1. systematic macro SYSDATE the date of the SAS invocation (DATE7.) SYSDATE9 the date of the SAS invocation (DATE9.) SYSDAY the day of the week of the SAS invocation SYSTIME the time of the SAS invocation SYSENV FORE (interactive execution) or BACK (noninteractive or batch execution) SYSSCP an abbreviation for the operating system that is being used, such as OpenVMS, WIN, HP 300 SYSVER the release of SAS that is being used SYSJOBID an identifier for the current SAS session or for the current batch job (the user ID or job name for mainframe systems, the process ID (PID) for other systems) Some automatic macro variables have values that automatically change based on submitted SAS statements. SYSLAST the name of the most recently created SAS data set, in the form LIBREF.NAME. This value is always stored in all capital letters. If no data set has been created, the value is _NULL_ SYSPARM text that is specified when SAS is invoked SYSERR contains a return code status that is set by the DATA step and some SAS procedures to indicate if the step or procedure executed successfully 2.user-defined macro The %LET Statement Displaying Macro Variable Values in the SAS Log The SYMBOLGEN Option The %PUT Statement _ALL_ Lists the values of all macro variables _AUTOMATIC_ Lists the values of all automatic macro variables _USER_ Lists the values of all user-defined macro variables %STR function hides the normal meaning of a semicolon and other special tokens and mnemonic equivalents of comparison or logical operators so that they appear as constant text. %NRSTR function performs the same quoting function as %STR, except it also masks macro triggers ( and % %BQUOTE function is used to mask (or quote) special characters and mnemonic operators Macro Functions The %UPCASE Function The %QUPCASE Function The %SUBSTR Function: enables you to extract part of a character string from the value of a macro variable The %QSUBSTR Function The %INDEX Function The %SCAN Function:enables you to extract words from the value of a macro variable The %QSCAN Function The %SYSFUNC Function Creating a Macro Variable During DATA Step Execution CALL SYMPUT(macro-variable,text); CALL SYMPUT(‘macro-variable’, ‘text’); CALL SYMPUT(‘macro-variable’,DATA-step-variable); CALL SYMPUTX(macro-variable,expression); CALL SYMPUT(‘macro-variable’,expression); PUT(source,format.) CALL SYMPUT(expression1,expression2); Creating Macro Variables During PROC SQL Step Execution PROC SQL NOPRINT; SELECT column1,column2,... INTO :macro-variable-1,:macro-variable-2,... FROM table-1 | view-1 WHERE expression other clauses; QUIT; PROC SQL NOPRINT; SELECT column1 INTO :macro-variable-1 SEPARATED BY ‘delimiter1’ FROM table-1 | view-1 WHERE expression other clauses; QUIT; Working with PROC SQL Views proc sql; create view subcnum as select student_name, student_company, paid from sasuser.all where course_number=input(symget(’crsnum’),2.); quit; %let crsnum=4; proc print data=subcnum noobs; title "Status of Students in Course Number crsnum"; run; Using Macro Variables in SCL Programs Referencing Macro Variables Indirectly Obtaining Macro Variable Values During DATA Step Execution SYMGET(macro-variable) CALL SYMPUTN(‘macro-variable’, value); SCL-variable = SYMGETN(‘macro-variable’); 3.macro??? Defining a Macro %MACRO macro-name; text %MEND macro-name; OPTIONS MCOMPILENOTE= NONE | NOAUTOCALL | ALL; Monitoring Execution with System Options OPTIONS MPRINT | NOMPRINT; -------- specified, the text that is sent to the SAS compiler as a result of macro execution is printed in the SAS log OPTIONS MLOGIC | NOMLOGIC; --------- prints messages that indicate macro actions that were taken during macro execution. 4.????macro??? (1)Positional Parameters %MACRO macro-name(parameter-1,...,parameter-n); text %MEND macro-name; (2)Keyword Parameters %MACRO macro-name(keyword-1=value-1,...,keyword-n=value-n); text %MEND macro-name; (3)Mixed Parameter Lists %MACRO macro-name(parameter-1,...,parameter-n, keyword-1=value-1,...,keyword-n=value-n); text %MEND; PARMBUFF option in a macro definition to create a macro that can accept a varying number of parameters at each invocation. 