Providing software solutions since 1976
Usage Note 24446: Are there corrections to the book "SAS System for Mixed Models"?
 |  |  |  |
Following are corrections to the first, second, and third printings of the first edition of SAS System for Mixed Models. Note that later editions are available.
*
For software releases that are not yet generally available, the Fixed
Release is the software release in which the problem is planned to be
fixed.
Errata for first and second printings of SAS System for Mixed Models1
A special thanks to Jørgen Granfeldt, Harold Henderson, and Cliff Pereira for contributing many of these.
/*-------------*/
/*---Preface---*/
/*-------------*/
page ix, last line, remove bar on y_{ij}
page x, end of first full paragraph, change i to i' in
the second y bar term
/*---------------*/
/*---Chapter 1---*/
/*---------------*/
page 2, after (1.1), change b_i to b_j
page 2, middle of the page, add Cov(e,b) = 0 to the assumptions for
the random effects
page 2, formula (1.2), add a bar over e_{1.}
page 2, last sentence, add a dot to the subscript of y bar_1
page 4, formula (1.8) and third line from the bottom, change
sigma hat to sigma hat **2
page 8, Interpretation line 3, change sigma hat to sigma hat **2
page 9, line 3, change 0 1 -1 to 1 -1 0
page 15, formula (1.12), remove the last )
page 16, formula (1.16), add generalized inverse symbol - after )
page 23, Table 1.5, change F-statistic 2.62 to 1.23
page 24, SAS code, change block to blk
/*---------------*/
/*---Chapter 2---*/
/*---------------*/
page 35, line 6, add a dot to the subscript of (alpha beta)
page 36, line 5, add a bar to both y terms
page 36, line 7, change u_{ij} to u_{1j}
page 37, middle of page, add hat to mu in var(mu_{1.}), and
five more lines down, add hats to mus in mu_{1.} - mu_{2.}
page 37, last equation, use apostrophe for transpose on K, K'
page 39, line 4, add dot to mu_4
page 41, third and fourth displayed equations, add hats to mus
page 41, fourth displayed equation and last line, add hats to
Vs
page 41, line 2 and line 16 from bottom, add dot to mu_4
page 41, use apostrophe for transpose on K, in both matrix
displays of K'
page 42, second equation, add hat to V
page 42, Table 2.2, change rank for ET from 4 to 3
page 44, after PROC MIXED code, change -2 REML log-likelihood from
58.4895 to 59.4895
page 46, end of line 2, change "If the random" to "If no random"
page 57, change std err(mu1.) to std err(mu1. hat)
page 60, lines 2 and 3 from bottom, change gamma_{ik} to w_{ik}
page 60, lines 1 and 2 from bottom, change both 4s to 3s
page 61, lines 1 and 2, change 4s to 3s
page 61, first equation, change 4s to 3s, 0.7154 to 0.7054, and
divide the entire quantity under the square root sign by 12
page 61, second equation, change 4 to 2
page 63, last line, change the (1/2)'s to (1/3)'s
page 64, line 2, change 1/2 to 1/3 and mu_i to mu_{i.}
page 64, last line, change 1/8 to 1/12 and 1/2 to 1/3
page 66, in the three formulas, change 1/2 to 1/3
page 69, Output 2.16, Test of Fixed Effects table, change DDF
for CULT to 2.28 and p-value to 0.5756
page 71, in the PROC GLM code, change rep to block
page 75, in formula for F, add hats to mus and V
page 77, sentence immediately preceding the mu expressions, change
"estimates" to "parameters"
page 78, add hats to all mus within var()
page 78, line 3, add bar on y_{i..}
page 78, line 4, the 2 after (1/27) should be an exponent
page 78, line 5, add exponent 2 after (1/27)
page 78, line 9, change mu_{.i} to mu_{.j}
page 78, line 15, change sigma^2 to (4 sigma_R^2 + sigma^2)
page 79, add hats to all mus within var()
page 80, add hats to beta and V in the expressions for se(K'beta)
and F and use apostrophe for transpose
page 80, in the expression for F, delete "K'(" after the "[" and add
a superscript -1 to (X'V^{-1}X) in the middle of the expression
pages 81-82, Least Squares Means table, change the DDFs for the
location lsmeans from 54 to 18
page 83, add hat to beta in expressions for F and use apostrophe for
transpose
page 83, middle of page, change "are given at the end of Output
2.21" to "can be produced by adding the E option to the first
ESTIMATE statement in the preceding PROC MIXED code"
page 84, Least Squares Means table, change DDFs from 54 to 9.43
page 85, add hat to beta in expressions for se(K'beta) and F and use
apostrophe for transpose
page 85, middle, change 2.3.3 to 2.8.3 and add hat to mu in var()
page 85, last sentence, change 6.5 to 6.6
/*---------------*/
/*---Chapter 3---*/
/*---------------*/
page 99, insert the following after the first paragraph:
You can add a RANDOM INTERCEPT / SUB=SUBJ(PROGRAM) statement
to this program to fit a random intercept in addition to the
AR(1) covariance structure. This addition is consistent
with the delta_ij term in model (3.1). It does not significantly
improve the model fit and results are not shown here.
