SUDAAN Resolved Bugs

NCSH style for RTF can always be considered as an alternative.  In addition, users can in 
some cases use a different output group to obtain the same results.  For example, the 
STEST output group can be used instead of the old TEST output group in PROC CROSSTAB to 
display the same results. 

Example:

Instead of using

PRINT / STYLE=BOX…

You can use

PRINT / STYLE=NCHS…

System	Release Reported	Release Fixed
PC	10.0.0	10.0.1
SUN	9.0.2	10.0.1
LINUX	9.0.2	10.0.1

There is no work-around for this problem.

Example:

The following PROC DESCRIPT program creates an output dataset which contains the variable _ONE_.

proc descript data="c:\RegressionTesting\data\brfshort" filetype=sudxport nomarg noprint;
nest _ststr _psu;
weight _finalwt;

/*  analyze only females */
subpopn sex=2 / name="females";

var  papexam  mamexam;
catlevel     1     1;

subgroup _one_ nrace;
tables _one_*nrace;
levels 1 3;

/* specify 3 levels for nrace, do not consider nrace='other' (4)  */

title "Comparing Incidence of Pap Tests and Mammograms"  " ";

output total percent sepercent nsum wsum / replace   filename="c:\RegressionTesting\OutFiles\Events\Standalone\Event289" filetype=sudxport;

proc records data="c:\RegressionTesting\OutFiles\Events\Standalone\Event289" filetype=sudxport contents countrec;
setenv linesize=150;

The contents of the output dataset in Release 10.0.1 follows:
Variables                                                                    
Name          Type       Output     Description                              
                         Format                                              
------------------------------------------------                             
PROCNUM       Numeric    F2.0       Procedure Number                           
TABLENO       Numeric    F2.0       Table Number                               
_ONE_         Numeric               SUDAAN Reserved Variable One               
NRACE         Numeric    F12.        Race Code                                 
VARIABLE      Numeric    F3.0       Variable                                   
_C1           Numeric    F4.0       Codes for Categorical Variable _ONE_       
_C2           Numeric    F4.0       Codes for Categorical Variable NRACE       
TOTAL         Numeric    F12.2      Total                                      
PERCENT       Numeric    F10.2      Percent                                    
SEPERCENT     Numeric    F10.2      SE Percent                                 
NSUM          Numeric    F8.0       Sample Size                                
WSUM          Numeric    F10.2      Weighted Size     
                         
In Release 10.0.0 of SUDAAN the variable _C1 is not included on the output dataset.

System	Release Reported	Release Fixed
PC	10.0.0	10.0.1
SUN

System	Release Reported	Release Fixed
PC	9.0.3	10.0.1
SUN	9.0.2	10.0.1
LINUX	9.0.2	10.0.1

In some cases in PROC CROSSTAB for design=JACKKNIFE the values for SEROW, SECOL, SETOT and related statistics such as confidence limits may be incorrect when there are multiple tables on the TABLES statement.  You should not trust results in this case, and instead should create separate jobs for each table to avoid this error.
In addition when you analyze multiply imputed data (all designs) the confidence limits and other results depending on the computed denominator degrees of freedom may be incorrect if there are multiple tables (CROSSTAB, RATIO, DESCRIPT) or marginals (REGRESS, LOGISTIC, LOGLINK, MULTILOG).  You should not trust results in this case, and instead should create separate jobs for each table / marginal to avoid this error.

The following code may lead to incorrect results for SEROW, LOWROW and UPROW:
PROC CROSSTAB DATA=Test;
. . .

CLASS A B 
TABLES A B;
PRINT ROWPER SEROW LOWROW UPROW;

Instead, use the following code:
PROC CROSSTAB DATA=Test;
. . .

CLASS A;  
TABLES A;
PRINT ROWPER SEROW LOWROW UPROW;

PROC CROSSTAB DATA=Test;
. . .

CLASS B;  
TABLES B;
PRINT ROWPER SEROW LOWROW UPROW;

System	Release Reported	Release Fixed
PC	10.0.0	10.0.1
SUN

In some cases when the Hosmer Lemeshow test is specified in PROC LOGISTIC the following message may be (inappropriately) displayed:

DATA WARNING
:
Degrees of freedom for OVERALL contrast are less than maximum number of estimable parameters
This is likely due to an over-parameterized model; you may wish to rerun your job and reduce
the number of parameters in your model by dropping variables or collapsing classes for
categorical variables.

System	Release Reported	Release Fixed
PC	9.0.3	10.0.1
SUN	9.0.2	10.0.1
LINUX	9.0.2	10.0.1

System	Release Reported	Release Fixed
PC	10.0.0	10.0.1
SUN

System	Release Reported	Release Fixed
PC	10.0.0	10.0.1
SUN

There is no work-around for this bug.

Example:

The follow program generates a PROGARMMER ERROR:

PROC DESCRIPT DATA="c:\RegressionTesting\Data\FORSUDN1" filetype=SUDXPORT DESIGN=WR MI_COUNT=5 TOTALS;
NEST SDPSTRA6 SDPPSU6 / MISSUNIT;
WEIGHT WTPFQX6;
VAR TCPMI;
/* No CLASS, SUBGROUP, LEVELS or TABLES statement */
SETENV COLWIDTH=14 DECWIDTH=2;
SETENV LINESIZE=100 PAGESIZE=60;
PRINT / TABLECELL=ALL STYLE=NCHS;

System	Release Reported	Release Fixed
PC	9.0.3 10.0.0	10.0.1
SUN	9.0.2	10.0.1
LINUX	9.0.2	10.0.1

If you merge your main data set and the PSUDATA or REPDATA dataset outside of SUDAAN there is no error.

Example:

Assume the variable BRRID is on both “mydat” and “brrwts”, and that replicate weights BRR1 – BRR24 are on “brrwts”.  This code leads to a PROGRAMMER ERROR:

PROC CROSSTAB DATA=”mydat” FILETYPE=SAS REPDATA=”brrwts” DESIGN=BRR;

REPWGTS BRR1-BRR24;
IDVAR BRRID;

< additional program statements>

OUTPUT / TABLECELL=DEFAULT FILENAME=”xtab” FILETYPE=SAS REPLACE;

If you merge “mydat” and “brrwts” using BRRID as the merge variable, creating a new data set “alldat”, you avoid the PROGRAMMER ERROR:

PROC CROSSTAB DATA=”alldat” FILETYPE=SAS DESIGN=BRR;

REPWGTS BRR1-BRR24;



OUTPUT / TABLECELL=DEFAULT FILENAME=”xtab” FILETYPE=SAS REPLACE;

System	Release Reported	Release Fixed
PC	10.0.0	10.0.1
SUN

To avoid this error, create a separate CROSSTAB job for each GOFIT test you wish to run rather than combining all in a single call to CROSSTAB.

Example:

The following code produces the PROGRAMMER ERROR:
PROC CROSSTAB DATA="C:\gofone" FILETYPE=SUDXPORT DESIGN=WR;
. . .