5. Global and local You can create a global macro variable with a %LET statement (used outside a macro definition) a DATA step that contains a SYMPUT routine a DATA step that contains a SYMPUTX routine (beginning in SAS 9) a SELECT statement that contains an INTO clause in PROC SQL The %GLOBAL statement -creates one or more macro variables in the global symbol table and assigns null values to them -can be used either inside or outside a macro definition - has no effect on variables that are already in the global symbol table. %GLOBAL macro-variable-1 ...macro-variable-n; local macro variables with parameters in a macro definition a %LET statement within a macro definition a DATA step that contains a SYMPUT routine within a macro definition a DATA step that contains a SYMPUTX routine within a macro definition (beginning in SAS 9) a SELECT statement that contains an INTO clause in PROC SQL within a macro definition a %LOCAL statement. The %LOCAL statement - can appear only inside a macro definition - creates one or more macro variables in the local symbol table and assigns null values to them - has no effect on variables that are already in the local symbol table Remember, the SYMPUT routine and the SYMPUTX routine can create local variables only if the local table already exists. it is possible to have a local macro variable and a global macro variable that have the same name and different values. OPTIONS MPRINTNEST | NOMPRINTNEST; ----------- allows the macro nesting information to be written to the SAS log in the MPRINT output. This has no effect on the MPRINT output that is sent to an external file. OPTIONS MLOGICNEST | NOMLOGICNEST; -----------allows the macro nesting information to be displayed in the MLOGIC output in the SAS log. 6. Structured statements (1)Conditionally %IF expression %THEN text; %ELSE text; %IF expression %THEN %DO; text and/or macro language statements %END; %ELSE %DO; text and/or macro language statements %END; (2)Iteratively %DO index-variable=start %TO stop %BY increment; text %END; (3)Using Arithmetic and Logical Expressions %EVAL(arithemetic or logical expression); evaluates integer arithmetic or logical expressions. Logical expressions and arithmetic expressions are sequences of operators and operands forming sets of instructions that are evaluated to produce a result. %SYSEVALF(expression, conversion-type); evaluates arithmetic and logical expressions using floating-point arithmetic.
1. create sample 2. create index 3. combine data 1. create sample (1) systematic sample of known # of obs data sasuser.subset; do pickit=1 to 142 by 15; set sasuser.revenue point=pickit; output; end; stop; run; (2) systematic sample of unknown # of obs data sasuser.subset; do pickit=1 to totobs by 10; set sasuser.revenue point=pickit nobs=totobs; output; end; stop; run; (3) random sample with replacement data work.rsubset(drop=1 sampsize); sampsize=10; do i=1 to sampsize; pickit=ceil(ranuni(0)*totobs); set sasuser.revenue point=pickit nobs=totobs; output; end; stop; run; proc print data=work.rsubset label; title 'A Random Sample with Replacement'; run; (4)random sample without replacement data work.rsubset(drop=obsleft sampsize); sampsize=10; obsleft=totobs; do while(sampsize0); pickit+1; if ranuni(0)sampsize/obsleft then do; set sasuser.revenue point=pickit nobs=totobs; output; sampsize=sampsize-1; end; obsleft=obsleft-1; end; stop; run; proc print data=work.rsubset label; title 'A Random Sample without Replacement'; run; 2. create index in data step manage index with proc datasets manage index with proc sql 3. combine data (1) filename statement filename qtr1('add1' 'add2' 'add3'); data work.firstqtr; infile qtr1; input Flight $ Origin $ Dest $ Date: date9. Revcargo: comma15.2; run; (2) infile statement data quarter (drop=monthnum midmon lastmon); monthnum=month(today()); midmon=month(intnx('month', today(), -1)); lastmon=month(intnx('month',today,-2)); do i=mnthnum, midmon, lastmon; nextfile=""!!compress(put(i,2.)