Providing software solutions since 1976
Sample 25033: %SMSUB macro provides additional capabilities for PROC SURVEYMEANS
 |  |  |  |  |  |
%SMSUB macro provides additional capabilities for SAS/STAT PROC SURVEYMEANS
| Contents: | Purpose / History / Requirements / Usage / Details / See Also / References |
- PURPOSE:
-
Provides additional subgroup capabilities beyond those provided by the
DOMAIN statement in PROC SURVEYMEANS. This includes:
- presenting subgroup and overall estimates in one table (TABLES=),
- computing ratio estimates for subgroups (RATIO=),
- computing contrasts for means, totals, and ratios (CONTRAST=),
- restricting table requests to a subpopulation (SUBPOP=), and
- incorporating missing values into the variance computations for all available statistics (mean, total, and ratio).
(In the SAS 9.2 release, the NOMCAR option was added to the SURVEYMEANS procedure to incorporate missing values into the variance computations for the mean and total statistic. In a future SAS release, the procedure will also incorporate missing values into the variance computation for the ratio statistic.)
- HISTORY:
-
Version Update Notes 3.1 Has the same functionality as 3.0, but has minor modifications to maintain compatibility with the SAS 9.2 release. 3.0 Specify multiple table requests (TABLES=)
Specify multiple contrast requests (CONTRAST=)
Create contrasts for any statistic (mean, total, or ratio)
Create data sets from any output object (OUTPUT=)
Compute ratio estimates (RATIO=)
Create formats for subgroup variables (FORMAT=)
Create Microsoft Word tables (TABLES /STYLE=WORDREADY)
Specify numerical formats (STATISTICS /Fkeyword=format)
Suppress printing of output (STATISTICS /NOPRINT)
Override options coded in macro (OPTIONS=)2.2 Cosmetic improvements. 2.1 Added display of contrast vector. 1.3 Added CONTRAST= parameter
Lifted restriction on the number of analysis variables
Lifted restriction on maximum value of subgroup levels1.2 Added more comparison operators for SUBPOP= parameter. 1.1 Initial coding. - REQUIREMENTS:
- Base SAS and SAS/IML Software.
- USAGE:
-
Follow the instructions in the Downloads tab of this
sample to save the %SMSUB macro definition. Replace the text within quotes in the following statement with the location of the %SMSUB macro definition file on your system. In your SAS program or in the SAS editor window, specify this statement to define the %SMSUB macro and make it available for use:
%inc "<location of your file containing the SMSUB macro>";
Following this statement, you may call the %SMSUB macro. See the Results tab for some examples.
The following parameters are available in the %SMSUB macro.
%SMSUB( DATA= SAS-data-set, STATISTICS= keywords < / options >, STRATA= variable, CLUSTER= variable, POPSIZE= variable, WEIGHT= variable, VAR= variables, RATIO= request < ... & request > TABLES= requests < / options >, CONTRAST= 'label' variable vector < ... & 'label' variable vector >, SUBPOP= variable operator value, OUTPUT= object=SAS-data-set < ... object=SAS-data-set >, ALPHA= value, FORMAT= variable value1='format1' ... valuen='formatn' < ... & variable value1='format1' ... valuen='formatn' >, COLWIDTH= value, OPTIONS= options, TITLE= 'text' )Required Macro Parameters
The required parameters are DATA=, WEIGHT=, and either VAR= or RATIO=. The VAR= and RATIO= parameters cannot be used at the same time.
DATA= SAS-data-set specifies the data set to be analyzed. WEIGHT= variable specifies the analysis weight. VAR= variables specifies variables for computing mean or total statistics. RATIO= request < ... & request > specifies ratio requests. The variables used in the each request can be either continuous or categorical. For example, the following requests use both types of variables: ratio= y/x & c=1/c=3, The first request computes the weighted y-total over the weighted x-total. The second request computes the weight sum of observations in c=1 over the weight sum of observations in c=3. The ampersand is used to separate the two requests.Sample Design Parameters
STRATA= variable specifies the first-stage strata. CLUSTER= variable specifies the first-stage sample units for a clustered or multistage sample. POPSIZE= variable specifies a variable containing stratum population totals (or overall population if your sample is not stratified) for computing finite population correction factors. The POPSIZE= parameter should only be used to compute fpc factors for one-stage samples that were selected without replacement and with equal probabilities. (For more information on sample design capabilities and when to use the POPSIZE= parameter, see SAMPLE DESIGN SPECIFICATIONS.) For a stratified non-clustered sample, the population totals would be the number of observational units in each stratum population. For a stratified one-stage cluster sample, the population totals would be the number of clusters in each stratum population. The POPSIZE= variable can also contain special values that tell %SMSUB to use different variance formulas for different strata. The applicable variance formula for each stratum corresponds to the type of sampling performed in each stratum. The value "0" indicates sample units were selected with certainty (probability equal to one). The value "-1" indicates sample units were selected with replacement. For example, all three possible variance formulas could be specified using the following codes: --------------------------------------------------------------- | STRATA | POPSIZE | Variance Formula | --------------------------------------------------------------- | 1 | 0 | with certainty, variance equals zero | | 2 | -1 | with replacement, no fpc | | 3 | 1000 | without replacement, fpc=(1-n/1000) | ---------------------------------------------------------------Optional Macro Parameters