SCORES HLTHSTAT=(10 20 30);
TEST  CMH ACMH TCMH/all;
CLASS HLTHSTAT HFB17;
/* Two GOFIT statements */
GOFIT HLTHSTAT=(.30 .40 .30)/all;
GOFIT HFB17=(.2 .8);
TABLES  HFB13*HLTHSTAT;
OUTPUT / GOF=default FILETYPE=SUDXPORT FILENAME=”GOFTEST” REPLACE;

The following code with two calls to CROSSTAB avoids the PROGRAMMER ERROR:

PROC CROSSTAB DATA="C:\gofone" FILETYPE=SUDXPORT DESIGN=WR;
. . .

SCORES HLTHSTAT=(10 20 30);
TEST  CMH ACMH TCMH/all;
CLASS HLTHSTAT HFB17;
/* Just the first GOFIT statement */
GOFIT HLTHSTAT=(.30 .40 .30)/all;
TABLES  HFB13*HLTHSTAT;
OUTPUT / GOF=default FILETYPE=SUDXPORT FILENAME=”GOFTEST1” REPLACE;

PROC CROSSTAB DATA="C:\gofone" FILETYPE=SUDXPORT DESIGN=WR;
. . .

SCORES HLTHSTAT=(10 20 30);
TEST  CMH ACMH TCMH/all;
CLASS HLTHSTAT HFB17;
/* Just the second GOFIT statement */
GOFIT HFB17=(.2 .8);
TABLES  HFB13*HLTHSTAT;
OUTPUT / GOF=default FILETYPE=SUDXPORT FILENAME=”GOFTEST” REPLACE;

System	Release Reported	Release Fixed
PC	10.0.0	10.0.1
SUN

System	Release Reported	Release Fixed
PC	10.0.0	10.0.1
SUN

Use PROC RECORDS to convert the data from type SUDAAN to another type. There is no PROGRAMMER ERROR in PROC RECORDS.  You can use the converted data set in other procedures. 

Example:

Suppose variable A in SUDAAN dataset TEST has 2 levels, but only has a code and stub associated level 1:

Values and Labels for Variable A:                                       
Value  Label                                                               
--------------                                                               
1   	“Purple” 

The following code will produce a PROGRAMMER ERROR:

PROC CROSSTAB DATA=”Test” FILETYPE=SUDAAN DESIGN=SRS;
CLASS A;
PRINT NSUM;

The following code will create a dataset of an alternate type that will not cause the PROGRAMMER ERROR in CROSSTAB:

PROC RECORDS DATA=”Test” FILETYPE=SUDAAN NOPRINT;
OUTPUT / FILENAME=”NewTest” FILETYPE=SPSS REPLACE;

PROC CROSSTAB DATA=”NewTest” FILETYPE=SPSS DESIGN=SRS;
CLASS A;
PRINT NSUM;

System	Release Reported	Release Fixed
PC	9.0.3	10.0.1
SUN	9.0.2	10.0.1
LINUX	9.0.2	10.0.1

System	Release Reported	Release Fixed
PC	9.0.3	10.0.1
SUN	9.0.2	10.0.1
LINUX	9.0.2	10.0.1

System	Release Reported	Release Fixed
PC	9.0.3	10.0.1
SUN	9.0.2	10.0.1
LINUX	9.0.2	10.0.1

There are no simple workarounds. One potential workaround is to change the event status of at least one observation at the highest order time to censored. This will prevent the Kaplan-Meier estimate from becoming 0.00 and avoid the error. 

Example: 

Consider this simple ordered data set:

PATIENT	EXTIME	COMPLETE
5	102	1
5	110	1
5	121	1
5	129	1
2	206	1
14	212	1
2	224	1
14	224	1
2	250	1
14	291	1
2	306	1
6	425	1
14	432	1
6	514	1
6	580	1
6	613	1

Running the following code will cause the error:

PROC KAPMEIER DATA=temp DESIGN=WR FILETYPE=SAS;
NEST _ONE_ PATIENT;
WEIGHT _one_;
EVENT COMPLETE;
TIME EXTIME ;
OUTPUT KM KMTIME /FILENAME=fulldat REPLACE ;

However, if the last observation is changed to 

PATIENT	EXTIME	COMPLETE
6	613	0

then no error will occur.  Note that the estimate for S(613) will not be included in the output as the event status has been changed to censored. 

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.2	10.0.0
LINUX	9.0.2	10.0.0

There is no work around for this problem.

Example:

This program uses the same variable for stratum and TABLE, hence values of ROWPER are constant within strata and variances should be exactly 0.  Because of numerical round-off, the variances displayed are missing values.

proc crosstab data="c:\data\brfshort" filetype=sasxport design=strwr;
nest _ststr ;
weight _one_;
class _ststr;
print nsum rowper serow lowrow uprow/style=nchs nsumfmt=f6.0;

Variance Estimation Method: Taylor Series (STRWR)                              
by: _STATE AND _STRATA COMBINED.                                               
                                                                               
-------------------------------------------------------------------------      
_STATE AND _STRATA                                    Lower      Upper         
  COMBINED                                            95%        95%           
                       Sample   Row        SE Row     Limit      Limit         
                       Size     Percent    Percent    ROWPER     ROWPER        
-------------------------------------------------------------------------      
Total                    5838     100.00       0.00        .          .        
111                      1501      25.71        .          .          .        
131                      1964      33.64        .          .          .        
361                      2373      40.65        .          .          .        
-------------------------------------------------------------------------      

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.2	10.0.0
LINUX	9.0.2	10.0.0

Use PREDICTED=DEFAULT on the OUTPUT statement or specify the individual keywords associated with the PREDICTED keyword group on the OUTPUT statement.

Example:

The PROGRAMMER ERROR occurs with the following program:

PROC REGRESS data="c:\data\boric" filetype=sasxport R=EXCHANGE;
  NEST _ONE_ DAMID;
  WEIGHT _ONE_;

  REFLEVEL DOSEGRP = 1;
  SUBGROUP DOSEGRP;
  LEVELS   6;

  MODEL BW = DOSEGRP; 
  OUTPUT / PREDICTED=ALL FILENAME=”out” FILETYPE=SASXPORT REPLACE;

Avoid the PROGRAMMER ERROR with the following program (different OUTPUT statement):

PROC REGRESS data="c:\data\boric" filetype=sasxport R=EXCHANGE;
  NEST _ONE_ DAMID;
  WEIGHT _ONE_;

  REFLEVEL DOSEGRP = 1;
  SUBGROUP DOSEGRP;
  LEVELS   6;

  MODEL BW = DOSEGRP; 
  OUTPUT / PREDICTED=DEFAULT FILENAME=”out” FILETYPE=SASXPORT REPLACE;

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.2	10.0.0
LINUX	9.0.2	10.0.0

Include one or more of the following keywords on the TEST statement: CMH, ACMH or TCMH test on the TEST statement.

Example:

The following code will cause the PROGRAMMER ERROR described above:

PROC CROSSTAB DATA=wicdat FILEYTPE=sas DEFT1 ;
NEST stratum psu;
WEIGHT analwgt1;
CLASS racemom educ bfeed  ;
TABLE racemom*educ*bfeed;
TEST llchisq  / SATADJF; 
PRINT stestval sadjdf spval;

The preceding code can be modified by adding the CMH keyword to the TEST statement. This will prevent the PROGRAMMER ERROR from occurring:

PROC CROSSTAB DATA=wicdat FILEYTPE=sas DEFT1 ;
NEST stratum psu;
WEIGHT analwgt1;
CLASS racemom educ bfeed  ;
TABLE racemom*educ*bfeed;
TEST llchisq CMH / SATADJF; 
PRINT stestval sadjdf spval;

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.3	10.0.0
LINUX	9.0.3	10.0.0

The workaround is to add a new variable to the end of the dataset. This can be a dummy variable.