!!".dat",' '); do until(lastobs); infile temp filevar=nextfile end=lastobs; input Flight $ Origin $ Dest $ Date: date9. Revcargo: comma15.2; output; end; stop; run; (3)proc append proc append base=work.acities data=work.airports force; run; (4) if-then/else statement data mylib.employees_new; set mylib.employees; if IDnum=1001 then Birthdate='01JAN1963'd; else if IDnum=1002 then Birthdate='08AUG1946'd; else if IDnum=1003 then Birthdate='23MAR1950'd; else if IDnum=1004 then Birthdate='17JUN11973'd; run; (5) array statement data mylib.employees_new; array birthdates{1001:1004} _temporary_ ( '01JAN1963'd '08AUG1946'd '23MAR1950'd '17JUN11973'd ); set mylib.employees; Bithdate=birthdates(IDnum); run; (6)format procedure proc format; value $birthdate '1001'= '01JAN1963'd '1002'='08AUG1946'd '1003'='23MAR1950'd '1004'='17JUN11973'd; run; data mylib.employees_new; set mylib.employees; Birthdate=input(put(IDnum,$birthdate.),date9.); run; (7) match-merge proc sort data=sasuser.expenses out=expenses; by flightid date; run; proc sort data=sasuser.revenue out=expenses; by flightid date; run; datarevexpns (drop=rev1st revbusiness revecon expenses); merge expenses(in=e) revenue(in=r); by flightid date; if e and r; Profit=sum(rev1st, revbusiness, revecon, -expenses); run; data sasuser.alldata; merge revexpns (in=r) acities (in=a rename=(code=dest) keep=city name code); by dest; if r and a; run; (8) sql proc sql; create table sqljoin as select revenue.flightid, revenue.date format=date9., revenue.origin, revenue.dest, sum(revenue.rev1st, revenue.revbusiness, revenue.revecon)-expenses.expenses as Profit, acities.city, acities.name from sasuser.expenses, sasuser.revenue, sasuser.acities where expenses.flightid=revenue.flightid and expenses.date=revenue.date and acities.code=revenue.dest order by revenue.dest, revenue.flightid, revenue.date; quit; (9) many-to-many match proc sql; create table flightemp as select flightschedule.*, firstname, lastname from sasuser.flightschedule, sasuser.flightattendants where flightschedule.empid=flightattendants.empid; quit; data fightemps3(drop=empnum jobcode) set sasuser.flightschedule; do i=1 to num; set sasuser.flightattendants(rename=(empid=empnum)) nob=num point=1; if empid=empnum then output; end; run; (10) summary data and detail data proc means data=sasuser.monthsum noprint; var revcargo; output out=sasuser.summary sum=Cargosum; run; data sasuser.percent1; if _n_=1 then set sasuser.summary(keep=cargosum); set sasuser.monthsum(keep=salemon revcargo); PctRev=revcargo/cargosum; run; data sasuser.percent2(drop=totalrev); if _n_=1 then do until(lastobs); set sasuser.monthsum(keep=revcargo) end=lastobs; totalrev+revcargo; end; set sasuser.monthsum (keep=salemon revcargo); PctRev=revcargo/totalrev; run; (11)index data work.profit work.errors; set sasuser.dnunder; set sasuser.sale200(keep=routeid flightid date rev1st revbusiness revecon revcargo) key=flightdate; if _iorc_=0 then do; Profit=sum(rev1st, revbusiness, revecon, revcargo, -expenses); output work.profit; end; else do; _error_=0; output work.errors; end; run; (12) multidimensional array data work.wndchill(drop=column row); array WC {4,2} _temporary_(-22, -16, -28, -22, -32, -26, -35, 29); set sasuser.flights; row=round(wspeed,5)/5; colunm=(round(temp,5)/5)+3; WindChill=wc{row, column}; run; (13) stored array values data work.lookup1; array Targets{1997:1999,12} _temporary_; if _n_=1 then do i=1 to 3; set sasuser.ctargets; array Mnth{*} Jan--Dec; do j=1 to dim(mnth); targets{year,j}=mnth{j}; end; end; set sasuser.monthsum(keep=salemin revargo monthno); year=input(substr(salemon,4),4.); Ctarget=targets{year,monthno}; format ctarget dollar15.2; run; (14) transpose and merge proc transpose data=sasuser.ctargets out=work.ctarget2 name=Month prefix=Ctarget; by year; run; proc sort data=work.ctarget2; by year month; run; data work.mnthsum2; set sasuser.monthsum(keep=SaleMon RevCargo); length Month $ 8; Year=input(substr(SaleMon,4),4.); Month=substr(SaleMon,1,1)||lowcase(substr(SaleMon,2,2)); run; proc sort data=work.mnthsum2; by year month; run; data work.merged; merge workmnthsum2 work.ctarget2; by year month; run;