STATISTICS= keywords < / options >, specifies the statistics to be computed. If the VAR= parameter is used, the available statistics are: NOBS number of nonmissing observations NMISS number of missing observations SUMWGT weight sum of nonmissing observations MEAN mean SEMEAN standard error of mean CLMEAN confidence limits for mean CVMEAN coefficient of variation for mean TMEAN t-value for Ho: contrast-mean=0 PMEAN p-value for Ho: contrast-mean=0 TOTAL total SETOTAL standard error of total CLTOTAL confidence limits for total CVTOTAL coefficient of variation for total ALL all statistics If the RATIO= parameter is used, the available statistics are: NOBS number of nonmissing observations NMISS number of missing observations SUMWGT weight sum of nonmissing observations NUMTOTAL numerator total DENTOTAL denominator total RATIO ratio SERATIO standard error of ratio CLRATIO confidence limits for ratio CVRATIO coefficient of variation for ratio TRATIO t-value for Ho: contrast-ratio=0 PRATIO p-value for Ho: contrast-ratio=0 ALL all statistics The default statistics for the VAR=, RATIO=, and CONTRAST parameters are presented below: ------------------------------------------------------------ | Parameters | Default Statistics | ------------------------------------------------------------ | VAR= | NOBS MEAN SEMEAN CLMEAN | | VAR=, CONTRAST= | NOBS MEAN SEMEAN TMEAN PMEAN | | RATIO=, | NOBS RATIO SERATIO CLRATIO | | RATIO=, CONTRAST= | NOBS RATIO SERATIO TRATIO PRATIO | ------------------------------------------------------------ You can use the following options to control how the statistics are printed: Fkeyword=format specifies a numerical format for individual statistics. NOPRINT suppresses the printing of output. For example, if you wanted to change the default 12.4 format for SEMEAN to 12.6, you could specify the format as follows: statistics= nobs mean semean /fsemean=12.6, The NOPRINT option is helpful when used in combination with the OUTPUT= parameter. TABLES= requests < / options > requests one-way, two-way, or three-way subgroup tables, where the subgroup variables are joined by asterisks. For example, the following requests would generate all three tables types: tables= a a*b a*b*c, If you omit the TABLES= parameter, the macro produces a table with overall results. You can also request an overall table by using the TABLES= parameter with a request using the special keyword, _OVERALL_. You can use the following options to control how the table rows are constructed: LEVELMISS displays statistics for missing subgroup levels. EXCLMISSMARG excludes missing subgroup levels in the marginal estimates. NOMARG suppresses marginal estimates. See Example 6 for an illustration of when LEVELMISS and EXCLMISSMARG are important options to consider. You can also use the STYLE= option to produce standard box tables, or to produce listing output ready for conversion to WORD tables. STYLE=BOX presents tables in a standard box style (the default). STYLE=WORDREADY generates listing output ready to convert to WORD tables. For the STYLE=WORDREADY option, the output objects contain pound signs that act as column deliminators. You create WORD tables by first pasting your entire listing output into a WORD document. Using your pointing device, select an output table (excluding the table title), and click Table-Convert-Text to Table, which will open the table conversion dialog box. Then choose the following options: AutoFit behavior: AutoFit to contents AutoFormat... "choose a table style" Separate text at: Other # After selecting the dialog box options, click OK to create the table. See Example 2 for an illustration of the listing output created by the STYLE=WORDREADY option. CONTRAST= 'label' variable vector < ... & 'label' variable vector > estimates a linear contrast. If subgroup tables are also requested, then contrast estimates are computed for each level. For multiple contrast requests, ampersands are used to separate each contrast. For example, the parameters below test pairwise differences of drug use for a variable c, by each level of gender: var= druguse, table= gender, contrast= 'c: 1-2' c 1 -1 0 & 'c: 1-3' c 1 0 -1 & 'c: 2-3' c 0 1 -1, When contrasts are requested, the statistics associated with the "mean", "total", and "ratio" actually refer to the contrast mean, contrast total, and contrast ratio. SUBPOP= variable operator value SUBPOP= variable IN (values) restricts the analysis to a subpopulation of the overall population. The available comparison operators are listed below: ----------------------------------------------------------------- | Symbol | Mnemonic | Definition | | | Equivalent | | ----------------------------------------------------------------| | = | EQ | equal to | | ^= or ~= | NE | not equal to | | > | GT | greater than | | < | LT | less than | | >= | GE | greater than or equal to | | <= | LE | less than or equal to | | | IN | equal to one from a list of numbers | ----------------------------------------------------------------- For example, to restrict an analysis to people aged 25 or older, the syntax could be written using either the ">=" or "ge" operators: subpop= age >= 