Example:

Say there is a simple data set that has the following variables (ordered):

Variables
Name         Type       Output     SAS        Description
                        Format     Format
----------------------------------------------------------
RACEMOM      Numeric    F2.0       R          Race-ethnicity
SITE         Numeric    F12.0                 Site
ANALWGT1     Numeric    F12.0                 Analysis
STRATUM      Numeric    F12.0                 STRATUM
RACE2        Numeric    F2.0       R          RACE2

The variable RACE2 is the last variable on the data set. SUDAAN will not properly apply the labels to this variable. 

The solution is to add a dummy variable to the end of the dataset. This can be done using the DATA step in SAS. For example:

data two;
 set one;
 varx=1;
 format race2 r.;
 format racemom r.;
run;

The dataset now contains the following variables:
 

Variables
Name         Type       Output     SAS        Description
                        Format     Format
----------------------------------------------------------
RACEMOM      Numeric    F2.0       R          Race-ethnicity
SITE         Numeric    F12.0                 Site
ANALWGT1     Numeric    F12.0                 Analysis
STRATUM      Numeric    F12.0                 STRATUM
RACE2        Numeric    F2.0       R          RACE2
VARX         Numeric    F12.0                 VARX

The variable VARX is now the last variable on the data set. Labels will be correctly applied to RACE2 as it is no longer the last variable on the dataset.

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.3
LINUX	9.0.3

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.3
LINUX	9.0.3

There isn't a workaround for this test using only CROSSTAB. A potential solution is to 
obtain the values of the SATADJCHI and the adjusted degrees of freedom from 
CROSSTAB and use a different software package to calculate the p-value. 

It is possible to use LOGISTIC to get this test in certain situations. 

Example:

The following example illustrates the problem and demonstrates a potential workaround. 

The 2-level categorical variable bfeed and the 3-level categorical variable educ are 
crossed in the TABLES statement and the Satterthwaite¡¯s adjusted chi is requested. 

PROC CROSSTAB data=ts.wicdat filetype=sas design=wr deft1;
NEST stratum psu;
WEIGHT analwgt1;
CLASS educ bfeed  ;
TABLE educ*bfeed;
TEST chisq llchisq / satadjchi; 
PRINT stestval sadjdf spval / spvalfmt=f8.5 sadjdffmt=f6.4 
filename="G:\Gordon\Tech Support\SATCHI\SAT example1" filetype=RTF replace;

These tests have more than 1 DF and generate the wrong p-value as is show in the output below:



Hypothesis Test
Test Statistic
Test Value
Adj DF
P-Value


CHISQ (Obs - Exp)
Satterthwaite-adj chi-sq
6.6706
1.6405
0.00265


LLCHISQ (Log-Lin Model)
Satterthwaite-adj chi-sq
6.4273
1.5701
0.00372



The correct p-values are calculated using a SAS DATA step. The Test Value and the 
Adj DF from the above table are needed to calculate the p-value. The CDF function in 
SAS is then used to calculate the correct p-values.  

DATA pvalue;
 pvalue_c=1-CDF('CHISQUARE', 6.6706 , 1.6405);
 pvalue_l=1-CDF('CHISQUARE', 6.4273 , 1.5701);
 
run;

proc print data=pvalue;
run;

The correct p-values are
	
	Obs    pvalue_c    pvalue_l
	  1     0.024057    0.025132

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.3
LINUX	9.0.3

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.3
LINUX	9.0.3

Try adding the USEVMEM=0 option to the PROC REGRESS statement to force SUDAAN 
not to use virtual memory for the analysis.  If SUDAAN is able to complete the 
analysis the variances will be correct.  If there is insufficient memory for the analysis, 
then no work-around is currently available.

Example:

If this problem occurs, add the USEVMEM statement to the PROC line of the code.

PROC REGRESS data=trend5 DESIGN=BRR USEVMEM=0;
REPWGT BRRWGT1-BRRWGT24; 

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.3
LINUX	9.0.3

If you have already specified a USEVMEM setting on the PROC statement, specify a 
smaller one.  If you have not already specified the USEVMEM setting you may need 
to experiment with the appropriate value to obtain satisfactory results.  The setting 
will depend on the proportion of records removed from the analysis.

Example:

Consider the following program. When run the results indicate that all of the 
replicates have been removed.   

PROC RLOGIST data=trend5 DESIGN=BRR;
REPWGT BRRWGT1-BRRWGT24; 

To correct the problem, the USEVMEM statement can be added to the PROC line. 
Keep reducing the value of the USEVMEM argument until the problem no longer 
occurs.   

PROC RLOGIST data=trend5 DESIGN=BRR USEVMEM=300;
REPWGT BRRWGT1-BRRWGT24; 

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.3
LINUX	9.0.3

Use the newer STESTS output group to output the CHISQ and LLCHISQ statistics.

Example:

Suppose the following program results in the floating point divide error just described:

proc crosstab data=master2 design=wr;
    nest strata_analysis;
    weight _one_;
    subgroup record_id censusdiv;
    levels   2         10;
    tables censusdiv*record_id;
    test chisq llchisq;
    print / tests=all style=nchs;
run;

Then simply replace the print statement above with this statement:
    print / stest=all style=nchs;

System	Release Reported	Release Fixed
PC	9.0.3	10.0.0
SUN	9.0.3
LINUX	9.0.3

The AVERAGE option is typically only used when the variable specified in the EFFECTS statement is involved in an interaction. This error will not occur if the variable on the EFFECTS is not involved in an interaction. Note that in this situation the AVERAGE option would have no affect on the results.

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

You can create the same contrasts using a series of either CONTRAST or DIFFVAR statements. Use the /NAME option following the statement to apply appropriate labels to the contrasts.

Example:

The following SAS program creates data and labels for the variable 'A', then uses this variable in two PROC DESCRIPT jobs. The labels for the levels of the PAIRWISE contrast in the first PROC DESCRIPT are incorrect. The second PROC DESCRIPT shows an alternate method of obtaining the contrasts with proper labels. Note that contrasted means and their SE's are identical in both programs.

proc format;
  value aval
    2 = 'value2'
    3 = 'value3'
    4 = 'value4'
  ;
run;

data dat;
  do a = 2 to 4;
    do b = 1 to 100;
      c = a + a*b;
      output;
    end;
  end;
  format a aval.;
run;

proc descript data=dat design=wr nomarg;
  nest _one_;
  weight _one_;
  var c;
  class a;
  pairwise a;
  print mean, semean/style=nchs;
run;

proc descript data=dat design=wr nomarg;
  nest _one_;
  weight _one_;
  var c;
  class a;
  diffvar a=(2 3) / NAME='value 2 vs value 3';
  diffvar a=(2 4) / NAME='value 2 vs value 4';
  diffvar a=(3 4) / NAME='value 3 vs value 4';
  print mean, semean/style=nchs;
run;

Incorrectly labeled results from the first DESCRIPT call using PAIRWISE:

Variance Estimation Method: Taylor Series (WR)
by: Variable, One, Contrast.

for: Variable = C.

