Looking for some interpretation help with a GLIMMIX output (as most the documentation is from logistic models). The very basic bivariable model is as follows: proc glimmix data=have noclprint noitprint method=quad gradient ;   class study_id ;   model score=race / solution dist=mult link=clogit ;   random intercept / type=un subject=study_id ;       covtest 'Need Random Intercept?' 0 ;   where race ^in(.,88,99) ; run ; I have 5500 observations, with a total of 1795 individuals (study_id) observed over time. Participants have multiple scores (range: 0-4) over time. I have made no assumption about the distance between adjacent levels, but they are ordered (i.e. 0=not present, 4=very bad levels present). Race is ethnic race (i.e. caucasian, African-American etc) with 5 levels. I am modeling the probabilities of levels of score having lower ordered values in the response profile table. Once run, I have 4 intercepts  (0-3), and 5 estimates for race. Numbers below are just made up. Effect           Score     Race     Estimate Intercept      0                                  0.5 Intercept      1                                   0.6 Intercept      2                                   0.7 Intercept      3                                   0.8 Race                        Race1               0.09 Race                         Race 2              0.1 Race                          Race3               0.11 Race                          Race4               0.12 Race                          Race5               0.13 I would like to be able to interpret this (eventually) in a multivariable model as well, then present it in a logical and meaningful way. Any direction would be appreciated.     ... View more

I have a conundrum that I just can't figure out...(epidemiologist and not a bio/statistician) I have a continuous outcome with a range of 0-5.0. This data was generated from a questionnaire from a Likert-like item as a score (i.e. 0=no event, 1=event+severity level 1, 2=event+severity level 2, etc). The data is right-skewed. I want to model the data using <proc glimmix> with gamma log-link (where my distribution is exponential, r=1 -- such that my gamma density specializes to the exponential density). This assumes a constant coefficient of variation over the range of X, a random variable. From my understanding, GLMM with gamma should be able to model a gamma function with limits (0,infinity). So why does proc glimmix restrict to interval data? I can't model this data as a count process (i.e. Poisson, etc), so that is not an alternative option. I could use proc genmod, but intuitively, a GLMM is more appropriate than GEE models for my given research question. Or am I just modeling the data as an exponential with log-link since r=1? In which case, can I then model my 0's? Alternatively, I could add an interval to each observation (e.g. +1) to avoid the issue of zeros in the data, but I don't know how this would impact my beta estimates (if at all). Thanks! ... View more

Two code problems that I was hoping to get some further insight: (1) Trying my hand at arrays and I continue to get the an error code = "Error: Array subscript out of range at line X column Y". I am trying to merge (I) questionnaire and (ii) medication data (both longitudinal, sorted by participant study_id). There are a maximum of 8 interviewdates per participant, but tens of medications possible corresponding to a given interview date. (2) I also want to "carry forward" a variable (mednaive) but am unsure on how to do this. In this particular variable, a participant can be on a medication at one questionnaire (mednaive=0) and then off at another (mednaive=1). However, I want to capture "ever exposed" to the medication such that, if an individual is not on a medication at a given questionnaire but was in the past, then they will ALWAYS be mednaive=0 (even if they are mednaive=1 at this particular questionnaire). data qSummary ;   set dataset ;   by study_id ; /*sorted by study_id*/   retain qCount qdate1-qdate11 ; / *variables created for interview dates and number of completed questionnaires*/   mednaive=. ; /*new variable to identify those naive vs those experienced to a particular medication class*/      if ((drug_class='X') or (drug_class='Y') or (drug_class='Z')) then mednaive=0 ; /*on medication at current questionnaires*/      else mednaive=1 ; /*no on medication at current questionnaire*/      format mednaive mednaive. ; /*NEWVAR needed to address "carry forward" concept noted above -- unsure of how to do this*/   array qdate {*} qdate1-qdate11  ;   format qdate1-qdate11 date9. ;      if first.study_id then do ; /*if first questionnaire per person*/        qCount=1 ; /*count of questionnaire (max=11 questionnaires)*/        qdate1=interviewdt ; /*the first questionnaire date is equal to the first interviewdate variable already defined in the dataset*/        do i=2 to 11 ;           qdate{i}=. ;        end ;      end ;      else do ; /*if not first questionnaire per person*/           qcount=sum(qcount,1) ;           qdate{qcount}=interviewdt; /* <-- ERROR: ARRAY SUBSCRIPT OUT OF RANGE AT (THIS) LINE COLUMN X*/      end ;      if last.study_id then output ; run ; ... View more