25, subpop= age ge 25, For the IN comparison operator, an example that restricts an analysis to certain age groups could be written as follows: subpop= agegrp in (3 4), OUTPUT= object=SAS-data-set < ... object=SAS-data-set >, creates output data sets from the macro output objects. The output objects are listed below: ---------------------------------------------------- | Object | Description | Invocation | --------------------------------------------------- | Summary | Data summary | always | | Tables | Table estimates | always | | Contrast | Contrast vectors | CONTRAST= | ---------------------------------------------------- For example, to create data sets for all three output objects, the OUTPUT= parameter could be written as follows: output= summary=mysummary contrast=mycontrast tables=mytables, The tables= data set will contain special codes for the marginal and missing categories of the subgroup variables. The value "0" identifies marginal categories. The value "9.9E16" identifies any missing categories if the table option LEVELMISS is specified. If you omit the OUTPUT= parameter, default output data sets are still created using the following names: _summary_, _contrast_, and _tables_. ALPHA= value specifies the confidence level for confidence limits. The default value of ALPHA=.05 produces 95% confidence limits.Optional Parameters for Output Appearance
FORMAT= variable value1='format1' ... valuen='formatn' < ... & variable value1='format1' ... valuen='formatn' > specifies value formats for subgroup variables in the TABLES= parameter. The ampersands separate the value formats given for each variable. For example, a three-way table could specify formats as follows: tables= agegrp*gender*race, format= agegrp 1='12-17' 2='18-34' & gender 1='Male' 2='Female' & race 1='White' 2='Black', OPTIONS= options overrides the SAS systems options coded inside the %SMSUB macro, which are: NOCENTER NONUMBER NODATE NONOTES NOSTIMER FORMDLIM=' ' FORMCHAR="|----|+|---+=|-/\<>*" ps=max ls=94 For example, to override the NODATE option, you would specify: OPTIONS=date, Only the date option would be changed. The remaining system options would remain in effect. COLWIDTH= value specifies the first column width in the output tables. The default is 26 spaces. TITLE= 'text' specifies a title for the output listing. - DETAILS:
-
Overview
The %SMSUB macro provides additional subgroup capabilities beyond those provided by the DOMAIN statement in PROC SURVEYMEANS. This includes:
- presenting subgroup and overall estimates in one table (TABLES=),
- computing ratio estimates for subgroups (RATIO=),
- computing contrasts for means, totals, and ratios (CONTRAST=),
- restricting table requests to a subpopulation (SUBPOP=), and
- incorporating missing values into the variance computations.
Because survey samples are from finite populations, the correct variance estimator for a given subgroup requires all the observations in your sample, including observations outside the subgroup. Consequently, elimination of observations outside the subgroup can often produce incorrect variance estimates. [For conditions that lead to incorrect variance estimates, refer to Korn, Graubard (1999, section 5.4).] Therefore, you should not subset your data set prior to analysis. Also, you should not use the BY statement or the WHERE statement to compute subgroup estimates.
The above consideration for subgroup variance estimation also applies when there is missing data in your data set. If your analysis variables contain missing values, the estimates for the mean, total, or ratio effectively estimate the mean, total, or ratio on the part of the population that, if sampled, would have given a response to the survey item. Analogous to any subgroup, the variance estimation for item-respondents requires all the observations in your sample, including the item-nonrespondents (i.e. the missing values). Of course, with nontrivial amounts of missing data, the total error will equal the variance of the estimator plus the squared bias, where the bias contribution from nonresponse is unknown.
For the mean and total statistics in PROC SURVEYMEANS, the subgroup variance estimators multiply the analysis weight by a zero-one subgroup indicator in each location the weight appears. However, the zero-one indicator does not identify the item response within the subgroup. Rather, it deletes the observations associated with the missing values. In regard to the ratio statistic, PROC SURVEYMEANS currently does not compute subgroup estimates.
The variance estimators in the %SMSUB macro for the mean, total, and ratio statistics start with the same formulas used in PROC SURVEYMEANS. But for each statistic, the analysis weight is multiplied by a zero-one indicator that identifies both the subgroup and item response membership. Thereby, the subgroup variance estimates are always computed over all sample observations. For details on subgroup estimation and variance estimation in the presence of missing values, refer to Cochran (1977), Levy/Lemeshow (1999), or Korn/Graubard (1999).
Acronyms in Documentation:
- PSU: Primary Sampling Unit
- refers to the first-stage sample units. For non-clustered samples, the PSUs are the observation units. For clustered samples, the PSUs are the clusters selected in the first stage.
- FPC: Finite Population Correction
- refers to the variance reduction multiplier, (1-fh), that accounts for sampling without replacement and with equal selection probabilities.