page 99, change the last full paragraph to the following:
Adding a RANDOM INTERCEPT / SUB=SUBJ(PROGRAM) statement to this
program would be redundant because the unstructured covariance
structure is the most general one possible.
page 102, first full paragraph, change "with values of the
criteria closest to zero are" to "with the largest values of
the criteria is"
page 102, in the second table change the p-value for PROGRAM*TIME
with UN from 0.0998 to 0.1297; in the subsequent paragraph,
delete the "marginally significant" clause
page 106, in the PROC GLM code, add TIME to the CLASS statement
page 113, output for TIME.4 should be changed to
Source DF Type III SS Mean Square F Value Pr > F
MEAN 1 0.08132173 0.08132173 0.12 0.7329
page 113, output for TIME.6 should be changed to
Source DF Type III SS Mean Square F Value Pr > F
MEAN 1 0.17490098 0.17490098 0.55 0.4617
page 117, third paragraph, change "Output 3.8" to "Output 3.6"
page 125, output for TIME.5 should be changed to
Source DF Type III SS Mean Square F Value Pr > F
MEAN 1 0.02127019 0.02127019 0.07 0.7883
page 134, line 4, change "Such structures" to "Similar structures"
/*---------------*/
/*---Chapter 4---*/
/*---------------*/
page 138, move (4.3) out to the right margin and move (4.4)
to the expression for the method-of-moment estimators; in (4.3),
change t to s and divide the expression for MSsite(I) by
n_{.} - s
page 139, expression for u hat, change b to beta hat
page 140, first and third displayed equations, change t to s
page 148, DATA SATT code, switch the comments on ms influent and ms
error
page 154, first paragraph of 4.3.4, change Output 4.9 to Output 4.11
and 56.17 to 15.11
page 158, last formula, add hat to sigma^2_L
page 163, output 4.17, line 4, column 'Ratio' , change "7.71827297" to
"17.71827297"
page 165, line 2, add sigma^2_l/4 + sigma^2_b/12 to the
expression for the variance of a family mean, and
change the denominator of sigma^2_{LF} from 3 to 4
page 166, sixth equation, between the two equal signs add 613.6/4
and 89.3/12 to the expression and change the denominator of 74.9
from 3 to 4; on the right-hand side of the second equal sign, change
217.3 to 371.8
page 166, seventh equation, change 217.3 to 371.8 and 0.865 to 0.506
page 166, line one of the second paragraph, change 0.865 to 0.506
and 86.5% to 50.6%
page 168, second paragraph, change "The only F-statistic with
meaning in Output 4.19 is that corresponding to BLOCK(LOC)"
to "The only F-statistics with appropriate denominators in
Output 4.19 are those corresponding to LOC*FAM and BLOCK(LOC)"
page 169, in Bryk reference, remove the comma before (1992)
/*---------------*/
/*---Chapter 5---*/
/*---------------*/
page 179, line 6, change WT*TRT to IWT*TRT
page 192, PROC MIXED code, add "ddfm=satterth" to the end of the
MODEL statement
page 202, PROC MIXED code, add "ddfm=satterth" to the end of the
MODEL statement
page 204, PROC MIXED code, add "/ ddfm=satterth" to the end of the
MODEL statement
page 205, PROC MIXED code, add "/ ddfm=satterth" to the end of the
MODEL statement
page 206, PROC MIXED code, add "/ ddfm=satterth" to the end of the
MODEL statement
page 207, PROC MIXED code, in the labels of the first eight
ESTIMATE statements, change "m-f" to "f-m"
page 211, line 12 from the bottom, change "two" to "three"
/*---------------*/
/*---Chapter 6---*/
/*---------------*/
page 233, first equation, change X_{ij} to X_{ji} and Z_{kj}
to Z_{jk}
page 233, add "+ e_j" to the end of the first equation, and
"e_j is the jth residual error" to the description list; add
bold "+ e" to the end of the first two occurrences of "X beta
+ Z u"
page 235, label of ESTIMATE statement at bottom, change "narrow"
to "broad"
page 238, Output 6.2, widen output so that p-values don't wrap
page 240, near bottom, insert "estimate of the" before "sire
variance component" and "within variance component"
page 241, second sentence of first full paragraph, insert "estimate
of the" before "within variance component"
page 241, middle, change "BLUP of sire 1" to "empirical BLUP
of sire 1"
page 246, Output 6.5, after the expected mean squares, insert
the header line
MACHINE Y Std Err Pr > |T|
/*---------------*/
/*---Chapter 7---*/
/*---------------*/
page 264, line 7, change "These estimates are computing" to
"These estimates are computed"
page 264, Table 7.1, change the ML3 value for omega_{11} from
0.09 to 0.009
/*---------------*/
/*---Chapter 8---*/
/*---------------*/
/*---------------*/
/*---Chapter 9---*/
/*---------------*/
page 306, line 3 of 9.4.1, change tau_I to tau_i
Page 319, PROC MIXED code, change "spherical" to "exponential" in the
TITLE statement
page 322, PROC MIXED code, add (1) after (18.1) in the PARMS
statement
page 323, Output 9.7(a), in the Covariance Parameter Estimates table,
change the Residual estimate to 1 and the ratios to the same values
as the estimates
/*----------------*/
/*---Chapter 10---*/
/*----------------*/
page 353, Interpretation, remove extra space around the expressions
of differences of mu bars
page 390, line 2 of last paragraph, change 10.8.2 to 10.6 .