I need to estimate the difference between interview dates (<interviewdt>), delete those interviews completed <300 days apart, and re-code questionnaire numbers to reflect the new dataset. This is longitudinal data. Variables: 1. study_id = unique id per patient 2. qnum = questionnaire number 3. interviewdt = date interview was conducted data datediff ;   input study_id qnum interviewdt ;   datalines ;   1 1 01Jan2007   1 2 04Jan2007   1 3 07July2008   2 1 15Feb2009   2 2 03Mar2009   2 3 30Mar2010   3 1 20Dec2012   3 2 15Feb2013 ; run ; Data should look like: Obs          study_id          qnum       interviewDt          1                    1                    1               01Jan2007 2                    1                    2               04Jan2007   3                    1                    3               07July2008 4                    2                    1              15Feb2009 5                    2                    2               03Mar2009       6                    2                    3                30Mar2010    7                    3                    1              20Dec2012           8                    3                    2              15Feb2013 --I need to understand how to bu program that would (1) Compute the difference between dates within an individual (i.e. from qnum=1 to qnum=2, and qnum=2 and qnum=3) etc (2) Delete observations who completes an interview <300d from the previous (i.e. should only complete one questionnaire approximately every 12 months) (3) renumber QNUM (e.g. in the above data, I would delete Obs 2 and Obs 8 -- so I want to create a new variable for Qnum reflecting the new interview number -- such that, for Obs2 where Qnum=2, this observation would be deleted and Obs3 where QNum=3 would become NewQnum=2). I am not familiar with <proc SQL> and would prefer to avoid (I know you all love SQL on the discussion boards!) ** ====================================================== ; This is what I tried but it didn't work : data dateDiff ;   set old ;   by study_id ;      diff_IntDt = dif(interviewDt) ; /*Compute difference between interviewDt*/      if not first.study_id then output ; run ; proc print data=dateDiff ;   var interviewYr qnum ; run ; PROBLEM: I ended up deleting ALL my first questionnaires (since I asked for <if not first.study_id then output>). If I don't include this condition, however, I don't get a dataset computing the difference between sequential <interviewdt>. The next step I would imagine is: data new ;   set dateDiff ;   by study_id qnum ;      if diff_intDt <300 then delete ; /*Delete any observation who's interview date occurs <300 days from the previous date*/   qCount = 0 ; /*Want to re-label QNUM to reflect the deleted observations*/      if qnum=1 then qCount=qnum+1 ;      label qCount = 'Number of Completed Questionnaires' ; run ; quit ; proc print data = dateDiff ;   var interviewYr qCount ; run ;  ... View more

Hi all -- I am a clinical researcher (not a trained analyst/statistician), so I'll do my best to be very clear about the problem I am currently having. I will use generic names for datasets and variables for illustrative purposes. I have 2 data sets (data1 and data2). Both are sorted on study_id. Data1 - Has multiple records per individual (i.e. study_id #1 has 6 records, representing 6 completed questionnaires -- therefore, "study_id" variable 1 is listed 6x, and "qnum" variable is 1-6). Period of observation: 01-01-07 to 31-12-13 Data2 - Also has multiple records per individual (i.e. study_id #1 has 72 records, representing 72 medications he/she has been prescribed). The "quest" variable is NOT included in Data2. Period of observation: 01-01-85 to 31-12-13. However, if a participant only completed a single questionnaire (Data1), he/she still has ALL previous prescriptions "linked" to that questionnaire (i.e. study_id=2 has only 1 questionnaire in 2010, but he/she has prescription records back to 1999 liked to qnum=1 when I attempted to merge the data below).  Goal: A merged data set that has one record per "questionnaire" (similar to Data1), and a new variable ("newvar") based on the prescribing of a drug reported in data2. To be more clear on "newvar": In Data1, there is a questionnaire interview date ("interviewdt") and in Data2 there is a prescription date ("medstartdate"). This "newvar" should be binary, such that IF a participant has started a particular prescription (represented by a particular code)(Data2) on or before the interview date (Data1), then "newvar"=1 (i.e. yes), ELSE, "newvar"=0 (i.e. no). Here is what I've  done : data Data3 ;      merge Data1 (keep=study_id qnum othervariables in=in_data1 where=(interviewdt^=.))           Data2 (in=in_data2 where (medstartdate^=. and (prescriptioncode="A" or prescriptioncode="B"))) ;      by study_id ;      if in_data1 and in_dadta2 ;      dtdiff=interviewdt - medstartdate ; /*ONLY want "interviewdt" where qnum=1 (i.e. first questionnaire/baseline) - I think this is part of my problem; how to only select qnum=1?*/      MedStartDtDiff = abs(dtdiff) ;           if dtdiff^=. then MedStartDtDiff=(-10220<=(interviewdt - medstartdate)<=10220) ; /*10220 represents the maximum number of days from the last day of observation, 31-12-13, to the first day, 01-01-85)*/      format MedStartDtDiff yesno. ; run ; proc sort data=Data3 (where =(MedStartDtDiff^=.)) ;      by study_id MedStartDtDiff ;           where qnum=1 ; run ; proc print=Data3 (obs=30) ; var study_id MedStartDtDiff qnum ; where qnum=1 ; run; SO this is what I get as an output --> study_id     MedStartDtDiff     MedStartDate     qnum      1               YES                    01NOV1996     1          --> This is correct; study_id1 has 6 questionnaires (with qnum=1 when interviewdt=14Feb2008 AND started med BEFORE interview date; therefore, MedStartDt is YES)      2               YES                    01JUN1999      1      2               YES                    01JUN1999      1      2               YES                    07NOV2006     1          --> This is NOT correct; study_id2 should only have a single entry. So this is my major issue! Why am I getting multiple study_id for each individuals? To further elaborate on study_id2 -- this participant only completed a single questionnaire (Data1), he/she still has ALL previous prescriptions "linked" to that questionnaire (i.e. study_id=2 has basline/only questionnaire in 2010, but he/she has prescription records back to 1999 liked to qnum=1). I really only want one line per study_id. What I'd like to see: study_id     MedStartDtDiff     MedStartDate     qnum      InterviewDt      1               YES                    01NOV1996     1              14FEB2008      2               YES                    01JUN1999      1              6JUN2010      3               YES                    15JUN1989      1              20AUG2008      4               NO                      07NOV2010     1              11NOV2011 etc etc.... I am happy to elaborate further or clarify if this isn't 100% clear. ... View more