Sample Design Specifications
The %SMSUB macro provides the same sample design capabilities as the SURVEYMEANS procedure. This includes any sample design that falls under one of the following three major categories:
Number of Stages Replacement Method Selection Probabilities single-stage with replacement (WR) equal or unequal single-stage without replacement (WOR) equal multistage with replacement at first stage equal or unequal For multistage samples, the variance estimator only uses the between-PSU variance component and is unbiased if the PSUs are selected with replacement. The variance components for the subsequent sample-stage units within the PSUs are not needed because the PSUs are independent when selected with replacement. This corresponds to the classical analysis of variance for nested random models. [Refer to Cochran (1977, section 11.9).]
The table below expands the major categories into all the possible sample designs and presents the corresponding required sample design parameters:
Number of Stages Stage 1 Features Stage 1 Method Stage 1 Probs Required Design Parameters Single WR Equal or Unequal Single WOR Equal POPSIZE= Single Strata WR Equal or Unequal STRATA= Single Strata WOR Equal STRATA= POPSIZE= Single Clusters WR Equal or Unequal CLUSTER= Single Clusters WOR Equal CLUSTER= POPSIZE= Single Strata, Clusters WR Equal or Unequal STRATA= CLUSTER= Single Strata, Clusters WOR Equal STRATA= CLUSTER= POPSIZE= Multi Clusters WR Equal or Unequal CLUSTER= Multi Strata, Clusters WR Equal or Unequal STRATA= CLUSTER= The remaining major categories currently not supported by the %SMSUB macro (or the SURVEYMEANS procedure) are as follows:
Number of Stages Replacement Method Selection Probabilities single-stage without replacement unequal multistage without replacement at first stage equal multistage without replacement at first stage unequal The unbiased variance estimators for these sample designs require additional computation. For multistage samples using without replacement sampling at the first stage, the unbiased variance estimator not only adjusts the between-PSU variance component (with fpc factors if equal probabilities), but also requires computing additional variance components associated with subsequent sample-stage units. [Refer to Cochran, (section 10.4).] Also, if without replacement sampling is performed with unequal selection probabilities, the unbiased variance estimator requires incorporating the joint selection probabilities instead of computing fpc factors. [Refer to Cochran, (section 9A.7).]
At first glance, the variance estimation limitations for without replacement sampling may seem restrictive. However, a common practice is to approximate the variances for these more complicated designs by assuming the PSUs were selected with replacement. If the PSU sampling fractions are small (say, less than 10 percent), the approximation is very close to the variances that would be obtained from the unbiased estimators. In fact, for multistage samples with equal PSU selection probabilities, the approximation results in negligible overestimation, and for conservative practices, may even be preferable. [Refer to Korn and Graubard (1999, section 2.3).]
Note on POPSIZE=
As stated in the SYNTAX section, the POPSIZE= parameter can be used to compute fpc factors for single-stage samples selected without replacement and with equal probabilities. It should not be used for multistage samples. For example, if the POPSIZE= parameter were used on a multistage sample that had equal selection probabilities at the first stage, the variance estimates would correctly reduce the between-PSU variance component with fpc factors, but would be missing the additional variance components associated with subsequent sample-stage units. Hence, the variances would be underestimated.
- SEE ALSO:
-
The %SREGSUB macro provides subgroup capabilities
currently not available in PROC SURVEYREG. This includes:
- restricting the regression analysis to a subpopulation (SUBPOP= ), and
- incorporating missing values into the variance computations.
- REFERENCES:
-
Cochran, W.G. (1977), Sampling Techniques, Third Edition, New York: John
Wiley & Sons, Inc.
Levy, P.S, Lemeshow, S. (1999), Sampling of Populations, Third Edition, New York: John Wiley & Sons, Inc.
Korn, E.L., Graubard, B.I. (1999), Analysis of Health Surveys, New York: John Wiley & Sons, Inc.
These sample files and code examples are provided by SAS Institute Inc. "as is" without warranty of any kind, either express or implied, including but not limited to the implied warranties of merchantability and fitness for a particular purpose. Recipients acknowledge and agree that SAS Institute shall not be liable for any damages whatsoever arising out of their use of this material. In addition, SAS Institute will provide no support for the materials contained herein.
These sample files and code examples are provided by SAS Institute Inc. "as is" without warranty of any kind, either express or implied, including but not limited to the implied warranties of merchantability and fitness for a particular purpose. Recipients acknowledge and agree that SAS Institute shall not be liable for any damages whatsoever arising out of their use of this material. In addition, SAS Institute will provide no support for the materials contained herein.
- EXAMPLE 1: Hospital Survey -- Stratified Random Sample
-
Sample Design Parameters: STRATA=, POPSIZE=
Other Parameters: VAR=, TABLES=Three hospital strata, partitioned by level of obstetric services. Sample 15 hospitals from a population of 158 hospitals.
Estimate total number of births by level of obstetric services (1=basic, 2=intermediate, 3=tertiary).
