SAS Macro Programs: mvinfluence

The mvinfluence macro ( get mvinfluence.sas)

Influence measures for multivariate regression

The MVINFLUENCE macro calculates and plots measures of influence in a multivariate regression model. These are generalizations of the usual influence measures for univariate regression described by Barrett & Ling (1992). This is an initial, experimental implementation designed to explore thiese methods.

Three types of plots are provided (controlled by the PLOTS= argument)

COOKD:

scatterplot of generalized Cook's D against generalized leverage

STRES:

bubble plot of generalized squared studentized residual against generalized leverage using Cook's D as bubble size

LR:

bubble plot of log residual factor against log leverage factor using Cook's D as bubble size. This has the property that diagonal lines with slopes = -1 are contours of equal influence.

Usage

The MVINFLUENCE macro is defined with keyword parameters. The arguments may be listed within parentheses in any order, separated by commas. For example:

  %mvinfluence(data=rohwer, x=n s ns na ss, y=SAT PPVT Raven)

Parameters

DATA=

The name of the input data set [Default: DATA=_LAST_]

Y=

Names of response variables

X=

Names of predictor variables

ID=

The name of an observation ID variable

BUBBLE=

Bubble proportional to: COOKD. No other choices in this implementation [Default: BUBBLE=COOKD]

OUT=

The name of the output data set [Default: OUT=COOKD]

PRINT=

Print the OUT= data set? Not yet implemented. [Default: PRINT=NONE]

PLOTS=

Which plots to produce? NONE, ALL, or one or more of COOK, STRES, LR [Default: PLOTS=ALL]

LABEL=

Points to label: ALL, NONE, or INFL [Default: LABEL=INFL]

INFL=

Criteria for influential [Default: INFL=%STR(((HAT>&HCRIT)|COOKD>.5))]

LSIZE=

Obs label size. The height of other text is controlled by the HTEXT= goption [Default: LSIZE=1.6]

LCOLOR=

Obs label color [Default: LCOLOR=BLACK]

LPOS=

Obs label position [Default: LPOS=5]

LFONT=

Obs label font

BSIZE=

Bubble size scale factor [Default: BSIZE=10]

BSCALE=

Bubble size proportional to AREA or RADIUS [Default: BSCALE=AREA]

BCOLOR=

Bubble color [Default: BCOLOR=RED]

BFILL=

Bubble fill? SOLID or GRADIENT

REFCOL=

Color of reference lines [Default: REFCOL=BLACK]

REFLIN=

Line style for reference lines; 0->NONE [Default: REFLIN=33]

NAME=

The name of the graph in the graphic catalog [Default: NAME=MVINFL]

GOUT=

The name of the graphics catalog

Dependencies

%gskip - gskipDevice-independent macro for multiple plots
%labels - labels Create an Annotate dataset to label observations

References

* Barrett, B. E. & Ling, R. F. (1992). General classes of influence measures for multivariate regression, JASA, 87, # 417, 184-191.

* Barrett, B. E. (2003): Understanding Influence in Multivariate Regression, Communications in Statistics - Theory and Methods, 32:3, 667-680

* Some code from Timm & Mieczkowski, Univariate and Multivariate General Linear Models, http://ftp.sas.com/samples/A55809, Program 5_6.sas, was used in the initial implementation.

Examples

  %include data(rohwer);
  data rohwer;
     set rohwer;
     where group=2;
     drop subjno;
     case = _n_;
  %include macros(mvinfluence);        *-- or include in an autocall library;
  %mvinfluence(data=rohwer, x=n s ns na ss, y=SAT PPVT Raven, id=case,
  bfill=gradient, plots=ALL);

The following plots are produced:

See also