Try using DEQUOTE.   CALL SYMPUTX('PRODUIT', DEQUOTE(X1) ); The rest of your code is a real hodgepodge and indicates maybe you need to learn a bit more before jumping into macro.     For example: This is the WRONG WAY to get the number of observations in a data set   DATA _null_; SET test END=eof; IF eof THEN DO; CALL SYMPUT('nb',put(_N_,8.)); END; run;     Without knowing the planned use of the macro I speculate this is too much overwriting of the target macro symbol PRODUIT %do i=1 %to &nb.; data _NULL_; set test(obs=&i); * CALL SYMPUTX('PRODUIT',X1); CALL SYMPUTX('PRODUIT',DEQUOTE(X1)); %* Regardless, this is where DEQUOTE might go; run; %put "&PRODUIT."; %end; ... View more

LENGTH is a character function. You are asking for the character length of a numeric variable. A read of the LENGTH documentation will tell you:   If string is a numeric constant, variable, or expression (either initialized or uninitialized), SAS automatically converts the numeric value to a right-justified character string by using the BEST12. format. In this case, LENGTH returns a value of 12 and writes a note in the SAS log stating that the numeric values have been converted to character values. Important take away.... read the log. ... View more

Consider sample data with a little more JAN2020, AND an extra variable HALF, between START and END.  These extras will demonstrate some of the problems you will encounter.    A common first try would be TABULATE, however the percentages you want are from n per variable (and n is not representing mutually exclusive conditions), so the COLPCTN concept won't apply -- neither data as is or data in transpose.   First look at SUMMARY counts and examine how the percentages could be calculated from that. DATA have; input ID START HALF END country $ time monyy7.; format id start half end 4.; attrib time format=monyy7. label='Month'; datalines; 1 3 5 6 FR FEB2020 2 5 7 8 US FEB2020 3 3 . . FR FEB2020 4 . . . FR JAN2020 5 . . . US JAN2020 6 4 6 . FR JAN2020 ; proc summary noprint data=have; class time country; var start half end id; output out=counts(index=(order=(time country))) n=/autoname; run; Summary output Arrows showing numerators and common denominator that would be used to compute each original variables fraction (aka percentage) within SUMMARY row.  I.e. START_N / ID_N is a fractional value [0..1] that can be displayed with format PERCENTn.d  For your desired output PRINT or REPORT procedures would require two groups of the summary variables (n and computed %) to be pivoted in parallel row-wise.   Transforming DATA step The following DATA step (in spoiler) computes values and pivots rows.  The column pivoted into, such as PERCENT, is also formatted here.  An additional variable , N_PCT, is calculated as "nnn ( ##.#% )" and is for use in a slightly alternate report structure.   data counts2; set counts(where=(_type_=3)) counts(where=(_type_=2)) /* row wise want time totals after time>country totals, this is setup for a later ORDER=DATA */ ; array counts ID_N START_N HALF_N END_N; array percents ID_PCT START_PCT HALF_PCT END_PCT; do _n_ = 1 to dim(percents); percents(_n_) = counts(_n_) / id_n; end; call missing(ID_PCT); if _type_ = 2 then country = 'Total'; length _NAME_ $32; * loop for array based transpose (i.e. pivot) of summary data and computed values; do _n_ = 1 to dim(percents); _NAME_ = scan(vname(percents(_n_)),1,'_'); COUNT = counts (_n_); PERCENT = percents(_n_); ************************************** ; * compute values "<n> ( <pct> )" for also reporting in a single columns under country; ************************************** ; length _NAME2_ $32 N_PCT $25; if _NAME_ ne 'ID' then _NAME2_ = _NAME_; else _NAME2_ = 'ID (n)'; if not missing(PERCENT) then N_PCT = vvalue(COUNT) || ' ( ' || vvalue(PERCENT) || ' )'; else N_PCT = vvalue(COUNT); **************************************; output; end; keep time country _name_ _name2_ COUNT PERCENT N_PCT; format COUNT comma9.; format PERCENT id_pct start_pct half_pct end_pct percent8.1; run; Summary data pivoted with additional computed values The data is now in a categorical layout.  The time, country and original variable names can be used in TABULATE (CLASS) or REPORT (stacked /ACROSS or /GROUP).   REPORT Two columns under each country, one for N, one for %.  Pretty standard reporting.  The header stacking (aka column dimension hierarchy) is specified with time, country, (COUNT PERCENT)   options missing = ' '; proc report data=counts2; title 'Counts and computed %''s REPORT, time&country across'; columns _name_ time, country, (COUNT PERCENT); define time / ' ' across order=internal; define country / ' ' across order=data style=[cellwidth=5em textalign=center] ; define _name_ / ' ' group order=data; define COUNT / 'n' format=comma9.; define PERCENT / '%' ; run;   Output   REPORT #2 One column under each country, nnn ( ##.#% ). This report code is more complicated and uses some tricks. dummy column Z / noprint custom row format for ID row /* Trick: * dummy column, Z, needed to display character variables under across variables * N_PCT is precomputed character value */ proc report data=counts2; title 'Counts and computed %''s REPORT, time&country across'; columns _name2_ time,country,(N_PCT) Z; define time / ' ' across order=internal; define country / ' ' across order=data style=[cellwidth=4em textalign=center] ; define _name2_ / ' ' group order=data; define N_PCT / 'n (%)' ; define Z / computed noprint; /* trick */ compute _name2_; if _name2_ =: 'ID' then call define (_ROW_, 'STYLE', 'style=[textalign=center]'); endcomp; run; Output       ... View more