Basic question for a new user: I have a class variable (Y) that has >2 levels (yes/no/don't know/refused). I would like to perform a 2-sample t-test to compare population means on a survey at baseline (Z = 1). However, I know that my class variable will return an error of too many levels. How do I exclude "don't know" and "refused" observations (thus, only leaving those participants who responded "yes" or "no" responses)? title X ; proc ttest data = X ;      class Y ;      var A ;           where Z = 1 ; run ; Thank you. ... View more

I have two variables; the second dependent upon the answer of the first. There are 5 variables (Q1) each with its own severity scale (Q2). Q1. Do you have the symptom? y/n/don't know/refused (1/2/88/99) --> If a participant answered Y(1) to Q1, they should go to Q2. Q2. Severity: 1/2/3/4/88/99 I would like to make a data set with no inappropriate missing data (where 'missing'=./don't know/refused). So far, I have been able to subset my data for Q1: data newdata;      set olddata;      if cmiss (of var1 var2 var3 var4 var5) then delete; run; This worked well and the expected number of observations were removed. However, I am unsure of how to move forward. If I ask SAS to remove the missing observations from Q2, then all those that answered N(2) to Q1 (and therefore, have appropriately missing data in Q2) would be removed. I only want to remove those individuals who answered Y(1) to Q1 and subsequently have missing data for Q2 (Q2 is conditioned on Q1). Any direction would be appreciated. ... View more

New, and slightly reluctant, SAS user (previously STATA user) SAS dates have me scratching my head a bit. I have a longitudinal dataset, but I am currently only interested in baseline (e.g. first questionnaire) data for which I have created a new dataset. Variables: (a) Date of Baseline Questionnaire: var1 = DDMMYYYY (SAS Informat: DDMMYYYY8) (b) Known Disease Exposure?: var2: yes(1), no(0), don'tknow(88), refused(99) --> ONLY those that responded YES had the next question asked: (c) Date of Disease Exposure: var3 =  MM (0-12, 0=unknown, 1=January etc.), var4 = YYYY Goal: Determine "Time Since Exposure" from baseline questionnaire What I'd like to do/think I should be doing: 1. Only analyse those individuals at baseline who responded "YES(1)" to var2 2. Assign/impute the middle of each month (e.g. 15th day) as a new variable to every individual (create: var5) 3. Create a new variable: 'DateExp' (var6) -- this will combine var3-5 in to a single variable represented as: DDMMYYY (equivalent to: var5-var3-var4) - I believe I need to use the MDY function; but the SAS examples are not easily understood (again, being a new user!) 4. Calculate time since exposure (TimeSinceExp=var2-var6/365.25) with an output in Years, rounded to a single decimal (e.g. 5.4 years) Any assistance would be appreciated. I am pondering this away today and will check in tomorrow to see how close (or far off!) my own code is....so far about 20% of my more complicated coding works (which I would say isn't too bad for 2nd week of use). Best & many thanks in advance. -Tyler ... View more