come from http://www.ats.ucla.edu/stat/sas/faq/bootstrap.htm SAS FAQ: How can I bootstrap estimates in SAS? Bootstrapping allows for estimation of statistics through the repeated resampling of data. In this page, we will demonstrate several methods of bootstrapping a confidence interval about an R-squared statistic in SAS. We will be using the hsb2 dataset that can be found here. We will begin by running an OLS regression, predicting read with female, math, write, and ses, and saving the R-squared value in a dataset called t0. The R-squared value in this regression is 0.5189. ods output FitStatistics = t0; proc reg data = hsb2; model read = female math write ses; run; quit; The REG Procedure Model: MODEL1 Dependent Variable: read reading score Number of Observations Read 200 Number of Observations Used 200 Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr F Model 4 10855 2713.73294 52.58 .0001 Error 195 10064 51.61276 Corrected Total 199 20919 Root MSE 7.18420 R-Square 0.5189 Dependent Mean 52.23000 Adj R-Sq 0.5090 Coeff Var 13.75493 Parameter Estimates Parameter Standard Variable Label DF Estimate Error t Value Pr |t| Intercept Intercept 1 6.83342 3.27937 2.08 0.0385 female 1 -2.45017 1.10152 -2.22 0.0273 math math score 1 0.45656 0.07211 6.33 .0001 write writing score 1 0.37936 0.07327 5.18 .0001 ses 1 1.30198 0.74007 1.76 0.0801 *store the estimated r-square; data _null_; set t0; if label2 = "R-Square" then call symput('r2bar', cvalue2); run; To bootstrap a confidence interval about this R-squared value, we will first need to resample. This step involves sampling with replacement from our original dataset to generate a new dataset the same size as our original dataset. For each of these samples, we will be running the same regression as above and saving the R-squared value. proc surveyselect allows us to do this resampling in one step. Before carrying out this step, let's outline the assumptions we are making about our data when we use this method. We are assuming that the observations in our dataset are independent. We are also assuming that the statistic we are estimating is asymptotically normally distributed. We indicate an output dataset, a seed, a sampling method, and the number of replicates. The sampling method indicated, urs, is unrestricted random sampling, or sampling with replacement. The samprate indicates how large each sample should be relative to the input dataset. A samprate of 1 means that the sampled datasets should be of the same size as the input dataset. So in this example, we will generate 500 datasets of 200, so our output dataset bootsample will have 100,000 observations. %let rep = 500; proc surveyselect data= hsb2 out=bootsample seed = 1347 method = urs samprate = 1 outhits rep = rep; run; ods listing close; The SURVEYSELECT Procedure Selection Method Unrestricted Random Sampling Input Data Set HSB2 Random Number Seed 1347 Sampling Rate 1 Sample Size 200 Expected Number of Hits 1 Sampling Weight 1 Number of Replicates 500 Total Sample Size 100000 Output Data Set BOOTSAMPLE With this dataset, we will now run our regression model, specifying by replicate so that the model will be run separately for each of the 500 sample datasets. After that, we use a data step to convert the R-squared values to numeric. ods output FitStatistics = t (where = (label2 = "R-Square")); proc reg data = bootsample; by replicate; model read = female math write ses; run; quit; * converting character type to numeric type; data t1; set t; r2 = cvalue2 + 0; run; Method 1: Normal Distribution Confidence Interval We will first create a confidence interval using the normal distribution theory. This assumes that the R-squared values follow a t distribution, so we can generate a 95% confidence interval by about the mean of the R-squared values based on quantiles from a t-distribution with 499 degrees of freedom. We find the critical t values for our confidence interval and multiply these by the standard deviation of the R-squared values that arose in our 500 replications. Our confidence interval using this method is symmetric about the R-squared value we saw in our original regression. We can see that the 95% confidence interval using this method is (0.432787, 0.605013). We have also calculated the bias in our original value of R-squared as the difference