```Date: Fri, 20 Nov 2009 16:44:04 -0600 Reply-To: Joe Matise Sender: "SAS(r) Discussion" From: Joe Matise Subject: Re: Survey Question Data Analysis Comments: To: "Pashi, Arthur" In-Reply-To: Content-Type: text/plain; charset=ISO-8859-1 Sorry, hit send somehow by mistake, here is the complete email. --- So is 112 1, 12 or 11, 2? In general, you need to code multipunch survey data in one of two ways: 1. Spread variables; so, instead of QUESTION_1, QUESTION_1_01 QUESTION_1_02 QUESTION_1_03 etc. 2. Comma separated values, so, QUESTION_1 = 1,12,15 3. Several variables: QUESTION_1_1 QUESTION_1_2 where _1 is the first response, _2 is the second response if present, etc up to whatever limit is reasonable. 4. Vertical question design, ie, one row per response rather than one row per respondent. I strongly suggest option 1, because it's easier to do work with it. Option 4 is also a great option if the rest of your data is like that, but for non-programmers it can be very difficult to wrap their heads around. So, for instance, convert your QUESTION_1 to spread, through whatever algorithm makes sense based on your rules [I guess that it is always ascending, so 112 is 1,12 always?]. Set each value to 1 if present or 0 if not present but the question was answered, or missing if the question was not answered at all. Then keep a 1/0/missing flag for the question as a whole; and if you wish, keep a 1/0/missing flag for 'other' (for a coded answer). So: QUESTION_1 originally is '346' becomes QUESTION_1_ANSWERED = 1 QUESTION_1_01=0 QUESTION_1_02=0 QUESTION_1_03=1 QUESTION_1_04=1 QUESTION_1_05=0 QUESTION_1_06=1 etc. QUESTION_1_OTHER=0 [no 'coded' answer] Then you can easily identify what percentage of people had any answer to the question, what percentage had a particular answer or answers, and what percentage had a 'coded' response. By using QUESTION_1_... you can use easily formulated arrays to do calculations; you may want to have a single character difference in the proper numerical responses as opposed to the flag responses - so, question_1f_answered, question_1f_other or something like that - to allow you to use question_1_: to define all of the codings. That however depends on your data usage needs. -Joe On Fri, Nov 20, 2009 at 4:39 PM, Joe Matise wrote: > > > On Fri, Nov 20, 2009 at 3:18 PM, Pashi, Arthur wrote: > >> I have the following survey question: >> >> >> >> >> What types of health providers have you ever seen or talked to about >> your illness? (Check all that apply) >> >> >> [QUESTION_1] >> >> 1 Medical Doctor >> >> 2 Chiropractor >> >> 3 Dentist >> >> 4 Priest/Pastor >> >> 5 Dietician >> >> 6 Nurse/Nurse Practitioner >> >> 7 Pharmacist >> >> 8 Physical Therapist >> >> 9 Psychiatrist or Psychologist >> >> 10 Other (Specify) >> >> 11 None/Haven't seen or talked to anyone [Go to next section] >> >> >> 77 Don't know / Not sure >> >> 99 Refused >> >> >> >> After looking at the contents of response #10 Other-specify, I found >> that the respondents provided the following responses which I have >> grouped according to the choices provided by Question_1. I have put >> those choices under Question_1other (variable derived from response >> #10). Question_1other has a much longer list of responses than >> Question_1 (the primary variable). >> >> >> >> QUESTION_1other >> >> 1 Medical Doctor >> >> 2 Chiropractor >> >> 3 Dentist >> >> 4 Priest/Pastor >> >> 5 Dietician >> >> 6 Nurse/Nurse Practitioner >> >> 7 Pharmacist >> >> 8 Physical Therapist >> >> 9 Psychiatrist or Psychologist >> >> 10 Other (Specify) >> >> 11 None/Haven't seen or talked to anyone [Go to next section] >> >> >> 12 Specialist >> >> 13 Clinic, ER, Hospital (or VA), Public Health, Rehab Center or Urgent >> Care >> >> 14 Pain Center or Clinic >> >> 15 Procedure - CAT Scan, MRI, Colonoscopy, Diagnostic Imaging >> >> 16 Pain management, Unspecified >> >> 17 Homeopathic >> >> 77 Don't know / Not sure >> >> 99 Refused >> >> Some of the records have multiple responses such as 110, or 710 with 10 >> being Other (specialist, medical doctor, nurse, etc). >> >> I have been asked to input any of the responses 1 to 11 of >> QUESTION_1other into Question_1, and leave responses 11 to 17 in >> Question_1other. >> >> My questions: 1) Is this a good approach for treating the main question >> to determine the frequencies for each type of provider? 2) If that is >> the case, should I use Arrays to sum similar responses? >> >> >> >> Thank you, >> >> >> >> Arthur K. Pashi >> >> >> > > ```

Back to: Top of message | Previous page | Main SAS-L page