Reference: Levy, P.S., Lemeshow, S. (1999), Sampling of Populations, Third Edition, New York: John Wiley & Sons, Inc. p. 135.
data hospsamp; input hospno oblevel sampsize tothosp births; weighta=tothosp/sampsize; datalines; 15 1 4 42 480 80 1 4 42 426 86 1 4 42 342 136 1 4 42 174 7 2 5 99 2022 26 2 5 99 576 62 2 5 99 1999 90 2 5 99 482 101 2 5 99 836 28 3 6 17 3108 34 3 6 17 4674 39 3 6 17 2539 102 3 6 17 1610 119 3 6 17 4618 149 3 6 17 1781 ; %inc "<location of your file containing the SMSUB macro>"; %smsub(data=hospsamp, statistics=nobs total setotal, strata=oblevel, popsize=tothosp, weight=weighta, var=births, tables=oblevel, title='--- Example 1 ---'); run; - RESULTS:
-
--- Example 1 --- The SURVEYMEANS SUBGROUP Macro Data Summary Number of Observations Read: 15 Weight Sum: 158 Number of Strata: 3 Number of PSUs: 15 Denominator Degrees of Freedom: 12 Table: oblevel Analysis Variable: births ----------------------------------------------------------------- |By: oblevel | | | Std Error | | | N | Total | of Total | |------------------------+------------+------------+------------| |Total | 15| 183983.0| 34014.3| |1 | 4| 14931.0| 2669.9| |2 | 5| 117117.0| 33067.7| |3 | 6| 51935.0| 7508.4| ----------------------------------------------------------------- - EXAMPLE 2: School Drug Use Survey -- Stratified Two-Stage Sample
-
Sample Design Parameters: STRATA=, CLUSTER=
Other Parameters: VAR=, TABLES/style=wordready, FORMAT=Two first-stage school strata, partitioned by region. Sample 12 schools in first stage. Sample six students per school in second stage (72 total).
Estimate mean drug use by agegrp*gender, and agegrp*race.
NOTE: The STYLE=WORDREADY option is used to generate listing output ready to convert to WORD tables. (See the TABLES= parameter for more information.)
data students; input studid$ region school gender race agegrp druguse @@; if region=1 then analwt=1380; else analwt=1120; datalines; 0101 1 01 2 2 2 1 0102 1 01 1 2 2 1 0103 1 01 1 1 2 1 0104 1 01 2 1 1 0 0105 1 01 1 2 2 1 0106 1 01 1 1 2 1 0201 1 02 2 1 2 0 0202 1 02 2 2 1 1 0203 1 02 2 1 1 1 0204 1 02 1 2 1 1 0205 1 02 2 2 2 1 0206 1 02 2 1 2 0 0301 1 03 2 1 1 1 0302 1 03 1 1 2 1 0303 1 03 1 2 1 0 0304 1 03 1 2 2 0 0305 1 03 2 1 1 0 0306 1 03 2 1 2 1 0401 1 04 1 1 1 1 0402 1 04 2 1 2 0 0403 1 04 2 2 2 1 0404 1 04 2 2 1 0 0405 1 04 1 2 1 1 0406 1 04 1 1 2 0 0501 1 05 2 2 2 0 0502 1 05 2 1 1 1 0503 1 05 1 2 2 1 0504 1 05 2 1 2 1 0505 1 05 1 2 1 0 0506 1 05 1 1 1 0 0601 1 06 1 1 2 0 0602 1 06 1 1 1 0 0603 1 06 2 2 1 1 0604 1 06 2 2 1 0 0605 1 06 1 1 2 1 0606 1 06 1 1 1 0 0701 2 07 1 1 1 1 0702 2 07 1 2 1 1 0703 2 07 1 1 1 0 0704 2 07 2 2 1 0 0705 2 07 2 2 2 1 0706 2 07 1 1 1 1 0801 2 08 2 2 2 0 0802 2 08 1 2 2 1 0803 2 08 1 2 1 0 0804 2 08 1 1 1 1 0805 2 08 1 1 2 0 0806 2 08 1 2 1 1 0901 2 09 2 1 1 0 0902 2 09 1 1 1 1 0903 2 09 2 1 2 0 0904 2 09 2 1 1 0 0905 2 09 2 1 1 0 0906 2 09 2 2 2 0 1001 2 10 1 2 1 0 1002 2 10 2 2 2 1 1003 2 10 2 2 1 1 1004 2 10 1 1 2 1 1005 2 10 2 1 2 1 1006 2 10 1 1 2 1 1101 2 11 2 1 1 0 1102 2 11 2 2 2 0 1103 2 11 1 2 2 1 1104 2 11 2 1 2 1 1105 2 11 1 1 2 0 1106 2 11 2 1 1 1 1201 2 12 1 2 2 0 1202 2 12 2 2 1 1 1203 2 12 1 1 1 1 1204 2 12 2 1 2 1 1205 2 12 1 2 2 0 1206 2 12 1 1 2 0 ; %inc "<location of your file containing the SMSUB macro>"; %smsub(data=students, statistics=nobs mean semean, strata=region, cluster=school, weight=analwt, var=druguse, tables=Agegrp*Gender Agegrp*Race /style=wordready, format= Agegrp 1='15-16' 2='17-18' & Gender 1='Male' 2='Female' & Race 1='White' 2='Black', title='--- Example 2 ---'); run; - RESULTS:
-
--- Example 2 --- The SURVEYMEANS SUBGROUP Macro Data Summary Number of Observations Read:#72 #Weight Sum:#90000 # Number of Strata:#2 Number of PSUs:#12 Denominator Degrees of Freedom:#10 Table: Agegrp*Gender Analysis Variable: druguse Agegrp#Gender#NOBS#MEAN#SEMEAN Total #Total #72 #0.5431 #0.0561 Total #Male #37 #0.5663 #0.0896 Total #Female #35 #0.5187 #0.0932 15-16 #Total #35 #0.5099 #0.0832 15-16 #Male #18 #0.5387 #0.1384 15-16 #Female #17 #0.4799 #0.1330 17-18 #Total #37 #0.5744 #0.1019 17-18 #Male #19 #0.5921 #0.1474 17-18 #Female #18 #0.5556 #0.1098 Table: Agegrp*Race Analysis Variable: druguse Agegrp#Race#NOBS#MEAN#SEMEAN Total #Total #72 #0.5431 #0.0561 Total #White #40 #0.5224 #0.0758 Total #Black #32 #0.5690 #0.0902 15-16 #Total #35 #0.5099 #0.0832 15-16 #White #20 #0.4947 #0.1024 15-16 #Black #15 #0.5297 #0.1009 17-18 #Total #37 #0.5744 #0.1019 17-18 #White #20 #0.5495 #0.1449 17-18 #Black #17 #0.6042 #0.1269 - EXAMPLE 3: Housing Development Survey -- Simple One-Stage Cluster Sample
-
Sample Design Parameters: CLUSTER=, POPSIZE=
Other Parameters: VAR=Sample two housing developments out of five. Interview all households from each sampled development.
Estimate total number persons over 65 in the five housing developments.
Reference: Levy, P.S., Lemeshow, S. (1999), Sampling of Populations, Third Edition, New York: John Wiley & Sons, Inc. p. 236.
data tab9_1a; input devlpmnt hh nge65 @@; wt1=2.5; M=5; datalines; 2 1 2 2 2 1 2 3 2 2 4 1 2 5 1 2 6 1 2 7 2 2 8 1 2 9 3 2 10 1 2 11 1 2 12 2 2 13 1 2 14 3 2 15 1 2 16 3 2 17 1 2 18 2 2 19 1 2 20 3 5 1 1 5 2 1 5 3 1 5 4 3 5 5 2 5 6 1 5 7 3 5 8 1 5 9 1 5 10 3 5 11 2 5 12 3 5 13 1 5 14 1 5 15 2 5 16 2 5 17 1 5 18 2 5 19 2 5 20 1 ; %inc "<location of your file containing the SMSUB macro>"; %smsub(data=tab9_1a, statistics=nobs total setotal, cluster=devlpmnt, popsize=M, weight=wt1, var=nge65, title='--- Example 3 ---') run; - RESULTS:
-
--- Example 3 --- The SURVEYMEANS SUBGROUP Macro Data Summary Number of Observations Read: 40 Weight Sum: 100 Number of Strata: 1 Number of PSUs: 2 Denominator Degrees of Freedom: 1 NOTE: Finite population correction to sample PSUs (devlpmnt) assumes inclusion of all listing units within sample PSUs. Table: Overall Analysis Variable: nge65 ----------------------------------------------------------------- |By: Overall | | | Std Error | | | N | Total | of Total | |------------------------+------------+------------+------------| |Total | 40| 167.5| 1.9| ----------------------------------------------------------------- - EXAMPLE 4: Health Status Survey -- Simple Random Sample
-
Sample Design Parameters: none
Other Parameters: VAR=, TABLES=, CONTRAST=, SUBPOP=, FORMAT=Perform pairwise comparisons of blood pressure between different levels of alcohol use. Restrict subpopulation to: age >= 25.