between that value and the mean of the 500 R-squareds in our bootstrap sample. * creating confidence interval, normal distribution theory method; * using the t-distribution; %let alphalev = .05; ods listing; proc sql; select r2bar as r2, mean(r2) - r2bar as bias, std(r2) as std_err, r2bar - tinv(1-alphalev/2, rep-1)*std(r2) as lb, r2bar + tinv(1-alphalev/2, rep-1)*std(r2) as hb from t1; quit; r2 bias std_err lb hb 0.5189 0.006616 0.043829 0.432787 0.605013 Method 2: Percentile Confidence Interval Another way to generate a bootstrap 95% confidence interval from the sample of 500 R-squared values is to look at the 2.5th and 97.5th percentiles in this distribution. This approach to the confidence interval has some advantages over the normal approximation used above. This interval is not symmetric about the original estimate of the R-squared and this method is unaffected by monotonic transformations on the estimated statistic. The first advantage is relevant because our original estimate is subject to bias. The second advantage is less relevant in this example than in an instance where the estimate might be subject to a transformation. The bootstrap estimates that form the bounds of the interval can be transformed in the same way to create the bootstrap interval of the transformed estimate. We can easily generate a percentile confidence interval in SAS using proc univariate after creating some macro variables for the percentiles of interest and using them in the output statement. We can see that the confidence interval from this method is (0.436, 0.6017). Since we have put the information of interest into a new dataset, pmethod, we have omitted the standard output from the proc univariate. %let alphalev = .05; %let a1 = %sysevalf(alphalev/2*100); %let a2 = %sysevalf((1 - alphalev/2)*100); * creating confidence interval, percentile method; proc univariate data = t1 alpha = .05; var r2; output out=pmethod mean = r2hat pctlpts=a1 a2 pctlpre = p pctlname = _lb _ub ; run; ... output omitted ... data t2; set pmethod; bias = r2hat - r2bar; r2 = r2bar; run; ods listing; proc print data = t2; var r2 bias p_lb p_ub; run; Obs r2 bias p_lb p_ub 1 0.5189 .0066164 0.436 0.6017 Method 3: Bias-Corrected Confidence Interval We can also correct for bias in calculating our confidence interval. We have calculated bias in the previous method as the difference between the R-squared we observed in our initial regression and the mean of the 500 R-squared values from the bootstrap samples. The R-squared estimate from the initial regression is assumed to be an unbiased estimate of the true R-squared. If we wish to correct for the bias in calculating our confidence interval, we can go through the steps below. These are described by Cameron and Trivedi in Microeconomics Using Stata. We first calculate the proportion of the bootstrap R-squareds that are less than our original value. We will adjust the percentiles used to define our confidence interval based on how this proportion differs from 0.5. We then find the probit of this proportion (z0) and the proportion associated with our alpha level (zalpha). Next, we calculate the two percentiles that will be used to find our confidence interval, p1 and p2, from these values. We then calculate our interval with proc univariate. From this method, our interval is (0.40575, 0.5936). %let alphalev = .05; %let alpha1 = %sysevalf(1 - alphalev/2); %put alpha1; proc sql; select sum(r2=r2bar)/count(r2) into :z0bar from t1; quit; 0.44 data _null_; z0 = probit(z0bar); zalpha = probit(alpha1); p1 = put(probnorm(2*z0 - zalpha)*100, 3.0); p2 = put(probnorm(2*z0 + zalpha)*100, 3.0); output; call symput('a1', p1); call symput('a2', p2); run; * creating confidence interval, bias-corrected method; proc univariate data = t1 alpha = .05; var r2; output out=pmethod mean = r2hat pctlpts=a1 a2 pctlpre = p pctlname = _lb _ub ; run; ... output omitted ... data t2; set pmethod; bias = r2hat - r2bar; r2 = r2bar; run; ods listing; proc print data = t2; var r2 bias p_lb p_ub; run; Obs r2 bias p_lb p_ub 1 0.5189 .0066164 0.40575 0.5936
标签: SAS经管大学堂：名校名师名课
京公网安备 11010802022788号
论坛法律顾问：王进律师
知识产权保护声明
免责及隐私声明