page 391, PROC MIXED code, drop SEQ from the CLASS statement
page 392, line 7 of output, change PRIORTRT to PRIORPRD
page 392, PROC MIXED code, drop SEQ from the CLASS statement
page 395, PROC MIXED code, drop SEQ from the CLASS statement
page 395, line 5 of second paragraph, change
SUBJECT=PERIOD*PERSON*BLK to SUBJECT=PERIOD*PERSON(BLK)
page 396, PROC MIXED code, drop SEQ from the CLASS statement
page 398, PROC MIXED code, drop SEQ from the CLASS statement
page 401, top, Output box should be open at the top
/*----------------*/
/*---Chapter 11---*/
/*----------------*/
page 427, line 12, add a transpose operator (') to the first D
pages 432-3, 435 in MME, 436 in C=, 442 in standard error, 444 in
displayed equations, use apostrophe ' for transpose consistently.
Often a different symbol is used.
page 436, line 3, add a transpose operator (') to the first D
page 439, end of first paragraph, parentheses is misspelled
page 440, GLIMMIX code, add a / CL at the end of the first
ESTIMATE statement
page 443, last line, change "convert" to "converted"
page 444, change X'WX to (X'WX)^- in four places
page 444, line 11, change 0.23(1-0.23) to [0.2354(1-0.2354)]^2
and change 0.239 to 0.102
page 444, line 12, change 0.275 to 0.135
page 444, use apostrophe for transpose
page 446, GLIMMIX code, delete the PARMS statement and replace
it with
repeated;
parms (2.032) (0.0813) (1.0) (1.0) / noiter;
page 446, middle paragraph, change EQCONS=3 to NOITER
page 446, replace Output 11.4 with the following:
GLIMMIX Model Statistics
Description Value
Deviance 8.1426
Scaled Deviance 8.6487
Pearson Chi-Square 6.0313
Scaled Pearson Chi-Square 6.4062
Extra-Dispersion Scale 0.9415
page 446, Interpretation, change 8.826 to 8.1426
page 447, GLIMMIX code, change data=a to data=new, fav/nij to y,
and logit to probit
page 447, replace the last paragraph with the following:
This program is similar to the previous analysis, but with
the following changes: A different data set, NEW, is used
instead of A. The NEW data set is an expanded version of A
in which all of the values of Y are 0s and 1s. This data
set is used to improve the stability of GLIMMIX. Also,
LINK=PROBIT is used instead of LINK=LOGIT. The results
from this analysis are given in Output 11.5.
page 450, replace the last three sentences with the following: Because
GLIMMIX has great difficulty converging for this analysis, we fix the
variance components using the NOITER option in the PARMS statement.
The results are displayed in Output 11.7 and are not intended to be
definitive, but rather are used for illustration purposes.
page 456, first paragraph, add "except the DDF for TRT changes
from 18 to 21" to the end of the first sentence
/*----------------*/
/*---Chapter 12---*/
/*----------------*/
page 465, line 5, delete second occurrence of "vector of"; hyphenate
"random effect"
page 465, first mathematical expression, change the last exponent on
W from 1/2 to -1/2
page 466, table 12.4, Methods 1 and 2, boldface the second Z_i and
add a -1 exponent to W_i; Method 3, change the last exponent on W_i
from 1/2 to -1/2
page 472, last line, change "intrasubject" to "intersubject"
page 479, Table 12.6, change the description of TLAG to
"the previous time value (if NEWSUB=0) or 0.0 (if NEWSUB=1)"
page 480, in top half of page, equations for c_i and v_i, use
standard (non-bold) font for u's, and in the equation for v_i,
change d_i to delta_i
page 484, change REML_LL to -442.910, -2REML_LL to 885.8209, and
Objective to 606.4636
page 485, change REML LL to -442.910, AIC to -445.910, BIC to
-450.446, and -2 REML LL to 885.8209
page 486, change the title of Section 12.6 to "Common Errors
and Convergence Problems"
page 487, third line of section 12.6, add "(exit abnormally)"
after "abend"
page 489, add the following reference:
Wolfinger, R. D. 1993. "Laplace's Approximation for Nonlinear
Mixed Models." Biometrika 80: 791-795.
