I don't know proc sql, so maybe I will just do the way I described.    Thanks for confirming that there isn't one code for inverse transformation!   ... View more

I think I meant to ask this:   I know we obtain inverse transformed values by using following formula.   new variable = 1/(K − old variable) (for negatively skewed data)                      or 1/(C + old variable) (for positively skewed data)   K = largest score + 1 If zero is in the range use this equation, where C = smallest score + 1.   I wondered if there is SAS code that I can simply use instead of figuring out the largest or smallest values and using the formula.   But thanks for your information though. ... View more

But apparently each level of TRIAL1 represents a level of another factor ("task"), complicated by the fact that the response of a subject to the different tasks may be autocorrelated and that the degree of autocorrelation may vary by subject.    --> My reply: Actually, each task is repeated ten times over ten trials, but these tasks were randomly mixed in each experimental block. So I am not sure if the trial1 really represents tasks. It may not.   After consulting with one faculty and students majoring in biostats, I decided to ignore the unequally spaced intervals among the trials because it is unlikely that variation in terms of the time each trial occurred caused by each individual would change how they perform each task.    I will determine the best covariance structure by examining the fit statistics.   And thanks for confirming that I can use info under Den DF as my second degrees of freedom.           ... View more

Thanks for your advice. After changing the order of variables and sort the R dataset, the normality test ran well.   Thanks.   Regarding the covariance structure, there is not specific interval and the time intervals among trials vary and depend on each individual. This is because each individual had the capability to move on to the next trial when they completed responding to the current task in each trial. In this type of unequally time spaced dataset, do you have any recommendation in terms of what code should be used to specify covariance structure? Can I just use "UN" instead of "AR(1), ARH(1) for heteroscedastic data, or SP(pow) (c-list)"?   Or can I use VC because following description about VC (http://documentation.sas.com/?docsetId=statug&docsetTarget=statug_mixed_syntax14.htm&docsetVersion=14.3&locale=en#statug.mixed.repeatedstmt_type) says that VC specifies standard variance components and is the default structure for both the RANDOM and REPEATED statements. In the RANDOM statement, a distinct variance component is assigned to each effect. In the REPEATED statement, this structure is usually used only with the GROUP= option to specify a heterogeneous variance model. I like the fact that VC with "Group=" option handles heterogeneous variance model, which is the case in my result.     SP(pow) (c-list) was something I found in  Littell et al.'s(2006) book.   If I ask one more question, I want to report the F statistics in my document. I wonder if I can assume the second degrees of freedom I need to report is written under the heading of Den DF in the second column of the SAS output. So if I write, "a significant group by SLC interaction (F(2,359)=18.14, p<.0001), the main effect of group (F(1,46.2)=48.96, p<.0001), and the main effect of SLC (F(2,359)=148.47, p<.0001) were observed." will this be a correct statement?   Thanks again in advance.   ex) SAS Output: ... View more

I still wanted to test the normality and homoscedasticity assumptions with codes and I faced an error message like this:   For normality assumption testing, the code I used is this:   proc mixed data=dissertation1 covtest method=ml;*plots=all; class group subjectid slc trial1; model gmp_syl_log = group|slc|trial1/ ddfm=kr2 outp=R; random intercept / subject=subjectid(group); repeated trial1 / subject=slc*subjectid(group) type=sp(pow)(trial1); /*lsmeans trial1 / diff adjust=simulate(seed=98375); lsmeans group*slc ;*/ run; ods graphics on; proc univariate data=R normal; BY group slc; var Resid; histogram / normal ctext = blue; run;   The error I received is this: ERROR: Data set WORK.R is not sorted in ascending sequence. The current BY group has SLC = 3 and the next BY group has SLC = 1.   Could you please help me one more time? In addition, I changed the covariance structure to sp(pow)(trial1) because the trials had unequally spaced time intervals. There were different numbers of fillers. Thanks in advance.   Linda ... View more

Could you please explain me what this one line means? There is not enough explanation in the proc mixed manual except that it says, "ADJUST=SIMULATE< (sim-options) >".   lsmeans trial1 / diff adjust=simulate(seed=98375);   Thanks in advance! ... View more

