spida-package: Collection of utility functions for mixed models, etc. for...

In gmonette/spida: Collection of miscellaneous functions for mixed models etc. prepared for SPIDA 2009+ (development version)

Description

Collection of utility functions for mixed models including

wald:	Wald test with functions to generate hypothesis matrices
gsp:	general spline with any degree and smoothness with additional function 'sc' to contruct hypotheses, smsp - easy smoothing splines with lme and nlme
capply:	create contextual variables. Also: cvar, dvar, up
Misc:	vif.lme; xqplot; tab; panel.subgroups (additional level of subgroups in xyplot)

Details

Package:	spida
Type:	Package
Version:	0.1-1
Date:	2010-07-19
License:	GPL (>= 2)
LazyLoad:	yes

A local R package currently available from

http://www.math.yorku.ca/~georges/R/spida_0.1.zip

containing functions and some datasets developed for the courses PSYC6140, MATH6630, MATH6643, MATH6627, SCS Workshops on Mixed Models, the Summer Program in Data Analysis organized by SCS at York in June 2009 and MGT 1382 (in the MBA program at Rotman).

Many functions are inspired by, extensions of, or simply cobbled from functions in the 'car' package by John Fox.

Please send any comments or changes to: Georges Monette <[e-mail]>

Partial list of functions:

Tables:

• atotal: border an array with sums

• abind : glue two arrays together on a selected dimension

• tab( fmla, data ) : includes NAs

Graphics:

td : easy front end to trellis.device and related functions xqplot: extended quantile plots panel.subgroups : to allow an extra level of subgroups in lattice graphics

Splines:

• lsp : linear

• qs : quadratic

• cs : cubic

• gsp : general spline with arbitrary degree in each interval and arbitrary constraint at each knot

• sc : function to generate hypothesis matrices for 'gsp' for use with Wald tests

• smsp: smoothing splines using mixed models

Inference

• glh - same as wald below but name retained for compatibility

• wald - General Linear Hypothesis of fixed effects using a Wald test

  • should work on lm, lme, nlme and lmer objects

  • wald( fit, L ) where 'L' is a hypothesis matrix performs a Wald test

  • wald( fit, 'pat' ) where 'pat' is a regular expression (see ?regex) uses an L matrix that tests all fixed effects whose names match 'pat'. For example > wald( fit, ':.*:' ) will test all 2- and higher-way interactions together.

  • wald( fit, c(2,3,5) ) will test the 2nd, 3rd and 5th coefficients.

• Lmat - Lmat ( fit, 'pat' ) constructs a hypothesis matrix.

• Lform - Lform( fit, list.of.formulas, data.frame ) constructs an L matrix applying each formula in the data.frame environment to create the corresponding column of the L matrix. Single scalars (e.g. 0 or 1) expand to fill the column.

• Leff and related tools (fdiff, flevs) simplify the visualization, estimation, testing and display of specific effects in models with complex interactions. See Leff.

• as.data.frame( wald( fit, Lform( fit, list.of.formulas, data.frame)), se = 2) returns a data frame with 'coef', 'coefp' = coef + se * std.error, 'coefm' = coef - se * std.error. This can be used for plotting fitted values with SEs for functions (e.g. lme) where the predict method does not produce standard errors, or where there is no predict method (e.g. lmer).

• Lmu

• Ldiff

• L2 - to come: will generate a hypothesis matrix that produces Wald-type Type II tests

• sc - hypothesis matrices for splines

• cell - a modified version of car::confidence.ellipse.lm that can add to a plot

• vif.lme - method for vif modelled on vif.lm from library car (note that model.frame and model.matrix are not implemented for lme objects)

Linear algebra

• OrthoComp - a basis for the orthogonal complement of the orthogonal projection of span(X) into span(Z) within span(Z).

• ConjComp - similar with respect to an inner product.

Graphics for linear algebra

• vplot - plots column vectors adding to current plot

• vell - ellipse as a 2 x n matrix

• vbox - box around unit circle

• orthog - 2 x 2 rotation

• orthog.prog 2 x 2 matrix of orthog projection

File utilities

• Read.spss : uses spss.get in library (Hmisc)

• Read.dta : uses read.dta in library (foreign) with trimming and conversion to factors. Reading an spss file that has been saved, in SPSS, as a Stata 'dta' file seems to produce proper date variables.

• trim : removes trailing blanks in character variables and factor levels - useful to post-process files from SPSS read with read.dta

Regression utilities:

• na.resid, na.fitted: pads with NAs to fit original data frame. These can be extended to additional classes by writing a method for 'na.pad'

• com, na.com: com(fit,'age') returns the component in 'fit' corresponding to effect that matches the regular expression 'age', na.com returns a vector padded with NAs to match the original data frame.

• comres, na.comres: component + residuals

Multilevel utilities:

• varLevel( dd, fmla ) shows the level of each variable in data.frame dd - e.g. varLevel( dd, ~Board/School/Class/Subject ) or just - varLevel( dd, ~ID ) to identify inner and outer variables

• up ( dd, fmla , keep ) creates a summary data set using the first row of each set of rows identified by 'fmla'. 'keep' is the level below which to keep variables. By default, only 'outer' variables are kept. Caution: currently uses library(nlme) which conflicts with library(lme4)

• capply ( x, data, FUN, ...) creates 'contextual' variables - like tapply but replicating the original shape - the ouput of FUN can also be time-varying with the same length as the argument

• xapply.formula ( fmla, by, FUN, ...) a formula interface to 'aggregate' - can be used with 'expand.grid' to create a variable whose values depend on other variables - useful to create 'prediction data frames'

• fill ( x, by, FUN , ... ) uses capply to create contextual variable by selecting a value. Works intelligently with factors and 'Date's. - the default FUN selects a unique value and can be used when a long data frame has been created with values for macro variables only in the first position.

• cvar ( x, id ) creates 'contextual' variables like 'capply' but using the mean. cvar also works when 'x' is a factor and returns a matrix with (length( levels(x) ) - 1) columns with group mean values of the coding matrix.

• dvar ( x, id ) is almost equivalent to x - cvar( x, id ) except that it is interpreted correctly when 'x' is a factor.

Miscellaneous utilities:

• labs( dframe ) <- c('rows','colums') adds an attribute 'labs' to a data frame and adds the class 'data.frame.lab'. 'print.data.frame.lab' uses 'labs' as headings for rownames and column names, in parallel to the effect

  If 'mat' is a matrix, labs( mat ) is equivalent to: > names(dimnames(mat)) <- c('rows','columns')

Recent Changes:

• December 8, 2008 predict.mer - a crude predict method for 'mer' objects requires a repetition of the model formula

• August 8, 2008 sc = generate coefficients for a spline for glh smsp : basis for a smoothing spline

• July 31, 2008 – new version: gsp ( x, k, degree, smooth) : general polynomial spline with variable degrees in each interval and variable smoothness at each knot.

• July 29, 2008 eg – INCOMPLETE a more efficient version expand.grid that takes level information from a data frame.

Author(s)

Georges Monette <[e-mail]>

References

Monette, G. (1990). Geometry of Multiple Regression and Interactive 3-D Graphics. In Fox, J. & Long, S. (ed.) Modern Methods of Data Analysis, Sage Publications, 209-256.

See Also

lme ~~

Examples

~~ simple examples of the most important functions ~~

gmonette/spida documentation built on May 17, 2019, 7:25 a.m.