The first thing that I would look is to go back to the software that collected the data and see about options on how the data is exported. Several data collection packages I have used have options on exporting data such as this. You may find that there is an option to have that questions values exported as 14 dichotomous (0/1 coded where 1 indicates selected) variables, or 14 different variables that record the order responses were "checked".   From what I see if you have a value like   1211 you do not know if you have two responses of 12 and 11 or 3 of 1, 2 and 11. I base this statement on the shown value of 1110. Since you apparently do not have possible single values of 110 then I have to parse that value as 11 and 10. Which means that the responses are not in value order but selection order. So you do not know which value 1211 has. You might have had other information in the raw file such that some values had leading 0, so 01, 02 instead of 1 2 which might make your shown value of 102 (which appears to be 10 and 2) originally a text value of 0102, which would allow parsing (having done so with a similar field with 45 categories). So examine your source data and if the values has leading zeroes for the 01 type values then re-read the data as text so you have the leading 0. Then you can parse the value from left to right two characters at a time into single variables. You would also be able to parse "lettuce" as   lettuce = (index(vegQ02,'01')>0 and mod(index(vegQ02,'01'),2)=1); which would have a 1 when the string '01' occurs in an "odd" starting spot, position 1,3, 5 etc. To avoid complications with values like 1012 data example; input vegq02 $; lettuce = (index(vegQ02,'01')>0 and mod(index(vegQ02,'01'),2)=1); datalines; 01 0201 040201 1011 ; You really have to explain what your second line of IF code was supposed to do. BTW it is a much better code scheme to use 1/0 for yes/no true/false etc. then 1/2. With a 1/0 scheme the Sum is the count of Yes/True/Present or what have you. The Mean is a percentage of Yes values. If you look at multiple variables with the 1/0 coding scheme such as Sum(lettuce,tomato,carrot) then you get the number of response marked Yes. Range=0 can tell if they were all the same choice, max=1 would indicate at least one was chosen, min=0 at least one choice not made. With a 1/2 type coding then you pretty much have to test each value and accumulate test responses.   Another approach would be to parse the values with an array instead of a bunch of If/then/else: data example; input vegq02 $; array r(14) (14*0); do i= 1 to (length(vegq02)/2) by 2; r[input(substr(vegq02,i,2),f2.)]=1; end; drop i; datalines; 01 0201 040201 1011 ; You could assign names like "lettuce" "tomatoe" "carrot" etc to the array elements if you want. Or just assign labels to r1 to r14. ... View more

Why do you think some values are missed?    Do you have more than 728+338 observations? Are you sure they weren't eliminated by something else in whatever code you used to generate the output pictured?   Or do you want to count some observations as both 0 and 1?  So that total of the number of 1's and number of 0's is more than the total number of observations? ... View more

@jag07g: One problem is that your in/output values aren't encoded as Booleans, which would be 1 = use and 0 = no use. That would stand to reason seeing that in SAS 1=true and 0=false. This way, the logical expression (sig and alc) would yield true in the case of both sig use and alc use and false otherwise. This is why in order to get what you require you need the IF-THEN-ELSE bunch or, more concisely, the IFN function, such as the expression: cig_and_alc = ifN (cig=1 and alc=1, 1, 2) ; Booleans can still be used; however, due to the odd way the values are encoded, some funny arithmetic has to be added: data _null_ ; input cig alc ; cig_alc = 2 - ((2 - cig) and (2 - alc)) ; put cig alc cig_alc= ; cards ; 1 1 1 2 2 1 2 2 ; run ; Which yields: 1 1 cig_alc=1 1 2 cig_alc=2 2 1 cig_alc=2 2 2 cig_alc=2 If the values were encoded as normal Booleans, the expression for cig_alc would be much simpler: data _null_ ; input cig alc ; cig_alc = cig and alc ; put cig alc cig_alc= ; cards ; 1 1 1 0 0 1 0 0 ; run ; and the result would be much more, uh, logical: 1 1 cig_alc=1 1 0 cig_alc=0 0 1 cig_alc=0 0 0 cig_alc=0 Kind regards Paul D.           ... View more

Yes the usual SAS modeling syntax of A*B on the MODEL statement also works in SURVEYLOGISTIC ... View more

@jag07g wrote: Thanks all for the helpful suggestions.    Here is what I have worked out so far using the surveyfreq command in SAS. This seems to give me what I want in terms of CIs and row percentages.    proc surveyfreq data=2017data; weight dataweight; table (gender age youth race grade)*(drink drug [other vars])/ cl row nocellpct nostd nofreq nowt; ods output CrossTabs=ResponseTable1; run; Would I be able to get the resulting cross tab data into Access so that I can further modify the names and labels of the columns?    My next step I believe would be to continue developing the macro so that I would not have to list every outcome variable in the table statement as I have over 50 outcome variables? Thanks for all your help! It is much easier to edit "names and labels" in SAS than in Access from my experience. Proc datasets is designed to do that without opening any existing data.     Again, though the output set gets a tad large you should be able to have all of your response variables in a single tables call. I might in some circumstances use a separate call to surveyfreq to create separate output if the responses are "grouped" in some natural way such as if I have responses only related to male or female respondents such as gender specific medical tests and remove gender from the variables. But a  macro to loop through response variables is likely not needed.   Another approach may be to use SURVEYMEANS with domain analysis if response variables are 0/1 coded.     ... View more