cancel
Turn on suggestions
Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.
Showing results for 
Search instead for 
Did you mean: 
palolix
3 hours ago
  • 204 Posts
  • 85 Likes Given
  • 0 Solutions
  • 7 Likes Received

Re: testing for a negative correlation between 2 dependent variables i...

by Pyrite | Level 9 in Statistical Procedures
‎12-05-2014 11:31 AM
‎12-05-2014 11:31 AM
Hi Steve! I am so glad you came to help me . Thank you!! I already check the distribution and the lognormal was the most suitable (is usually the case for nematode counts). harEggsT (total eggs) is one of the dependent variable that I want to test for negative corr with harEggsD (diseased eggs), so I did not know how to include both dependent variables in the same model, but I will try what you are suggesting. Thank you very much Steve, Caroline ... View more

testing for a negative correlation between 2 dependent variables in a ...

by Pyrite | Level 9 in Statistical Procedures
‎12-05-2014 03:35 AM
‎12-05-2014 03:35 AM
Dear SAS Community, I have analyzed my data as repeated measures, as the dependent variables were measured every year in a multi-year trial. I have 2 independent variables; nem and year. When plotted my data I observed that the population dynamics of two dependent variables (number of total eggs and number of diseased eggs) show opposite oscillations during the 5 years evaluated, so I would like to test for a negative correlation between these 2 variables. Is there any way to modify this repeated measures code in order to test this or should I test the correlation for each year separately? Proc glimmix data=one; class nem blk year; model harEggsT= nem|year/dist=lognormal ddfm=kr; random intercept/subject=blk; random year/residual subject=blk*nem type=ar(1); run; I would greatly appreciate any help! Thank you, Caroline ... View more

Re: MANOVA in GLIMMIX?

by Pyrite | Level 9 in Statistical Procedures
‎05-21-2014 02:29 PM
‎05-21-2014 02:29 PM
Thank you dear Steve!  method=laplace works wonderful for binomial and negbin when analyzing the sequences as univariate.  I tried the multivariate approach analyzing the two nematode populations separately, due to the hugh differences we observed between both, but still did not work.  I think that the problem is due to the high variability among replicates within a population.  For most bacterial species (dependent var) happens that for one replicate 400 sequences (more or less) were registered whereas for the other 4 replicates 0 sequences were found. I guess that is also the reason why for most of the species (in the univariate analysis) I do not find any significant differences between the two populations. Caroline ... View more

Re: MANOVA in GLIMMIX?

by Pyrite | Level 9 in Statistical Procedures
‎05-15-2014 08:56 AM
‎05-15-2014 08:56 AM
Hi Steve, thank you very much for your good Ideas and suggestions.  I´m sorry I could not reply before, but my SAS License expired.  You are right, I did not expected such a differences between the two populations, so I understand the problems you were facing when trying with different optimizations.  The graphs and the tables were for sure a good Idea to see those differences.  I think before getting deeper and deeper with this multivariate approach including the 2 populations, I would rather leave this on standby and go ahead with an univariate analyze for comparing the two populations, and a multivariate analyze for each population separately. In relation to the univariate analyze, I would like to ask you something; when should you use the option method=quad or laplace for a binomial or negbin distribution?  Is it preferable in order to optimize your model? Thanks a lot Steve!! Caroline ... View more

Re: MANOVA in GLIMMIX?

by Pyrite | Level 9 in Statistical Procedures
‎05-01-2014 02:35 PM
‎05-01-2014 02:35 PM
Mmmm.... it cannot be, I am probably doing something wrong or we are missing something. I reduced the species to 5 but still cannot converge.  I tried the genmode, changing the code a little bit due to the offset statement (I hope was ok), but did not work either.  I will send you my data attached so you can better see if we are missing something (the analysis with 10 species). Thank  you for still helping me Steve!! Caroline ... View more

Re: MANOVA in GLIMMIX?

by Pyrite | Level 9 in Statistical Procedures
‎04-29-2014 02:59 PM
‎04-29-2014 02:59 PM
I have 2 populations of nematodes with 5 replicates each, which makes a total of 10 experimental units.  In long format, I have 140 rows;14 bacteria species (dependent variable) evaluated for each exp. unit. I tried the negbin dist, but unfortunately did also not converge.  Do you think it may have something to do that the dependent var "species" in the random statement is not a numeric variable? Thanks a lot Steve! Caroline ... View more

Re: MANOVA in GLIMMIX?