Variables:bp (systolic blood pressure) age (age, in years) exercise (1=regular exercise, 2=no regular exercise) alcohol (1=heavy use, 2=moderate use, 3=no use) data bp; input bp age exercise alcohol w @@; cards; 124 26 1 3 20 136 20 2 2 20 116 37 1 1 20 113 51 1 3 20 137 43 1 1 20 126 36 2 2 20 111 13 1 2 20 143 50 2 1 20 120 22 2 2 20 . 28 2 3 20 . 61 1 2 20 118 36 1 1 20 132 25 1 1 20 101 37 1 2 20 . 34 2 3 20 150 53 2 1 20 136 45 2 3 20 125 33 1 2 20 131 39 2 2 20 116 37 2 2 20 139 21 2 1 20 112 30 1 3 20 . 35 1 1 20 143 72 2 2 20 137 15 2 2 20 124 27 1 2 20 129 35 1 3 20 151 43 1 2 20 138 45 2 3 20 126 24 2 2 20 126 36 2 3 20 . 20 2 2 20 124 35 2 3 20 136 58 1 1 20 118 38 2 3 20 119 14 1 1 20 120 54 1 2 20 112 40 2 2 20 121 44 2 3 20 122 9 1 2 20 141 44 1 1 20 . 42 1 3 20 120 54 2 3 20 115 8 1 2 20 126 26 2 3 20 124 26 1 3 20 . 11 2 2 20 . 46 1 2 20 138 33 2 1 20 141 24 2 1 20 ; %inc "<location of your file containing the SMSUB macro>"; %smsub(data=bp, statistics=nobs mean tmean pmean/ fnobs=10.0 fmean=10.4 ftmean=10.2 fpmean=10.4, weight=w, var=bp, tables=Exercise, contrast='Alcohol 1-2' alcohol 1 -1 0 & 'Alcohol 1-3' alcohol 1 0 -1 & 'Alcohol 2-3' alcohol 0 -1 1, format= exercise 1='Regular Exercise' 2='No Exercise', subpop= age >= 25, title='--- Example 4 ---'); run; - RESULTS:
-
--- Example 4 --- The SURVEYMEANS SUBGROUP Macro Data Summary Subpopulation: age >= 25 Number of Observations Read: 50 Weight Sum: 1000 Subpopulation Observations: 38 Weight Sum: 760 Number of Strata: 1 Number of PSUs: 50 Denominator Degrees of Freedom: 49 Contrast Vector(s) Label Variable Values Coefficients Alcohol 1-2 alcohol 1 1 2 -1 3 0 Alcohol 1-3 alcohol 1 1 2 0 3 -1 Alcohol 2-3 alcohol 1 0 2 -1 3 1 Table: Exercise *Contrast Analysis Variable: bp ---------------------------------------------------------------------- |By: Exercise, Contrast | | | t Value | Pr > |t| | | | N | Mean | Mean=0 | Mean=0 | |------------------------+----------+----------+----------+----------| |Total | | | | | | Alcohol 1-2 | 19| 9.6556| 1.71| 0.0933| | Alcohol 1-3 | 22| 10.6325| 2.60| 0.0124| | Alcohol 2-3 | 23| -0.9769| -0.20| 0.8411| |Regular Exercise | | | | | | Alcohol 1-2 | 11| 5.8000| 0.71| 0.4840| | Alcohol 1-3 | 11| 9.6000| 1.93| 0.0596| | Alcohol 2-3 | 10| -3.8000| -0.49| 0.6294| |No Exercise | | | | | | Alcohol 1-2 | 8| 18.0667| 3.14| 0.0029| | Alcohol 1-3 | 11| 17.5417| 4.66| 0.0000| | Alcohol 2-3 | 13| 0.5250| 0.09| 0.9250| ---------------------------------------------------------------------- - EXAMPLE 5: Hospital Survey -- Simple Random Sample
-
Sample Design Parameters: POPSIZE=
Other Parameters: RATIO=, TABLES=Sample 56 hospitals from a population of 101 hospitals.
Estimate ratio of newborns whose mothers' hepatitis B surface antigen status (HBsAG) were recorded on the newborn medical records by level of obstetric services (1=basic, 2=intermediate, 3=tertiary).
Ratio Variables:agrecs (number of records having mother's antigen recorded) births (number of births) Reference: Levy, P.S., Lemeshow, S. (1999), Sampling of Populations, Third Edition, New York: John Wiley & Sons, Inc. p. 201.
data hospsamp; input hosp oblevel agrecs births @@; tothosp=101; w=101/56; cards; 1 2 2898 2898 2 2 1095 1304 3 2 1860 2022 4 2 1227 1395 5 2 618 773 6 2 1499 1630 7 2 1321 1436 8 2 0 2525 9 3 3595 4674 10 2 1732 2279 11 2 1020 1252 12 3 304 2539 13 2 1708 1780 14 3 1195 3320 15 3 3706 4029 16 2 837 1436 17 2 597 844 18 2 490 511 19 2 39 331 20 2 2438 2540 21 2 2222 2526 22 2 1859 1859 23 2 1999 1999 24 2 683 2136 25 2 0 1378 26 2 1004 1142 27 1 342 342 28 2 1050 1094 29 2 1614 1869 30 2 866 1204 31 2 520 651 32 2 771 836 33 3 2911 4044 34 2 1777 1777 35 3 2170 2261 36 2 793 863 37 3 1809 1809 38 2 1302 1302 39 3 2669 2669 40 3 1279 1512 41 2 434 1242 42 2 1178 1286 43 2 253 576 44 2 52 1324 45 2 985 1027 46 2 37 602 47 2 435 518 48 2 1225 1613 49 1 365 381 50 2 719 782 51 2 740 740 52 1 689 752 53 1 365 365 54 2 760 760 55 2 602 1369 56 2 651 651 ; %inc "<location of your file containing the SMSUB macro>"; %smsub(data=hospsamp, statistics=nobs ratio seratio, popsize=tothosp, weight=w, ratio=agrecs/births, tables=oblevel, title='---Example 5---'); - RESULTS:
-
---Example 5--- The SURVEYMEANS SUBGROUP Macro Data Summary Number of Observations Read: 56 Weight Sum: 101 Number of Strata: 1 Number of PSUs: 56 Denominator Degrees of Freedom: 55 Table: oblevel Ratio: agrecs/births ----------------------------------------------------------------- |By: oblevel | | | Std Error | | | N | Ratio | of Ratio | |------------------------+------------+------------+------------| |Total | 56| 0.7526| 0.0304| |1 | 4| 0.9571| 0.0137| |2 | 43| 0.7560| 0.0360| |3 | 9| 0.7312| 0.0593| ----------------------------------------------------------------- - EXAMPLE 6: Simple Random Sample
-
Sample Design Parameters: none
Other Parameters: VAR=, TABLES/levelmiss exclmissmarg, OUTPUT=Sample 14 people out of 500. Estimate proportion for variable v, by one-way tables of race and gender. Variables v, race, and gender variables have different missing value patterns.
