Response-Surface Analysis

Provides functions to generate response-surface designs, fit first- and second-order response-surface models, make surface plots, obtain the path of steepest ascent, and do canonical analysis. A good reference on these methods is Chapter 10 of Wu, C-F J and Hamada, M (2009) "Experiments: Planning, Analysis, and Parameter Design Optimization" ISBN 978-0-471-69946-0. An early version of the package is documented in Journal of Statistical Software .


NEWS for rsm package

Changes for version 2.10 () ------

  • Changes to avoid loss of significance in
  • Informative error message in contour.lm() and relatives when non-existing variables are specified in 'form'
  • Some S3 methods (e.g. contour.lm) also exported visibly to save confusion
  • 'emmeans' support (recover_data.rsm, emm_basis.rsm) now dynamically registered

Changes for version 2.9 (22 October 2017)

  • Changed license to ease interdependencies with other packages
  • Moved development codebase to github repository rvlenth/rsm
  • Added more robust function
  • Added support for emmeans package

Changes for version 2.8 (14 October 2016)

  • Added 'adjust' argument to 'rsm()'
  • Added support for lsmeans package

Changes for version 2.7-4 (6 October 2015)

  • Fixed error incurred when PQ() has only one argument

Changes for version 2.7-3 (2 September 2015)

  • Added NAMESPACE imports of non-base packages referenced

Changes for version 2.7-2 (12 May 2015)

  • Modified the "rs-illus" vignette slightly to make it less confusing

Changes in version 2.07 (1 October 2014)

  • Including factors specified in 'at' in the slice labels

Changes in version 2.06 (18 April 2014)

  • Fixed a bug in contour.lm wherein if xlabs is supplied, decoding was disabled
  • Additional messages from contour.lm for misspecified axis labels.

Changes in version 2.05 (14 April 2014)

  • Added back some details (e.g. R^2, F) earlier omitted from summary.

Changes in version 2.04 (5 February 2014)

  • Correction in bbd: Wrong design was generated when k = 6

Changes in version 2.03 (21 February 2013)

  • Corrected error in slice labels present since 2.00 (Thanks AGAIN go to Keith Ponting, Avaya)

Changes in version 2.02 (9 February 2013)

  • Corrected deparse bug that occurs when model formula wraps a line. (Thanks go to Keith Ponting, Avaya)

Changes in version 2.01 (29 December 2012)

  • Added .ccd.1.41 and .bbd.141 functions, so that given the random seed, one can reproduce a design that was generated in version 1.41 or earlier with the identical randomization. These functions are NOT exported.

  • Renamed 'join' to 'djoin' to avoid clashing with 'join' function in the package

  • and generate automatic codings if no formulas are provided.

Changes in version 2.00 (7 December 2012)

This is a major update of 'rsm'. It includes more pervasive and complete integration of, and improves flexibility in generating designs and RSM analysis.

  • Old objects created before this version may need to be updated (via in order to use some new functions such as join

  • Allows more general rsm models, e.g. FO(x1,x2,x3,x4) + TWI(formula = ~x1:(x2+x3+x4)) + PQ(x1,x3)

  • Can deal with singular second-order effects (e.g. stationary ridge). New threshold option in canonical analysis to actually force singularity -- which likely will identify a stationary point nearer the design center.

  • All design-generating functions ALWAYS create objects, and they have columns for run.order and std.order. There are additional (under the hood) enhancements to to keep track of primary variables, blocks, etc.

  • New cube(), star(), foldover(), dupe(), and join() functions for creating and combining designs. These should usually be used in preference to ccd() when doing iterative response-surface experimentation

  • Added a oneblock argument to ccd() so we can create an un-blocked CCD

  • A star block may be added to ANY design, including ones imported from other packages. Options alpha = "orthogonal" and alpha = "rotatable" calculate the the design moments, check suitability and set alpha to meet the criterion, or raise an error condition if it cannot be met.

  • New 'block' argument in coded-data-generating functions. May be needed to identify blocking factor(s).