/*----------------*/
/*---Appendix 1---*/
/*----------------*/
page 492, in the beta vector, the last subscript should be p and
not n
page 493, in the R matrix, add squares to all of the sigma_1's
page 498, the formulas listed for ML and REML refer to the case when
the residual variance is profiled; change these to the non-profiled
versions
page 500, first sentence of Statistical Properties,
u hat is BLUP of u, not beta
page 500, last equation, add a generalized inverse (^-) on the
right-hand side matrix
page 501, first matrix expression, change lower right corner element
from hat(C)_21 to hat(C)_22
page 501, last sentence of next-to-last paragraph, change "variance
component equals 0" to "variance component equals its lower
boundary constraint of 0"
page 504, Searle (1982) reference, change "for Statisticians"
to "for Statistics"
page 504, Swallow and Monahan reference, change the volume number
to 26
/*----------------*/
/*---Appendix 2---*/
/*----------------*/
/*----------------*/
/*---Appendix 3---*/
/*----------------*/
/*----------------*/
/*---Appendix 4---*/
/*----------------*/
page 553, Data Set 2.2(a), change rep to block
page 600, "categorical" is misspelled in the second line of
the reference at the end of Data Set 11.5
page 600, Data Set 11.6 A, change time to mix, and add
the line
y = count + 1;
after the INPUT statement
Errata for third printing of SAS System for Mixed Models1
| PAGE | CORRECTION |
| 3 | Change "Milliken and Johnson (1984)", to "Milliken and Johnson (1992)". |
| 3 | Change "Winer (1991)" to "Winer, Brown, and Michels (1991)". |
| 7 | Change "(MS(Blocks)-MS(Error))/7" to "(MS(Blocks)-MS(Error))/3". |
| 17 | In the displayed matrices and in the following text, change Jr to Jt, Φr to Φt, Ir to It, and "r×r" to "t×t". |
| 19 | Change "random block" to "random blk" in the SAS code. |
| 29 | Change date on Milliken and Johnson to 1992. |
| 29 | Change Winer reference to the following: Winer, B. J., D. R. Brown, and K. M. Michels. 1991. Statistical Principles in Experimental Design. New York: McGraw-Hill Inc. |
| 129 | For SAS Releases 6.12 and later, add
parms (0.1) (100);to the PROC MIXED code. The code without it converges to a local maximum of the likelihood for which ρ is negative. The PARMS statement provides a better starting value and convergence is to the global minimum. |
| 305 | In correlation formulas for 4 and 5, change "<2" to "<1". |
| 318 | Change the first bullet to the following: The "DIAG Diagonal" parameter, whose estimate is 3.236, is called the "partial sill" (Cressie, Statistics for Spatial Data, pg 67, just below equation 2.3.23). The sill is the sum of the partial sill and nugget. |
| 320 | In the first bullet, change "sill" to "partial sill". |
| 440 | Change the last two ESTIMATE statements to the following:
estimate 'trt 1 clinic 1 blup' intercept 1 trt 1 0
| clinic 1 0 trt*clinic 1 / e;
estimate 'trt 1 clinic 6 blup' intercept 1 trt 1 0
| clinic 0 0 0 0 0 1 0 0
trt*clinic 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 / e;
|
| 441 | Change two output lines to the following:
TRT 1 CLINIC 1 BLUP -1.1788 0.3223 7 -3.66 0.0081 TRT 1 CLINIC 6 BLUP -2.8710 0.7592 7 -3.78 0.0069 |
| 500 | Change "The covariance matrix of βhat and μhat is" to "The covariance matrix of (βhat - β, μhat - μ) is". |
| 500 | In third from last line, change "as the approximate variance-covariance matrix of βhat and μhat" to "as the approximate variance-covariance matrix of (βhat - β, μhat - μ)". |
1Ramon Littell et al., SAS System for Mixed Models (Cary, NC: SAS Institute Inc., 1996).
Operating System and Release Information
| Product Family | Product | System | SAS Release | |
| Reported | Fixed* | |||
| SAS System | SAS/STAT | All | n/a | |
| Type: | Usage Note |
| Priority: | low |
| Topic: | SAS Reference ==> Procedures ==> MIXED |
| Date Modified: | 2007-10-11 16:02:53 |
| Date Created: | 2005-10-06 16:54:40 |