----------------------------------------------
One                    Cntrst       SE Cntrst
   Contrast            Mean         Mean
----------------------------------------------
1
   CONTRAST_1:
     (1,value2)            -51.50        10.43
   CONTRAST_1:
     (1,value3)           -103.00        12.93
   CONTRAST_1:
     (value2,value3)       -51.50        14.46
----------------------------------------------

Appropriately labeled results from the second DESCRIPT call using DIFFVAR:

Variance Estimation Method: Taylor Series (WR)
by: Variable, One, CONTRAST.

for: Variable = C.

----------------------------------------------
One                    Cntrst       SE Cntrst
   CONTRAST            Mean         Mean
----------------------------------------------
1
   value 2 vs value
     3                     -51.50        10.43
   value 2 vs value
     4                    -103.00        12.93
   value 3 vs value
     4                     -51.50        14.46
---------------------------------------------- 

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

Any LOGISTIC model may also be run using SUDAAN's MULTILOG procedure. The only difference in output is that the signs on all parameter estimates are reversed in MULTILOG. The only change required to run the model in MULTILOG is to place the dependent variable on the CLASS statement. The defect described above is not present in MULTILOG.

Example:

The following (incorrect) output is produced using PROC RLOGIST:

Independent
  Variables and        Beta
  Effects              Coeff.          SE Beta
------------------------------------------------
Intercept                   -2.80         0.03
A       
  0                          0.00         0.00
  1                          1.05         0.07
  2                      69331.92      1984.23
  3                        799.09      4281.51
------------------------------------------------

The results for the same model in MULTILOG follow:

Independent
  Variables and        Beta
  Effects              Coeff.          SE Beta
--------------------------------------------------
0 vs 1
   Intercept               2.802989       0.027877
   A
     0                     0.000000       0.000000
     1                    -1.046960       0.067144
     2                    -3.620946       0.085806
     3                    -2.891921       0.133746
--------------------------------------------------

This LOGISTIC procedure is equivalent to the following MULTILOG procedure:

PROC LOGISTIC DATA=mydat FILETYPE=SUDAAN;
NEST STRATUM PSU;
WEIGHT WGT;
CLASS A;
REFLEVEL A=0;
MODEL Y = A;

PROC MULTILOG DATA=mydat FILETYPE=SUDAAN;
NEST STRATUM PSU;
WEIGHT WGT;
CLASS Y A;
REFLEVEL A=0;
MODEL Y = A;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

There is no direct work-around for this problem, however you can use PROC CROSSTAB to obtain the cell counts for the tables created by crossing the DESCRIPT or RATIO TABLES with the standardization variables. You can then check for 0-cells before using the DESCRIPT and RATIO procedures.

Example:

SUDAAN provides no warning if there are 0 cell frequencies during standardization in a RATIO or DESCRIPT program such as the following:

PROC DESCRIPT DATA=mydat DESIGN=WR;
NEST STRATUM PSU;
WEIGHT WT;
VAR X;
CLASS A B C;
TABLES A*B;
STDVAR C;
STDWGT .2 .3 .4 .1;

By running the following CROSSTAB program prior to the job above, you can check 
for 0 cell frequencies:

PROC CROSSTAB DATA=mydat DESIGN=WR;
NEST STRATUM PSU;
WEIGHT WT;
VAR X;
CLASS A B C;
TABLES A*B*C;
PRINT NSUM;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

Rename the variable _ONE_ to something else, or remove the NOPRINT option, and specify a PRINT statement. To minimize output you can specify printing of only one variable on only one record.

Example:

The following program will display a record count of 0 if the data set MYDAT contains the variable _ONE_.

PROC RECORDS DATA=MYDAT COUNTREC NOPRINT;

The following program will correctly display the record count even if _ONE_ is present on the data set.

PROC RECORDS DATA=MYDAT COUNTREC;
PRINT X/MAXREC=1;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

There is no work-around within SUDAAN for this problem. However, you can correct this error outside SUDAAN by multiplying the displayed design effect within each cell by (n-1/n) where n is the cell frequency.

Example:

If the displayed type 4 design effect within a cell is 1.235 and the cell frequency is 256 then the corrected design effect is 1.235 * (255/256) = 1.230.

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

This problem only occurs if you specify printing or output of type 4 design effects for row, column or total percents (keywords DEFFROW, DEFFCOL and DEFFTOT with the option DEFT4 explicitly set, or with no DEFT option set). There is no problem with design effects DEFT1, DEFT2 and DEFT3, so a partial work-around is to specify a different design effect on the PROC CROSSTB statement.

Another workaround would be to decrease the number of cells produced from the TABLES statement in CROSSTAB by splitting the current run of the procedure into 2 or more calls to the PROC.

Example:

The following CROSSTAB program may fail due to insufficient memory:

PROC CROSSTAB DATA=mydat FILETYPE=SUDAAN DESIGN=WR;
NEST STRATUM PSU;
WEIGHT WT;
SUBGROUP A;
LEVELS 800;
PRINT / TABLECELL=ALL;

Even though the DEFT keywords are not specified explicitly, they are implicitly requested by the use of TABLECELL=ALL. Since there is no DEFTn parameter on the PROC statement, DEFT4 is assumed by default. You can avoid the out-of-memory problem either by specifying a different design effect, or by eliminating the request for the design effect keywords by using TABLECELL=DEFAULT. The following program should run without memory errors.

PROC CROSSTAB DATA=mydat FILETYPE=SUDAAN DESIGN=WR DEFT1;
NEST STRATUM PSU;
WEIGHT WT;
SUBGROUP A;
LEVELS 800;
PRINT / TABLECELL=ALL;

An alternate workaround to the problem above would be to define two variables (for example), A1 and A2. A1 could be set to 1,...,400 corresponding to A=1,...,400 and A2 could be set 1,...,400 corresponding to A=401,...,800. To get the desired design effects, two calls to PROC CROSSTAB could be used as follows:

PROC CROSSTAB DATA=mydat FILETYPE=SUDAAN DESIGN=WR;
NEST STRATUM PSU;
WEIGHT WT;
SUBGROUP A1;
LEVELS 400;
PRINT / TABLECELL=ALL;

PROC CROSSTAB DATA=mydat FILETYPE=SUDAAN DESIGN=WR;
NEST STRATUM PSU;
WEIGHT WT;
SUBGROUP A2;
LEVELS 400;
PRINT / TABLECELL=ALL;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.2	9.0.3
LINUX	9.0.1	9.0.3

SUDAAN displays the following error message in PROC DESCRIPT and in PROC RATIO when there is a TABLES statement with the variable _ONE_ on it, _ONE_ is not on the SUBGROUP or the CLASS statement, and there is no CONTRAST, DIFFVAR, PAIRWISE or POLY statement in the program.