  • New varfcn() function for examining the scaled variance function for a design

  • contour.lm, persp.lm, image.lm now have "decode" argument (TRUE by default) that displays decoded values on the coordinate axes when codings are available

  • contour.lm et al. now average over any factor levels rather than picking the first level of each. (May still specify particular levels in 'at')

  • New vignette with a simulated example of a series of response-surface experiments

  • Fixed some scoping bugs so that rsm can be called within a function. (This should make it possible to retire the spotRsm function in the SPOT package)

  • parse.coding function replaced by .parse.coding (and not exported)

  • row.names attribute for a design is updated to 1:nrow(design) every time it is randomized. 'ccd' does not create the same weird row names that it used to

  • now has argument to decide whether to show coded values or actual values (actual is the default)

  • function "codings<-" to change the codings

  • function "[" implemented for smarter handling of subsetted data; for example, if a coded column is excluded, so is its coding.

  • New functions "names<", "", "truenames<" for smart handling of variable names, keeping coding consistent with names

  • xs() function to obtain stationary point of an rsm object

  • TWI() function now has optional 'formula' argument to aid in specifying reduced version of second-order models

Changes in version 1.41 (July 28, 2012)

  • Added a NAMESPACE so it will work in newer versions of R

Changes in version 1.40 (July 28, 2010)

  • Fixed scoping bugs in '' (affects 'contour.lm' etc.)

  • Fixed scoping bugs in 'rsm'

  • Fix to 'rsm' -- coding worked incorrectly when 'data' keyword was not given explicitly. Side effect: 'data' is no longer a separate argument, it's just part of '...'

  • Added 'inscribed' optional argument to 'ccd' (Thanks to Ron Behlinger for suggesting this)

  • 'contour.lm' checks to ensure that elements of 'at' and 'bounds', if provided, are named.

  • New 'atlab' argument in 'contour.lm' and related functions. Now by default, the 'at' values are displayed for variables not on the coordinate axes; 'atlab' specifies where this information is shown, or suppresses it.

Changes in version 1.31 (December 19, 2009)

  • Fixed logic bug in contour.lm - didn't handle xlabs correctly

Changes in version 1.30 (December 16, 2009)

  • Enhanced contour.lm with hooks and improved axis labeling Also changed default to image=FALSE -- seems more "pure"

  • Added persp.lm for perspective plots akin to contour.lm

  • Added image.lm for image-only plots akin to contour.lm

  • Added 'codings' methods for and rsm objects

  • Added 'canonical' method for rsm for more convenient access

  • Updated references to 3rd ed. of Myers, Montgomery, Anderson-Cook (also corrected misspelling of Myers -- my apologies to Ray.)

  • Additional vignette "rsm-plots" describing contour, image, persp in more detail

Changes in version 1.21 (October 30, 2009)

  • Bug fix in contour.lm (axis labeling was reversed whern image=FALSE) Thanks to Eduard Sturm for reporting this

Changes in version 1.20 (October 20, 2009)

  • Replaced citation to point to paper published in JSS

  • Replaced vignette by JSS article

Changes in version 1.13 (12 October, 2009)

  • Fixed one misused macro in documentation

Changes in version 1.12 (5 May, 2009)

  • Modified CITATION file to conform to standards

Changes in version 1.11 (6 March, 2009)

  • Changed last variable name to 'logSD' in 'heli' and clarified documentation for that dataset

  • Fixed bug in contour.lm relating to original use of model formula, e.g. it now works with for model fitted by 'lm(swiss)'

Changes from version 1.10 (30 December, 2008)

  • Added functions ccd() and bbd() for generating RS designs

  • Added function ccd.pick() to help in choosing good CCDs

  • Added checks in rsm() for aliased parameters

  • Added NEWS and CITATION files.

  • Tweaks/corrections/additions to documentation and vignettes.

Changes from version 1.01 (5 December 2008)

  • Added canonical.path function

  • Modified formula handling in and relatives so that any legal expression will work (as long as it's linear!)

  • Overhauled contour.lm, with different argument list. Now it is much easier to use, and can produce multiple plots with one call.

  • Provided a true vignette. Other improvements to documentation.

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.


2.10.3 by Russell Lenth, 2 months ago

Browse source code at

Authors: Russell Lenth [aut, cre]

Documentation:   PDF Manual  

Task views: Design of Experiments (DoE) & Analysis of Experimental Data

GPL (>= 2) license

Imports estimability

Suggests emmeans, Vdgraph,, DoE.base, FrF2

Imported by HoRM, OptimaRegion, SPOT, mclcar, rsurface.

Depended on by DoE.wrapper.

Suggested by emmeans, industRial, mistat, mlr, vdg.

See at CRAN