关于多个变量proc freq进行卡方检验的结果整理

shiyiming

在用proc freq进行卡放检验时，想把每个变量对应的p值的结果显示在同一个数据集里，发现output语句只能显示最后一个分析变量！于是想，是否可以一个变量一个output文件，然后再set呢，却发现少了一列变量名。如果在分析每个变量时，output的数据集里能够添加上文件名就好了，我的文件名是用所分析的变量命名的。但是却不知道如何能够为每个output的数据集添加一个变量名？
   这样的做法可能不太好，但是能够解决问题。而且难免会有的数据集为空。不知道还有没有更好的方法？

shiyiming

my 2 cents,
(1) you may use a dataset label to identify the table generated by the OUTPUT statement. if you want a new variable, you have to hard code it in a separate step;
(2) the OUTPUT statement may not be able to generate a dataset if some categories of your data are missing. if you add some zero-weighted observations for the missing categories and use the WEIGHT statement with ZEROS option, then there will always be a dataset.

As a computer slave, I do this a lot. The following is an example. It did slightly more than your needs because I transposed the frequency table before merging it horizontally with the p-value table. You sure can make the whole process a lot easier!
[code:3baztzel]
%macro draw_two_(        dsn=,
                        outset=,
                        xvars=,
                        yvar=,
                        xcats=%str(0 1),
                        ycats=%str(0 1),
                        stat=%str(CHISQ|CHISQ|_PCHI_ P_PCHI)
);
%local i j k var;
%local        N_xvar; %*# of X variables;
%local        N_xcat; %*# of X categories;
%local        N_ycat; %*# of Y categories;
%local         stat_request
                stat_out
                stat_vars;
%let i=1;
%let var = %scan(&xvars, &i, %str( ));
%do %while(&var ne %str());
        %local xvar&i;
        %let xvar&i = %upcase(&var);
        %let i=%eval(&i+1);
        %let var = %scan(&xvars, &i, %str( ));
%end;
%let N_xvar=%eval(&i-1);

%let i=1;
%let var = %scan(&xcats, &i, %str( ));
%do %while(&var ne %str());
        %local xcat&i;
        %let xcat&i = &var;
        %let i=%eval(&i+1);
        %let var = %scan(&xcats, &i, %str( ));
%end;
%let N_xcat=%eval(&i-1);

%let i=1;
%let var = %scan(&ycats, &i, %str( ));
%do %while(&var ne %str());
        %local ycat&i;
        %let ycat&i = &var;
        %let i=%eval(&i+1);
        %let var = %scan(&ycats, &i, %str( ));
%end;
%let N_ycat=%eval(&i-1);

%let stat_request = %scan(&stat, 1, %str(|));
%let stat_out = %scan(&stat, 2, %str(|));
%let stat_vars = %scan(&stat, 3, %str(|));

/*
generate 6 names for temporary datasets;
*/
%do i=0 %to 5;
data;stop;run;
%let tmp&i = %sysfunc(tranwrd(&syslast, WORK., ));
%end;

PROC CONTENTS DATA=&dsn OUT=&tmp1(KEEP=name type) NOPRINT;
PROC SQL NOPRINT;
        SELECT type INTO :ytype
        FROM &tmp1
        WHERE upcase(name) = upcase("&yvar");       
        SELECT type INTO :xtype
        FROM &tmp1
        WHERE upcase(name) = upcase("&xvar1");
QUIT;

%*padding all combinations of &&xvar&i * &yvar w/ weight 0 to the input dataset;
DATA &tmp5;
        SET &dsn(KEEP=&xvars &yvar) END=eof;
        LENGTH wt 3;
        RETAIN wt 1;
        OUTPUT;
        IF eof THEN DO;
                wt = 0;
               
        %do i=1 %to &N_xcat;
                %do k=1 %to &N_xvar;
                        %if &xtype=1 %then %str(&&xvar&k = &&xcat&i;); %else %str(&&xvar&k = "&&xcat&i";);
                %end;        %*end of k=1 to # of x variables;
                %do j=1 %to &N_ycat;
                        %if &ytype=1 %then %str(&yvar = &&ycat&j;); %else %str(&yvar = "&&ycat&j");
                        OUTPUT;
                %end;        %*end of j=1 to # of y categories;
        %end;        %*end of i=1 to # of x categories;
        END;
RUN;

PROC DATASETS NOLIST;
        DELETE &tmp0 &tmp1 &tmp2 &tmp3 &tmp4;
RUN;
%do i = 1 %to &N_xvar;
PROC FREQ DATA=&tmp5 NOPRINT;
        TABLES &&xvar&i*&yvar / OUT=&tmp1 OUTPCT &stat_request;
        %if %length(&stat_out) %then OUTPUT OUT=&tmp2(KEEP=&stat_vars) &stat_out;;
        WEIGHT wt/ZEROS;
RUN;