SEMANTIC ERROR
:
(Message 594)
No variables on the CONTRAST, DIFFVAR, PAIRWISE or POLYNOMIAL statement should also be on the
TABLES statement

If you wish to use _ONE_ on the TABLES statement in RATIO or DESCRIPT, always include it on the SUBGROUP or CLASS statement.

Example:

The following code will cause SUDAAN to display the SEMANTIC ERROR message quoted above:

PROC DESCRIPT DATA=mydat DESIGN=SRS;
CLASS A B;
TABLES _ONE_ A B;
VAR X;
PRINT / STYLE=NCHS;

Adding the variable _ONE_ to the CLASS statement or to a SUBGROUP statement avoids this problem:

PROC DESCRIPT DATA=mydat DESIGN=SRS;
CLASS A B _ONE_;
TABLES _ONE_ A B;
VAR X;
PRINT / STYLE=NCHS;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.3
LINUX	9.0.1	9.0.3

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.3
LINUX	9.0.1	9.0.3

For the first problem, use the SUBPOPN statement to create an output data set with only the first MAXOBS records in it. The following statement will suffice:

SUBPOPN _OBSNO_ <= maxobs_value;

For the second problem, do not use the MAXREC option when using the SORTBY statement unless there is also an OUTPUT statement, or another PRINT with MAXREC+STARTREC greater than or equal to the number of records in the data.

Example:

The following program will not work correctly, and will output all of the observations on the input data set:

PROC RECORDS DATA=mydat FILETYPE=SUDAAN MAXOBS=50;
OUTPUT / FILENAME=mydat50 FILETYPE=SPSS REPLACE;

The next program will correctly output only the first 50 records:

PROC RECORDS DATA=mydat FILETYPE=SUDAAN MAXOBS=50;
/* SUBPOPN statement limits the output to the first 50 records */
SUBPOPN _OBSNO_ <= 50;
OUTPUT / FILENAME=mydat50 FILETYPE=SPSS REPLACE;

Suppose mydat contains 300 records. The following program will not work correctly, and will sort only the first 40 records in the data:

PROC RECORDS DATA=mydat FILETYPE=SUDAAN;
SORTBY X Y;
PRINT / MAXREC=40;

The next program will correctly sort the entire data set, and print the first 40 records of sorted data.

PROC RECORDS DATA=mydat FILETYPE=SUDAAN;
SORTBY X Y;
PRINT / MAXREC=40;
/* Extra PRINT to force SUDAAN to sort all of the data */
PRINT / STARTREC=300 MAXREC=1;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.3
LINUX	9.0.1	9.0.3

Example:

In this example the variable XYZ is not on the input data set, but the following RFORAMT statement is included in the program.

RFORMAT XYZ YN.;
The SUDAAN output looks like this:
9   RFORMAT  XYZ
 
FILE ERROR
 in Statement 9
:
(Message 105)
Variable XYZ not found
 
PROGRAMMER ERROR
 in Object 88
:
NOTICE:
You may have found a bug in the SUDAAN software.
Please contact Research Triangle Institute with
complete information about this job so that we can
attempt to locate and correct the problem.


(Message 505)
Offset out of range.
Offset is -1 object type is OARRAY (12)
Expected type is 12
FATAL ERROR in SUDAAN

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.3
LINUX	9.0.1	9.0.3

In some cases a shorter version of the lengthy statement may be possible. For example use of the '-' and '--' notations to specify variable lists, and use of the '*' notation to specify repeated values may be helpful. These notations are discussed fully in the SUDAAN language manual in Section 2.6.7, Guidelines for Number Lists, Variable Names, Multi-way Tables, and Models.

Example:

Suppose you wish to specify a long list of CLASS variables which occur consecutively in the input data set. Rather than listing each variable name, you can use the double-hyphen notation to specify the first and last variables in the list only:

CLASS age_gp - income_gp;

If you wish to specify a long list of LEVELS parallel to a SUBGROUPS statement, use the '*' notation to shorten the list whenever there are consecutive duplicate values:

LEVELS 8*2 3 5 9*2;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.3
LINUX	9.0.1	9.0.3

If you use the EST_NO parameter on the PROC statement and specify a record count as least as great as the number of records on each of the multiply imputed data sets, then there is no problem

Example:

Suppose the multiply imputed data sets test1, test2, test3, test4, and test5 have record counts as follows:

TEST1		2467 records
TEST2		2480 records
TEST3		2450 records
TEST4		2495 records
TEST5		2477 records

Use of the EST_NO parameter on the PROC RLOGIST statement in the example program below avoids the problem described here.

PROC RLOGIST DATA=in.test1 FILETYPE=SAS DESIGN=WR MI_COUNT=5 EST_NO=2500;
NEST STRATUM PSU;
WEIGHT WGT;
MODEL EMPLOYED = AGE SEX RACE EDUC;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.3
LINUX	9.0.1	9.0.3

The only work around at present is to recode the existing variable or create a new one outside of SUDAAN. This variable should have the same values as the original variable on all records in the subpopulation, and should have a missing value on all records not in the subpopulation. If you then specify this variable on a CLASS statement with the option INCLUDE=NONMISSING, the analysis will be done correctly.

Example:

Suppose the variable A has values 0, 2, and 5 on the data set TEST, but that for the subpopulation X>0, the value 2 never occurs for the variable A. Then the following program will return an error even though within the subpopulation the CLASS variable A has only 2 levels:

SEMANTIC ERROR: The RISK statement is not available unless the last two dimensions of each TABLE request are 2x2.

PROC CROSSTAB DATA=TEST DESIGN=WR FILETYPE=SUDAAN;
NEST STRATUM PSU;
WEIGHT WGT;
CLASS A;
SUBGROUP B;
LEVELS 2;
SUBPOPN X>0;
TABLES A*B;
RISK OR;
PRINT / STRRISK=DEFAULT;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.3
LINUX	9.0.1	9.0.3

There is no work-around for this problem.

Example:

Suppose the CLASS variable EXPOSURE has actual levels 0, 30, 100 and 300. Then the row labels for EXPOSURE in the output STRRISK group should be these values. Instead the labels are 1, 2, 3, 4 in the following program:

PROC CROSSTAB DATA=Test FILETYPE=SUDAAN DESIGN=WR;
NEST STRATUM PSU;
WEIGHT WGT;
CLASS EXPOSURE BABYSEX BFEED;
TABLES EXPOSURE * BABYSEX * BFEED;
RISK /all;
OUTPUT / STRRISK=DEFAULT FILENAME=SRISK FILETYPE=SUDAAN REPLACE;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.3
LINUX	9.0.1	9.0.3

The same contrast can be specified using the CONTRAST statement. This statement does produce the correct estimates and variances.

Example:

The following code illustrates the problem and its work-around. The CONTRAST and DIFFVAR statements below should produce identical results, but in Releases 9.0.x they don't. The results produced by the CONTRAST statement are correct, and those produced by DIFFVAR are not. Until this bug is fixed, do not use the DIFFVAR statement in RATIO and DESCRIPT when specifying a cross of two or more simple contrasts.

SUBGROUP  A B;
LEVELS    2 4;
CONTRAST A=(1 -1) * B=(1 -1 0 0);
DIFFVAR  A=(1 2) * B=(1 2);

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.3
LINUX	9.0.1	9.0.3

Carefully check any REFLEVEL statements to ensure that all levels specified are valid.