Table Pros Cons Default Nobs in total row is consistent across both tables Nobs in cells do not add up to total row Levelmiss Nobs in total row is consistent across both tables
Nobs in cells add up to total row.Missing category may be of no interest ExclmissMarg Nobs in cells add up to total row. Nobs in total row is not consistent across both tables
Nobs in total row excludes some nonmissing values of vdata sample; input race gender v; w=500/14; datalines; . . . . . . . . 0 . . 1 . 1 1 . 2 0 1 1 1 1 2 0 1 1 0 1 2 1 2 1 1 2 2 0 2 1 1 2 2 1 ; %inc "<location of your file containing the SMSUB macro>"; *-----------------------* * Default Table * *-----------------------*; %smsub(data=sample, statistics=nobs mean semean /noprint, weight=w, var=v, tables=Race Gender, output= summary=summary tables=table6a); run; *-----------------------* * Levelmiss Table * *-----------------------*; %smsub(data=sample, statistics=nobs mean semean /noprint, weight=w, var=v, tables=race gender /levelmiss, output=tables=table6b); run; *-----------------------* * Exclmissmarg Table * *-----------------------*; %smsub(data=sample, statistics=nobs mean semean /noprint, weight=w, var=v, tables=race gender /exclmissmarg, output=tables=table6c); run; options ls=75; proc print data=summary; title'--- Example 6: Data Summary ---'; run; proc print data=table6a noobs; var tableindex variable race gender nobs mean semean; title'--- Default Table ---'; run; proc print data=table6b noobs; var tableindex variable race gender nobs mean semean; title'--- Levelmiss Table ---'; run; proc print data=table6c noobs; var tableindex variable race gender nobs mean semean; title'--- Exclmissmarg Table ---'; run; - RESULTS:
-
--- Example 6: Data Summary --- Obs NSTRATA NPSU DDF WSUM NOBSREAD 1 1 14 13 500 14 --- Default Table --- TABLEINDEX VARIABLE RACE GENDER NOBS MEAN SEMEAN 1 V 0 . 12 0.58333 0.14769 1 V 1 . 4 0.50000 0.25944 1 V 2 . 4 0.75000 0.22468 2 V . 0 12 0.58333 0.14769 2 V . 1 5 0.80000 0.18564 2 V . 2 5 0.40000 0.22736 --- Levelmiss Table --- TABLEINDEX VARIABLE RACE GENDER NOBS MEAN SEMEAN 1 V 0 . 12 0.58333 0.14769 1 V 1 . 4 0.50000 0.25944 1 V 2 . 4 0.75000 0.22468 1 V 9.9E16 . 4 0.50000 0.25944 2 V . 0 12 0.58333 0.14769 2 V . 1 5 0.80000 0.18564 2 V . 2 5 0.40000 0.22736 2 V . 9.9E16 2 0.50000 0.36690 --- Exclmissmarg Table --- TABLEINDEX VARIABLE RACE GENDER NOBS MEAN SEMEAN 1 V 0 . 8 0.625 0.17762 1 V 1 . 4 0.500 0.25944 1 V 2 . 4 0.750 0.22468 2 V . 0 10 0.600 0.16077 2 V . 1 5 0.800 0.18564 2 V . 2 5 0.400 0.22736
Right-click on the link below and select Save to save
the %SMSUB macro definition
to a file. It is recommended that you name the file
smsub.sas.
| Type: | Sample |
| Topic: | SAS Reference ==> Procedures ==> SURVEYMEANS Analytics ==> Survey Sampling and Analysis Analytics ==> Descriptive Statistics |
| Date Modified: | 2007-08-14 03:03:12 |
| Date Created: | 2005-01-17 12:13:02 |
Operating System and Release Information
| Product Family | Product | Host | SAS Release | |
| Starting | Ending | |||
| SAS System | SAS/STAT | All | 8 TS M0 | n/a |