You create a hash instance that memory leaks (albeit very small leak) declare hash h(); * create uninitialized instance whilst adding to pdv; In HoH the hash data element h  only needs to be declared.  Of course, the _new_ instance must have occurred prior to  .add() or .replace()  time. So, simply declare the host reference, h, that will interact with your anonymous hashes. declare hash h; * add h, a hash reference, to PDV; The  table  variable should be $41 to accommodate longest plain <libref>.<dataset>, that being $8 + $1 + $32.  If the table value is to further accommodate data set options you might want the length to be more like $200.  However, since the  where  clause is part of the key, you should exhibit discipline and assign only plain data set names to  table  and use the separate  where , say $200, for loading subsets.   As stated elsewhere, the 'master' hash  hoh  does not need to be multidata in your situation.  Also, you do not need to repeatedly code  table="<whatever>";  The value will not be altered by a prior  hoh.add()   Example data have; do id=1 to 50 by 1; value=19; output; end; run; data _null_; if 0 then set have; length table $41; length where $200; if _n_=1 then do; declare hash hoh(); hoh.defineKey("table", "where"); hoh.defineData("h", "table", "where"); hoh.defineDone(); declare hiter i_hoh("hoh"); declare hash h; table="have"; where=" "; h = _new_ hash(dataset: catx(" ", table, where), multidata: "no"); h.defineKey("id"); h.defineData(all: "yes"); h.defineDone(); hoh.add(); where="(where=(id<20))"; h = _new_ hash(dataset: catx(" ", table, where), multidata: "no"); h.defineKey("id"); h.defineData(all: "yes"); h.defineDone(); hoh.add(); where="(where=(id>40))"; h = _new_ hash(dataset: catx(" ", table, where), multidata: "no"); h.defineKey("id"); h.defineData(all: "yes"); h.defineDone(); hoh.add(); end; do while (i_hoh.next() = 0); rows = h.num_items; put (table where rows) (=); end; stop; run; Log NOTE: There were 50 observations read from the data set WORK.HAVE. NOTE: There were 19 observations read from the data set WORK.HAVE. WHERE id<20; NOTE: There were 10 observations read from the data set WORK.HAVE. WHERE id>40; table=have where=(where=(id<20)) rows=19 table=have where=(where=(id>40)) rows=10 table=have where= rows=50   ... View more