Example:

The following program will produce the error described above, because 0 is an invalid level for the variable SEX. If the SEX variable is in fact coded 0-1, then it should be on the CLASS statement instead of the SUBGROUP statement. If this change is made, the REFLEVEL statement is correct. Alternately, if the SEX variable is coded 1-2, then the REFLEVEL statement must be changed to specify SEX=1 or SEX=2.

PROC REGRESS DATA=test DESIGN=WR FILETYPE=SAS;
NEST STRATUM PSU;
WEIGHT WGT;
SUBGROUP RACE SEX;
LEVELS   4    2;
REFLEVEL SEX=0;
MODEL Y = RACE SEX;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

When using multiple CLASS statements in the same procedure call, make sure all include the NOFREQS option, or that none include the NOFREQS option.

Example:

The following code within a single procedure call will cause a PROGRAMMER ERROR in SUDAAN:

CLASS RACE EDUC / NOFREQS INCLUDE=MISSING;
CLASS SEX;

The following code avoids a PROGRAMMER ERROR in SUDAAN because neither CLASS statement has the NOFREQS option:

CLASS RACE EDUC / INCLUDE=MISSING;
CLASS SEX;

The following code avoids a PROGRAMMER ERROR in SUDAAN because both CLASS statements have the NOFREQS option:

CLASS RACE EDUC / NOFREQS INCLUDE=MISSING;
CLASS SEX/NOFREQS;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

Supply a short label using the NAME=option on the SUBPOPN statement to avoid this problem.

Example:

The following SUBPOPN statement causes the buffer overflow which may lead to a PROGRAMMER ERROR or access violation:

SUBPOPN rfv13dn = 1455 and age < 18 and (spec3=1 or spec3=2 or spec3=3) 
                and othbactinfx = 2 and ((settype = 1 and (yr = 3 or yr = 4 or 
                yr = 5 or yr = 6 or yr = 7 or yr = 8)) or (settype = 2 and (yr = 3 or yr = 4 
                or yr = 5 or yr = 6 or yr = 7 or yr = 8)) or (settype = 3 and (yr = 7 or 
                yr = 8)));

Adding a NAME="label" option avoids the problem:

SUBPOPN rfv13dn = 1455 and age < 18 and (spec3=1 or spec3=2 or spec3=3) 
                and othbactinfx = 2 and ((settype = 1 and (yr = 3 or yr = 4 or 
                yr = 5 or yr = 6 or yr = 7 or yr = 8)) or (settype = 2 and 
                (yr = 3 or yr = 4 or yr = 5 or yr = 6 or yr = 7 or yr = 8)) or
                (settype = 3 and (yr = 7 or yr = 8))) / NAME="Valid Responses to Question4";

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

There is no work around for this problem.

Example:

Consider the following program statements which are part of a PROC LOGISTIC job. Here AGE_P is a continuous variable in the model.

   SUBGROUP SEX  EDUC_3  REGION  MARRY_3;
   LEVELS   2    3       4       3;

   REFLEVEL SEX=1 REGION=1 MARRY_3=1;
   MODEL CANTAF01 = AGE_P  SEX  EDUC_3  REGION  MARRY_3;

   CONDMARG AGE_P / AGE_P=(25 45 65 75);

The output for the conditional marginal looks like this:

EFFECT OF SEX, AGE, EDUCATION, REGION & MARITAL STATUS
ON CAN'T AFFORD MEDS

Marginals with AGE Continuous

by: Conditional Marginal #1.

---------------------------------------------------
Conditional Marginal #1       Conditional
                              Marginal           SE
---------------------------------------------------
Age, Age
  1,  25.0000                      0.0710    0.0038
  1,  45.0000                      0.0450    0.0018
  1,  65.0000                      0.0283    0.0017
  1,  75.0000                      0.0223    0.0018
---------------------------------------------------
NATIONAL HEALTH INTERVIEW SURVEY (NHIS, 1997)

The next table shows how the output should be labeled:

EFFECT OF SEX, AGE, EDUCATION, REGION & MARITAL STATUS
ON CAN'T AFFORD MEDS

Marginals with AGE Continuous

by: Conditional Marginal #1.

---------------------------------------------------
Conditional Marginal #1       Conditional
                              Marginal           SE
---------------------------------------------------
Age
   25.0000                         0.0710    0.0038
   45.0000                         0.0450    0.0018
   65.0000                         0.0283    0.0017
   75.0000                         0.0223    0.0018
---------------------------------------------------
NATIONAL HEALTH INTERVIEW SURVEY (NHIS, 1997)

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

Associate the desired format with the variable in a DATA step before using it on the SUBGROUP or CLASS statement in SUDAAN.

Example: The following statements create a SAS format which is associated with the variable DOSEGRP only at runtime, and the format DOSEGRP is not associated with the variable DOSEGRP on the output data set, as the partial output from PROC CONTENTS shows:

SAS Statements

PROC FORMAT; 
VALUE DOSEGRP 

  1='Control' 
  2='0.025%' 
  3='0.05%' 
  4='0.075%' 
  5='0.1%' 
  6='0.2%' ; 

RUN; 

PROC DESCRIPT DATA=IN.BORIC NOMARG DESIGN=WR NOPRINT; 
   NEST _ONE_ DAMID;
   WEIGHT _ONE_; 

   VAR BW; 
   SUBGROUP DOSEGRP;
   LEVELS 6; 

RFORMAT DOSEGRP DOSEGRP.;
OUTPUT NSUM WSUM MEAN SEMEAN TOTAL/ FILENAME=OUT.DSCTAB1 FILETYPE=SAS REPLACE;
RUN;

Partial Output From PROC CONTENTS:

                  Alphabetic List of Variables and Attributes
 
  #   Variable   Type   Len   Format   Label

  3   DOSEGRP    Num      8   12.      DOSEGRP
  8   MEAN       Num      8   8.4      Mean
  6   NSUM       Num      8   8.4      Sample Size
  1   PROCNUM1   Num      8   2.       Procedure Number
  9   SEMEAN     Num      8   8.4      SE Mean
  2   TABLENO    Num      8   2.       Table Number
 10   TOTAL      Num      8   8.4      Total
  4   VARIABLE   Num      8   3.       Variable
  7   WSUM       Num      8   8.4      Weighted Size
  5   _C1        Num      8   4.       Codes for Categorical Variable DOSEGRP

If you create a temporary data set which associated the format DOSEGRP with the variable DOSEGRP before executing the SUDAAN procedure, the format does get saved in the output data set.