Thank you! Yes, I will continue to study about SAS.   Thanks so much!   ... View more

Thanks a lot for your help. I am surprised by the level of help I can get from this SAS community. Thank you!   I have a few questions regarding the codes you provided.   1. I understood that you chose AR(1) after transforming the data by log. However, I don't see any code to test normality/homoscedasticity assumption except you included codes to observe the distribution of the data. I wonder how you test (or code) it. Or do you think I can use the same codes I used to test these two assumptions?   2. I see many places where it says "98375." Where is this number coming from?   3. In the second to the last commands where it starts with "proc mixed" I wonder if I can include "group*slc" next to your code in the model statement to check interaction effect. Or is the vertical line that divides independent variables already include meaning of any possible interactions among these variables?   4. In this proc mixed command, I do not see the empirical statement. As far as I know, if errors in your data is not normally distributed, you have to use "empirical" statement to accommodate it. Do I not need it because it is normally distributed after the log tranformation??   After writing above question, I read your explanation about why my data does not meet the requirement for empirical statement. You cited that "this approach assumes that the number of subjects per treatment is substantially greater than the number of times of observation. When the number of observation times is equal to or greater than the number of subjects per treatment, as often happens in agricultural experiments, the empirical estimate of Var(K'b) may actually be less than the model-based estimate and the resulting test-statistics may be wildly inflated."   However, now I am curious isn't my data number of subject (24 for each group) exceeds the number of observation (10 for each condition)? Or is the degree too small? Or is this citation talking about the size of number of subject (48 altogether) and the number of observation at the whole study level (30*48=1440)?      When you mentioned to use "ddfm=kr2", does it mean to replace empirical statement in proc mixed?   When you said, "you need to specify the existence of 48 subjects rather than 24, which you can do by specifying subject=subjectid(group) rather than subject=subjectid.", do you mean when I try to replace "empirical statement" to "ddfm=kr2"?   If the response mean could theoretically change with levels of Trial1, then Trial1 needs to be included as a fixed effect in the MODEL statement.   --> Thanks for your advice! I see that you included them after model in both proc mixed and proc glimmix.   5. I see that glimmix provides really neat results. When I try to justify the reason for going with glimmix, I wonder if I can describe my study following the flow chart of CH1 in Littell, et al. (2006). How do you view my design?   Response: Non-normal     Errors: Correlated   Random Effects: Yes   Non-linearity: None   Tool in SAS/STAT: Glimmix   6. Regarding the levels of the design: I view a little differently from yours, and I wonder if it will change any of the SAS codes. Subject (was not assigned randomly to Group, native English vs. native Mandarin speakers), Session within Subject (assigned--not randomly--to SLC), and RepeatedMeasure nested within Session within Subject (assigned--not randomly--to Trial1).   Could you please help me to answer above questions further? Thanks so much again! ... View more

