base font size. The default is 12.

prtnum

An option for printing covariate information on the screen, or not. The default is TRUE.

newwd

A logical variable to indicate whether to print graph in a new window. The default is TRUE.

permlist

A model parameter list produced by the function permmodels. When permlist != NULL, the option Df will be non-functional. This is a key option for pairwise comparisons via permutation tests.

Value

Predicted Means

A table of predicted means.

Standard Error of Means

A table of standard errors of predicted means.

Standard Error of Differences

Standard errors of differences between predicted means.

LSD

Least significant differences between predicted means.

Back Transformed Means

When trans!=NULL, a table of back transformed means with CIs are shown.

Pairwise p-value

A matrix with t-values above the diagonal and p-values below the diagonal, or matrix of pairwise comparison p-values for each level of atvar.

Note

The predictmeans function becomes confused if a factor or covariate is changed to the other in a model formula. Consequently, formulae that include calls as.factor, factor, or numeric (e.g. as.factor(income)) will cause errors. Instead, create the modified variables outside of the model formula (e.g., fincome <- as.factor(income)) and then use them in the model formula.

Factors cannot have colons in level names (e.g., "level:A"); the predictmeans function will confuse the colons with interactions; rename levels to avoid colons.

For predictmeans function, it is assumed that methods coef, vcov, model.matrix, model.frame and terms are available for model.

Author(s)

Dongwen Luo, Siva Ganesh and John Koolaard

References

Torsten Hothorn, Frank Bretz and Peter Westfall (2008), Simultaneous Inference in General Parametric Models. Biometrical, Journal 50(3), 346–363.

Welham, S., Cullis, B., Gogel, B., Gilm our, A., & Thompson, R. (2004), Prediction in linear mixed models, Australian and New Zealand Journal of Statistics, 46(3), 325-347.

Examples

1
2
3
4
5
6
7
8
  library(predictmeans)
  ftable(xtabs(yield ~ Block+Variety+nitro, data=Oats))
  Oats$nitro <- factor(Oats$nitro)
  fm <- lme(yield ~ nitro*Variety, random=~1|Block/Variety, data=Oats)
# library(lme4)
# fm <- lmer(yield ~ nitro*Variety+(1|Block/Variety), data=Oats)
  predictmeans(fm, "nitro", adj="BH")
  predictmeans(fm, "nitro:Variety", pair=TRUE, atvar="Variety", adj="BH")