%*transpose the frequency table;
DATA &tmp3(DROP=&yvar count percent pct_row pct_col RENAME=(&&xvar&i=xcat));
        LENGTH varname $ 32;
        RETAIN varname "&&xvar&i";
        DO UNTIL(LAST.&&xvar&i);
                SET &tmp1;
                BY &&xvar&i;
                SELECT (&yvar);
                %do j=1 %to &N_ycat;
                        %if &ytype=1 %then %str(WHEN (&&ycat&j)); %else %str(WHEN ("&&ycat&j"));
                        DO;
                                "count_&&ycat&j"n = count;
                                "percent_&&ycat&j"n = percent/100;
                                "pct_row_&&ycat&j"n = pct_row/100;
                                "pct_col_&&ycat&j"n = pct_col/100;
                        END;
                %end;
                        OTHERWISE;
                END;
        END;
RUN;

%*merge frequency table with statistics table horizontally;
DATA &tmp4;
        %do j=1 %to &N_xcat;
        MERGE &tmp3(FIRSTOBS=&j OBS=&j) &tmp2;
        OUTPUT;
        %end;
RUN;

PROC APPEND BASE=&tmp0 DATA=&tmp4;
RUN;
PROC DATASETS NOLIST;
        DELETE &tmp1 &tmp2 &tmp3 &tmp4;
RUN;
%end;
DATA &outset;
        SET &tmp0;
RUN;
PROC DATASETS NOLIST;
        DELETE &tmp0 &tmp5;
RUN;
%mend draw_two_;

data test;
        array vars[*] var1-var9;
        array p[9] _temporary_ (.1, .2, .3, .4, .5, .6, .7, .8, .9);
        do i=1 to 100;
                do j=1 to dim(vars);
                        vars[j] = ranbin(0,1,p[j]);
                end;
                y = ranbin(0, 2, .5);
                output;
        end;
        drop i j;
run;
%draw_two_(dsn=test, outset=dummy, xvars=var1 var2 var3 var4 var5 var6 var7 var8 var9,
                        yvar = y, ycats=0 1 2)
PROC FORMAT;
        VALUE pvaluecolor
        0.01-<0.05 = 'brown'
        0-<0.01 = 'red'
        0.05-high = 'white'
        other = 'gray'
        ;
PROC REPORT DATA=dummy NOWD  Split=' '
        style(column)={background=white font_face='helvetica'}
        style(header)={background=khaki font_face=times};
        COLUMN varname ('y=0' count_0 pct_col_0)
                                        ('y=1' count_1 pct_col_1)
                                        ('y=2' count_2 pct_col_2)
                        p_pchi;

        DEFINE varname/display width=8;
        DEFINE count_0 / 'n' display format=comma5.;
        DEFINE pct_col_0 / '%' display format=percent8.2;
        DEFINE count_1 / 'n' display format=comma5.;
        DEFINE pct_col_1 / '%' display format=percent8.2;
        DEFINE count_2 / 'n' display format=comma5.;
        DEFINE pct_col_2 / '%' display format=percent8.2;
        DEFINE p_pchi / display 'p-value' width=7 format=pvalue. style=[background=pvaluecolor.];
        WHERE xcat = 1;
RUN;
[/code:3baztzel]

shiyiming

多谢urchin！
继续探索中！

shiyiming

这是我最后成功实现的结果：基本思路是先用第一个变量生成一个result的表，这个表也是经过转置了的表。然后用宏来实现其他变量的分析结果，在进行其他变量分析时，直接和前面的结果合并，然后被下一个变量的结果替换掉，感觉似乎会少占一些空间。有一点点麻烦，请各位多指教，继续改进！
proc freq data=ana noprint;/**把第一个文件先生成作为一个格式文件**/
     exact fisher;
     tables group*r1/chisq sparse;
     output out=result chisq fisher;
run;
data result;
   set result;
   keep P_PCHI XP2_FISH; run;
proc transpose data=result out=result;   run;
data result; set result;rename col1=r1;drop _LABEL_;run;
%macro tihuan(var);
proc freq data=ana noprint;
     exact fisher;
     tables affected*&var/chisq sparse;
     output out=d chisq fisher;
run;
data d;   set d;   keep P_PCHI XP2_FISH; run;
proc transpose data=d out=d;   run;
data d; set d;rename col1=&var;drop _LABEL_;run;
data result;merge result d;by _NAME_;run;
%mend tihuan;   
%tihuan(r2);        %tihuan(r3);        %tihuan(r4);        %tihuan(r5);       
%tihuan(r6);                /**输入多个var*/
proc transpose data=result out=result; id _NAME_;var r1--r?;  run;
data result; set result;rename _NAME_=name;run;
proc export data=result
    outfile="F:\result.txt" replace;
run;