by Pyrite | Level 9 in Statistical Procedures
‎04-29-2014 04:06 AM
‎04-29-2014 04:06 AM
Thank you very much for your help Steve.  I followed your first advice changing to V, but unfortunately did not work.  I reduced then the species number from 26 to 14, trying this 2 codes, but still getting following warnings; proc glimmix data=one method=laplace; class blk pop species; model No_seq/total_seq=pop species pop*species/dist=binomial link=logit; random species/sub=blk*pop type=chol GCORR; run; WARNING: The initial estimates did not yield a valid objective function. proc glimmix data=one; class blk pop species; model No_seq/total_seq=pop species pop*species/dist=binomial link=logit; random species/residual sub=blk*pop type=chol V; run; WARNING: Pseudo-likelihood update fails in outer iteration 0. Do you think would be a good Idea to try with a negbin distribution (because of the zeroes)? Or should reduce the number of species to 10? Thank you Steve!! Caroline ... View more

Re: MANOVA in GLIMMIX?

by Pyrite | Level 9 in Statistical Procedures
‎04-28-2014 11:20 AM
‎04-28-2014 11:20 AM
Thank you Steve. I tried random species/sub=blk*pop type=chol GCORR; but got this WARNING: The initial estimates did not yield a valid objective function. I tried then the last code you gave me (with the residual option), and got this: ERROR 22-322: Syntax error, expecting one of the following: ;, (, ALPHA, CL, G, GC, GCI, GCOORD,               GCOORDS, GCORR, GI, GROUP, KNOTINFO, KNOTMAX, KNOTMETHOD, KNOTMIN, LDATA,               RESIDUAL, RSIDE, SOLUTION, SUBJECT, TYPE, V, VC, VCI, VCORR, VI. ERROR 202-322: The option or parameter is not recognized and will be ignored. Caroline ... View more

Re: MANOVA in GLIMMIX?

by Pyrite | Level 9 in Statistical Procedures
‎04-28-2014 07:58 AM
‎04-28-2014 07:58 AM
Hi Steve!  thank you so much for your suggestion.  It sounds like a very clever Idea to analyze it as "Repetead measures", I guess because of the possibility of correlation between observations on the same subject. I run your code, but I had a little problem due to syntax error because of the Chol R (I guess this is not available for SAS 9.3).   I tried with chol only, but did not yield any result.  I also tried with UN and it works, but I don´t know if it is a good Idea. I guess I could also try all the others to select the one with the lowest AIC.  Thank you Steve! Caroline ... View more

MANOVA in GLIMMIX?

by Pyrite | Level 9 in Statistical Procedures
‎04-24-2014 11:54 AM
‎04-24-2014 11:54 AM
Dear SAS Community, Is it possible to perform a MANOVA with GLIMMIX? I could only finf examples with GLM.  I have a RCBD with data having many zeroes.  I want to compare bacterial DNA sequences (26 dependent variables) found in 2 different nematodes populations (pop) in a multivariate analyze.  Because these are proportions (event/trial) response variables (bacterial sequence/total number of sequences), I would use a binomial or negbin distribution. Here is my data with factor pop having 2 levels, the total number of DNA sequences found for every population, and the number of sequences of each one of the 26 dependent variables.  I analyzed the first dependent variable as a univariate, but  I would greatly appreciate if someone could help me with an approach for a multivariate analyze with GLIMMIX.  Thank you in advance!! data one; input blk pop total_seq  A_junii A_oryzae A_brasilense B_vestrisii B_caledonica C_pinensis C_hispanicum C_metalli D_chinhatensis D_neptuniae F_feengrott M_polyspora M_saperdae N_simplex P_parvula    P_carboxy R_pickettii R_solanace R_syzygii R_giardinii R_blasticus S_thermophilus S_maltophilia S_flavogriseus S_vinaceus Z_ramigera; datalines; 1    1    1116    0    7    0    9    0    1    0    2    1    0    0    715    92    0    0    2    5    3    10    14    0    0    0    0    0    0 1    2    1557    0    5    0    0    0    0    0    0    0    0    3    0    2    0    0    0    74    141    431    19    0    0    0    0    0    170 2    1    1203    0    6    200    0    0    683    0    0    0    0    0    0    0    0    0    0    4    6    37    1    0    0    0    0    0    28 2    2    995    0    21    0    0    0    0    0    0    618    0    0    0    0    0    0    5    45    25    67    0    0    0    0    0    0    0 3    1    230    0    0    0    0    0    0    0    0    0    53    0    0    0    0    0    0    0    1    0    0    0    92    0    0    0    0 3    2    1217    0    29    0    0    0    0    0    0    0    0    0    0    0    0    0    16    37    26    85    0    0    0    20    54    700    0 4    1    1287    0    1    0    1    1249    0    0    0    0    0    0    0    0    0    0    3    0    3    4    0    0    0    1    0    0    1 4    2    464    0    0    0    8    0    0    0    2    0    0    236    0    9    51    0    4    1    1    1    0    0    0    0    0    0    1 5    1    151    0    0    0    0    0    3    0    0    0    0    0    0    0    0    0    0    0    0    0    98    0    0    0    0    0    0 5    2    1738    68    108    0    110    0    0    68    50    0    0    0    0    0    0    82    73    87    26    45    178    104    0    99    0    0    7 proc glimmix data=one method=quad; class blk pop; model A_junii/total_seq=pop/dist=binomial link=logit; random int/sub=blk; run; Thank you!! Caroline ... View more