Thanks for your information. I am reading the book, but if anyone can help me that will be great, too! Somehow my reply yesterday was not saved so I am writing again. Yes, although the order of stimuli presentation was randomized, the order was pre-determined and every participant underwent the same tasks in the same order.    Also, I am attaching my data file with the codes. The additional question I have is whether I need to stick to proc mixed with empirical statement because proc glimmix does not provide complicated covariance structure when homoscedasticity assumption does not meet. Thanks for your further advice in advance.   libname a1 'C:\Data Analysis\Experimental Results\SAS'; proc import datafile="C:\Data Analysis\Experimental Results\SAS\dissertation.csv" out=a1.Dissertation dbms=csv replace; run; data a1.dissertation1; set a1.dissertation; where Whole_1st=1; run; proc mixed data=a1.dissertation1 ic; class SubjectID SLC Group Trial1; model GMP_syl= Trial1/solution; random intercept Trial1/sub=SubjectID; run; proc sort data=a1.dissertation1 out=a1.dissertation2; by SLC Group; run; proc mixed data=a1.dissertation2 ic; class SubjectID SLC Group Trial1; model GMP_syl= Trial1 /solution outp=R; random intercept Trial1/sub=SubjectID; run; data R1; set R; absr=ABS(RESID); run; proc glm data=R1; class SLC Group; model ABSR=; run; proc mixed data=a1.dissertation2 covtest method=ml ; class SLC Group SubjectID Trial1; model GMP_syl=SLC Group SLC*Group/solution outp=R; random intercept Trial1/subject=SubjectID type=ARH(1); run; proc mixed data=a1.dissertation2 covtest method=ml ; class SLC Group SubjectID Trial1; model GMP_syl=SLC Group SLC*Group/solution outp=R; random intercept Trial1/subject=SubjectID type=CSH; run; ods graphics on; proc univariate data=R normal; BY SLC Group; var Resid; histogram / normal ctext = blue; run; proc mixed data=a1.dissertation2 empirical covtest ic; class SLC Group SubjectID Trial1; model GMP_syl=SLC Group SLC*Group/solution ddfm=bw outp=R; random intercept Trial1/type=ARH(1) subject=SubjectID g; run; proc mixed data=a1.dissertation2 empirical covtest ic; class SLC Group SubjectID Trial1; model GMP_syl=SLC Group SLC*Group/solution residual; random intercept/subject=SubjectID solution; repeated Trial1/subject=SubjectID type=ARH(1); lsmeans SLC /diff slice=SLC; contrast 'SLC1 vs SLC2' SLC 1 -1 0; contrast 'SLC1 vs SLC3' SLC 1 0 -1; contrast 'SLC2 vs SLC3' SLC 0 1 -1; run; proc glimmix data=a1.dissertation2; class SLC Group SubjectID Trial1; model GMP_syl=SLC Group SLC*Group/dist=gamma; random intercept /subject=SubjectID; run;   ... View more

Thanks for your advice. I have 24 subjects in each group and there are two groups. So there are 48 participants altogether. The SLC has three levels of experimental conditions each participant participated. Each participants attempt each task ten times in each SLC condition. These ten trials were indicated by the variable name "Trial1". I had to go with ARH (1) because homoscedasticity assumption did not meet. Does this explanation help you help me better? Also, I wonder if I can email you my dataset and my codes from importing dataset to testing assumptions. Thanks in advance for your help.     ... View more

It did actually import pretty well. I could not copy paste the exact screen I saw in SAS. Here I am attaching the screen shot again.   Please let me know if you still think the data was not imported correctly.   Thanks. ... View more

Hi. Thanks for your advice. I changed my code to this and there was no error message, but still the result did not print. Could you please advise me one more time? Also, I don't know how to upload the SAS datset, this webpage seems not uploading the file.   Thanks in advance.   proc mixed data=a1.dissertation2 empirical covtest ic; class SLC Group SubjectID Trial1; model GMP_syl=SLC Group SLC*Group/solution residual; random intercept/subject=SubjectID; repeated Trial1/subject=SubjectID type=ARH(1); lsmeans SLC /diff slice=SLC; contrast 'SLC1 vs SLC2' SLC 1 -1 0; contrast 'SLC1 vs SLC3' SLC 1 0 -1; contrast 'SLC2 vs SLC3' SLC 0 1 -1; run;   Result:   SAS Output The SAS System The Mixed Procedure Model InformationData SetDependent VariableCovariance StructuresSubject EffectsEstimation MethodResidual Variance MethodFixed Effects SE MethodDegrees of Freedom Method A1.DISSERTATION2 GMP_Syl Variance Components, Heterogeneous Autoregressive SubjectID, SubjectID REML None Empirical Containment   Class Level InformationClass Levels ValuesSLCGroupSubjectIDTrial1 3 1 2 3 2 1 2 24 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 10 1 2 3 4 5 6 7 8 9 10   DimensionsCovariance ParametersColumns in XColumns in Z per SubjectSubjectsMax Obs per Subject 12 12 1 24 60   Number of ObservationsNumber of Observations ReadNumber of Observations UsedNumber of Observations Not Used 1440 1317 123 ... View more

I see. Thank you for your help and advice!     ... View more