Additional SAS Statements

DATA temp;
set in.boric;
format dosegrp dosegrp.;
run;

PROC DESCRIPT DATA=temp NOMARG DESIGN=WR NOPRINT;
  NEST _ONE_ DAMID;
  WEIGHT _ONE_;

  VAR BW;

  SUBGROUP DOSEGRP;
  LEVELS   6;

  RTITLE "Fetal Body Weight in a Teratology Study";
  OUTPUT NSUM WSUM MEAN SEMEAN TOTAL/ FILENAME=OUT.DSCTAB2 FILETYPE=SAS REPLACE;
  RUN;

Partial PROC CONTENTS Output
                  Alphabetic List of Variables and Attributes
 
 #   Variable   Type   Len   Format     Label

 3   DOSEGRP    Num      8   DOSEGRP.   DOSEGRP
 8   MEAN       Num      8   8.4        Mean
 6   NSUM       Num      8   8.4        Sample Size
 1   PROCNUM1   Num      8   2.         Procedure Number
 9   SEMEAN     Num      8   8.4        SE Mean
 2   TABLENO    Num      8   2.         Table Number
10   TOTAL      Num      8   8.4        Total
 4   VARIABLE   Num      8   3.         Variable
 7   WSUM       Num      8   8.4        Weighted Size
 5   _C1        Num      8   4.         Codes for Categorical Variable DOSEGRP

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

The only work-around is to ignore the expected values produced by MULTILOG when one or more covariates have missing values for a particular observation.

Example:

The partial output below illustrates the problem. The bolded values should all be set to missing by MULTILOG. Since they are not, users should ignore these estimates.

EXPECTED01   EXPECTED02   EXPECTED03   EXPECTED04       BABYSDAD      MOINCOM
-----------------------------------------------------------------------------
      0.62         0.16         0.19         0.03           1.00         8.00
      0.59         0.25         0.12         0.04           1.00         8.00
      0.64         0.12         0.22         0.01           1.00        11.00
      0.65         0.12         0.21         0.02           1.00         9.00
      0.48         0.27         0.10         0.15           1.00          .
      0.66         0.10         0.23         0.01           1.00        11.00
      0.66         0.09         0.24         0.01           1.00        12.00
      0.00         0.00         0.00         1.00            .           6.00
      0.00         0.00         0.00         1.00            .           1.00
      0.00         0.00         0.00         1.00            .           7.00
      0.55         0.29         0.10         0.06           1.00         6.00

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

If you have a related model which does converge in SUDAAN, and therefore have some idea of the appropriate values for the betas you can supply these values as initial values on the MODEL statement. This may prevent SUDAAN from producing the very large values for the parameters following the first iteration.

Example:

The following program illustrates the use of the BETAS option on the MODEL statement to specify initial values for the parameters.

proc survival data=in.surv1;
 nest strata psu;
 weight weight;
 event senvar;
 model t1 t2=x1 x2 x3/betas=(0.02 0.0 -0.1);
 run;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

There is not a work-around for this problem.

Example:

Incorrect standard errors will be generated using the statement:

PROC REGRESS data='myfile' R=EXCHANGE SEMETHOD=MODEL;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

The only general workaround is to shorten the label(s) and/or level stubs for the variable(s) in question, before using them in SUDAAN.

Example:

Suppose the variable X has a label such as:

"This is an example of an extremely long label for the variable X. This label is created to illustrate the problems which may occur in PROC DESCRIPT or PROC RATIO when such a variable is used in conjunction with the CATLEVEL or NUMCAT/DENCAT statements and one or more level stubs are also very long."

Suppose in addition that the level stubs for the variable are:

1 = "This is a very long level stub for the first level of the variable X"
2 = "This is an even longer level stub for the second level of variable X"
3 = "This is the longest level stub of all for the third level of the variable X"

PROC DESCRIPT DATA=mydata;
NEST STRATUM PSU;
WEIGHT WT;
VAR X X X;
CATLEVEL 1 2 3;
PRINT PERCENT SEPERCENT;

System	Release Reported	Release Fixed
PC	9.0.1	9.0.3
SUN	9.0.1	9.0.2
LINUX	9.0.1	9.0.2

C:\PROGRAM FILES\SUDAAN\Release900\SAS9Ind for the SAS9-Callable Individual version

Or

C:\PROGRAM FILES\SUDAAN\Release900\SAS9Net for the SAS9-Callable Network version

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	N/A	N/A

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

MI_VAR A1 A2 ….. A20;

Note that you may have as many MI_VAR statements as you need in a single SUDAAN program call. For further details and examples of the MI_VAR statement see the SUDAAN Language Manual, Section 3.7.5, Using Multiple Variables on the Same Data Set.

Example:
The following statements illustrate the use of the MI_VAR statement in CROSSTAB.

PROC CROSSTAB DATA=MI_ALL FILETYPE=SAS DESIGN=WR;
NEST STRATUM PSU;
MI_VAR A1-A20;
CLASS A1;
PRINT WSUM ROW SEROW/STYLE=NCHS;

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

DATA ERROR
:
(Message 410)
Invalid value for F-value parameter
fval=-3.78956126e-014 f1=1 f2=1000

Note that this DATA ERROR message follows the DATA WARNING message:

DATA WARNING
:
Degrees of freedom for OVERALL contrast are less than maximum number of estimable parameters
You may wish to rerun this job with a tolerance (TOL) of 1.000000e-007 and 1.000000e-005

Both messages are an indication that the data may not support the hypothesis tests requested. In future releases SUDAAN will produce a clearer error message for this problem.

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

Example:
The following program may have a PROGRAMMER ERROR:

PROC CROSSTAB DATA=TEST1 MI_COUNT=4;
NEST STRATUM PSU;
WEIGHT WGT;
SUBGROUP A;
LEVELS 4;

Changing the SUBGROUP and LEVELS statements, and adding an appropriate TABLES statement can avoid the error:

PROC CROSSTAB DATA=TEST1 MI_COUNT=4;
NEST STRATUM PSU;
WEIGHT WGT;
SUBGROUP _ONE_ A;
LEVELS 1 4;
TABLES _ONE_ * A;

If you are using the CLASS statement, the same type of solution works:

PROC CROSSTAB DATA=TEST1 MI_COUNT=4;
NEST STRATUM PSU;
WEIGHT WGT;
CLASS _ONE_ A;
TABLES _ONE_ * A;

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

Example:

The following program output illustrates the problem:

1. PROC DESCRIPT DATA="brfshort" FILETYPE=SAS
   NOPRINT;
2. NEST _STSTR _PSU;
3. WEIGHT _FINALWT;
4. SUBPOPN SEX=2 / NAME="females";
5. SUBGROUP MAMEXAM ;
6. LEVELS 2;
7. VAR PAPEXAM;
8. CATLEVEL 1;
9. OUTPUT / FILENAME="COVPCT" PCTCOV=ALL REPLACE;

SEMANTIC ERROR in Request 69:
(Message 517)
All input BATs must have the same nest

Instead of the OUTPUT statement use a PRINT statement such as the following:

PRINT / PCTCOV=ALL COVPCTFMT=F8.4;

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

Example:

The following program may end with a PROGRAMMER ERROR:

PROC RECORDS DATA=mysasxdata LEVFILE=mysasxlev FILETYPE=SASXPORT COUNTREC CONTENTS;

The next program, which is identical to the previous one except for the bolded text, does not end with a PROGRAMMER ERROR:

PROC RECORDS DATA=mysasxdata FILETYPE=SASXPORT COUNTREC CONTENTS;

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

Example:

The following program will lead to a PROGRAMMER ERROR:

PROC CROSSTAB DATA=FORSUDN1 MI_COUNT=5;
NEST SDPSTRA6 SDPPSU6 / MISSUNIT;
WEIGHT WTPFQX6;
SUBGROUP HSSEX POVERTY HAT28MI;
LEVELS 2 2 3;
TABLES   HAT28MI *HSSEX * POVERTY;
risk or;
print srisk;

The following program will NOT lead to a PROGRAMMER ERROR:

PROC CROSSTAB DATA=FORSUDN1 MI_COUNT=5;
NEST SDPSTRA6 SDPPSU6 / MISSUNIT;
WEIGHT WTPFQX6;
SUBGROUP HSSEX POVERTY HAT28MI;
LEVELS 2 2 3;
TABLES   HAT28MI *HSSEX * POVERTY;
risk or rr1;
print srisk;

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

Example:

The following program output illustrates the problem:
1   PROC CROSSTAB DATA="brfshort" FILETYPE=SAS NOPRINT; 
2   NEST _STSTR _PSU; 
3   WEIGHT _FINALWT; 
4   SUBPOPN SEX=2 / NAME="females"; 
5   SUBGROUP  MAMEXAM ;  
6   LEVELS    2;  
7   OUTPUT / FILENAME="COVROW" ROWCOV=ALL REPLACE;

SEMANTIC ERROR:
DDF, ESTIMATE and COVAR are required for COVAR output

Instead of the OUTPUT statement use a PRINT statement such as the following:

PRINT / ROWCOV=ALL COVROWFMT=F8.4;

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

Example:

Suppose the variable WGTVAR in your data set MYDATA has non-positive or missing values. You can use PROC RECORDS to create a data set with all weights positive as follows:

PROC RECORDS DATA=MYDATA NOPRINT;
SUBPOPN WGTVAR>0;
OUTPUT / FILENAME=MYPOSDATA FILETYPE=SUDAAN REPLACE;

The dataset MYPOSDATA can then be used safely in subsequent SUDAAN procedures.

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

Example:

The following program may lead to a PROGRAMMER ERROR:
PROC DESCRIPT DATA=MYMI1 MI_COUNT=3 NOPRINT DESIGN=SRS;
VAR X;
OUTPUT MEAN SEMEAN / BY_MI FILENAME=MIMEAN REPLACE;

The following program will avoid this PROGRAMMER ERROR:
PROC DESCRIPT DATA=MYMI1 MI_COUNT=3 DESIGN=SRS /*NOPRINT*/;
VAR X;
OUTPUT MEAN SEMEAN / BY_MI FILENAME=MIMEAN REPLACE;
PRINT MEAN SEMEAN;

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

  X6                                                                             
  0       0.10         0.92        -2.25         2.46         0.11     0.9136    
  1       0.11         0.68        -1.64         1.85         0.16     0.8808    
  2      -0.03         0.70        -1.82         1.77        -0.04     0.9712    
  3      -0.54         0.45        -1.70         0.62        -1.20     0.2844    
  4      -0.95         0.83        -3.09         1.19        -1.14     0.3051    
  5       0.14         0.55        -1.27         1.54         0.25     0.8112    

  6       0.00         0.00         0.00         0.00          .        .        
-----------------------------------------------------------------------------    

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

ERROR:  Read Access Violation In Task [ CROSSTAB )

Example:

Suppose the variable X in the data set BADFMT has no associated format. To associate a SAS format outside of SUDAAN and then use the variable in SUDAAN:

DATA GOODFMT;
SET BADFMT;
FORMAT X BEST6.;
RUN;

PROC CROSSTAB DATA=GOODFMT;

CLASS X;
RUN;
To associate a SAS format using the RFORMAT statement in SUDAAN:
PROC FORMAT;
VALUE XF;
  1=”One”
  2=”Two”
;
RUN;

PROC CROSSTAB DATA=BADFMT;

CLASS X;
RFORMAT X XF.;
RUN;

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

PROGRAMMER ERROR
 in Object 345
:
NOTICE:
You may have found a bug in the SUDAAN software.
Please contact Research Triangle Institute with
complete information about this job so that we can
attempt to locate and correct the problem.
  
(Message 505)
Offset out of range.
Offset is 0 object type is SARRAY (4)
Expected type is 4
FATAL ERROR in SUDAAN

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

Example:
To compute the estimates using DESIGN=SRS, simply comment out all design statements, and change the DESIGN parameter on the PROC statement to SRS to obtain the correct test on the maximum relative difference. Use your original design to produce variances and tests of hypothesis.

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

In all procedures frequency tables will display correctly for the following CLASS statement:

CLASS variable(s) / INCLUDE=MISSING SORT=INTERNAL DIR=DESCENDING;

Example:

Suppose the variable A has the following values:

	.:	70
	1:	150
	2:	115

The CLASS statement

CLASS A / INCLUDE=MISSING;

Produces the following truncated frequency table:

Frequencies and Values for CLASS Variables
by: Sample Variable A. 

------------------------------------
Sample                                  
Variable                                
A             Frequency      Value
------------------------------------
Ordered                               
Position:                                
1                  70            .

To see all levels of A correctly displayed, use the CLASS statement

CLASS A / INCLUDE=MISSING SORT=INTERNAL DIR=DESCENDING.

This CLASS statement produces the following frequency table:

Frequencies and Values for CLASS Variables 
by: Sample Variable A.

------------------------------------ 
Sample                                   
Variable                                  
A             Frequency      Value 
------------------------------------  
Ordered                               
Position:                                
1                   115          2    
Ordered                          
Position:                              
2                   150          1 
Ordered                           
Position:                          
3                    70          .

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	9.0.0	9.0.1
SUN	9.0.0	9.0.1

System	Release Reported	Release Fixed
PC	8.0.2	9.0.0
SUN	8.0.2	9.0.0

SUDAAN cannot handle SAS formats containing negative levels. This problem occurs with SAS-callable SUDAAN with SAS formats, and in standalone SUDAAN with a SAS format catalog in SASXPORT format.

For example, the following creates a format which cannot be handled by SUDAAN:

proc format library=library cntlout=out.fmtlib;
value BADSTUFF
-1 = "Negative Level"
0 = "0 level"
1 = "Positive level"

System	Release Reported	Release Fixed
PC	8.0.2	9.0.0
SUN	8.0.2	9.0.0

SUDAAN cannot properly interpret Filenames to on the PROC, PRINT and OUTPUT statements, if the path contains embedded ‘.’

For example, SUDAAN cannot properly handle the following:

PROC RECORDS DATA=”C:\TEST.DATA\MYDATA” FILETYPE=SPSS;

System	Release Reported	Release Fixed
PC	8.0.2	9.0.0
SUN	8.0.2	9.0.0

In cases where the individual(s) with the lowest end time all are associated with non-events, the hazard for these individuals should be 0. However, due to a logic error in the program, the hazard for these individuals may be set to a different value. There is no problem with the hazards assigned from the first time point associated with an event onward.

The following records illustrate the type of input that currently has a problem. In this case SUDAAN incorrectly assigns a non-zero hazard to the first person on the file.

System	Release Reported	Release Fixed
PC	8.0.1	9.0.0
SUN	8.0.1	9.0.0

System	Release Reported	Release Fixed
PC	8.0.1	9.0.0
SUN	8.0.1	9.0.0