Re: Code to display Mean square errors when using proc mixed

by Pyrite | Level 9 in Statistical Procedures
‎02-10-2014 05:05 AM
‎02-10-2014 05:05 AM
Thank you very much for the comment!  I´m glad to know this. Eventhough works for my data what data null suggested, I will keep in mind what you said about proc mixed and will be carefull when using the information obtained. Thanks a lot!! Caroline ... View more

Re: Code to display Mean square errors when using proc mixed

by Pyrite | Level 9 in Statistical Procedures
‎02-10-2014 05:01 AM
‎02-10-2014 05:01 AM
Thank you very much data null for the trick, it works for my data!  Nevertheless I will keep in mind what lvm said. Thanks a lot!! Caroline ... View more

Code to display Mean square errors when using proc mixed

by Pyrite | Level 9 in Statistical Procedures
‎02-07-2014 06:41 AM
‎02-07-2014 06:41 AM
Hi all, When using GLM I always get a table with the mean square errors for each of my dependent variables, but this is not the case when using proc mixed; is there a code that I can add to my model when using proc mixed in order to get this information?? Thank you!! Caroline ... View more

Re: Estimates for split-plot with missing treatment combinations

by Pyrite | Level 9 in Statistical Procedures
‎09-16-2013 12:57 PM
‎09-16-2013 12:57 PM
Hello again Steve, When analyzing some dependent variables, where I only need 2 levels of the factor 'cultivar' (which has 3 levels; Nordstern, Sang and Beretta), so the number of observation that sas use are 27 out of 40, I only get the LSM but not the estimates.  The problem is that in my Type 3 Test of Fixed Effects Table I only get the 'nem' and 'cultivar' main effects but not the interaction nem*cultivar (which I suppose is not estimable), so I think that is why I am not getting the estimates.  When writing the estimates statements I think I should only write he main factors but not the interaction: lsmestimate nem cultivar 'label' [1, 1 2][-1, 2 2], instead of nem*cultivar 'label', but if I just write nem cultivar 'label', sas tells me that is expecting a * (nem*cultivar), so it does not work either. Here is my code: proc mixed data=one; where cultivar="Nordstern" or cultivar="Sang"; class blk nem cultivar; model ahreave= nem cultivar nem*cultivar/ddfm=kr; random blk blk*nem; lsmeans nem*cultivar; lsmestimate nem*cultivar  'nem1 vs nem2 | cultivar=Norstern' [1, 1 2][-1, 2 2],                                       'nem1 vs nem3 | cultivar=Sang' [1, 1 3][-1, 3 3],                                                                      'cultivarNord vs cultivarSang | nem=1' [1, 1 2][-1, 1 3];                                                               run;   Is there a way to write the lsmestimate without the interaction nem*cultivar (since this is not estimable)?  Do you think this is the reason why I am not getting the estimates results? I would greatly appreciate if you could give me some light. Thank you very much Steve! Caroline ... View more

Re: Estimates for split-plot with missing treatment combinations

by Pyrite | Level 9 in Statistical Procedures
‎09-05-2013 04:12 PM
‎09-05-2013 04:12 PM
Thank you very much Steve for your good suggestions and great support!! All thes best, Caroline ... View more
Contact Me
Online Status
Offline
Date Last Visited
3 hours ago
Likes given to
Likes from
Quick links