About lvm

lvm · ‎03-01-2016

You should not be using the Scheffe adjustment option. This is way too conservative. Change to Tukey, or see what happens without the option at all.

lvm · ‎03-01-2016

...One more thing.... Quoting from the HPMIXED User's Guide (applies also to MIXED): "You should exercise caution in defining the GROUP effect, because strange covariance patterns can result with its misuse. Also, the GROUP effect can greatly increase the number of estimated covariance parameters, which can adversely affect the optimization process".

lvm · ‎03-01-2016

You could switch to PROC HPMIXED, designed for large-scale problems. However, I think you want the residual to have different groups, not the random effect. repeated / group=fungicide*isolate; In fact, based on the limited information, plate(trial) could be your residual, which would mean that you have an over-parameterized model (since the residual is already there). You might want to try: random int / sub=trial; repeated / group=fungicide*isolate; With all that said, it is usually not good statistical practice to fit so many variances in a model. This usually reduces your power, and can lead to all kinds of estimation problems, even when there is enough memory.

lvm · ‎02-23-2016

Use MIXED or GLIMMIX, and treat company and year as random effects.

lvm · ‎02-23-2016

You really need to switch to MIXED or GLIMMIX for this. See SAS for Mixed Models, 2nd edition (2006). The syntax is a bit different compared with GLM. In particular, you do NOT put random effects in a model statement; random effects are given only in one or more random statements. Based on your description, I think blocks are nested within sites. Thus, there would be no block main effect, but there would be a block(site) effect. The procedure automatically figures everything out that you are concerned about. PROC MIXED DATA=work1; CLASS Treatment Site Block; MODEL Tb_BVTV= Treatment Site Treatment*Site Covariate(1) Covariate (2) / SS3; random block(site); LSMEANS Treatment Site Treatment*Site / PDIFF=ALL ADJUST=TUKEY ; RUN;

lvm · ‎02-18-2016

Beta does not allow 0 or 1 (defined only between those values). You would get a missing value for any 0s or 1s. This is a challenge when using the beta distribution. Skewness is not a problem. Beta can be very skewed. If you use beta, then you don't need the random _residual_ term I mentioned (there is a scale parameter automatically). If you want beta with 1s, you need to define (in an ad hoc way) a new variable that is always between 0 and 1. Something like yprime = (y-.005)/1.01. Be careful, results will depend on the constants you use.

lvm · ‎02-17-2016

Many SAS procedures print -2LL as a convenience, since one just has to subtract values without remembering to multiply by 2 or -2. The smaller the -2LL, the better the fit (ignoring the differneces in parameter numbers). The LRT is just the difference.

lvm · ‎02-17-2016

You missed the point, maybe I wasn't clear. You need to specify a distribution with link=log, otherwise you will get a normal distribution. If you want a binomial, then use dist=binomial or some other distribution (there is no such thing as "dist=option"). If you are using normal distribution (the default), then you are using pseudo-likelihood (the default).

lvm · ‎02-16-2016

You have several issues/problems. With your code with link=log, you are using a normal distribution (the default). This is likely not what you want. Put in the dist= option. Also, your syntax is wrong for GLIMMIX. One does not use "y ~ x". One uses "y=x". Also, you are using pseudo-likelihood, not quasi-likelihood. Quasi-likelihood is when the likelihood is not defined (or definable). If you put in: random _residual_; then you would have quasi-likelihood with the binomial. Another way of getting quasi-, is to specify the mean and variance functions directly. Both ways are defining a variance:mean structure that does not correspond to a real probability distribution.

lvm · ‎02-15-2016

Just add the diff option to the lsmeans statement.

lvm · ‎02-09-2016

This forum is not set up for such general questions. We don't write complete code from scratch. There is plenty of material in the User's Guide to get started using NLMIXED for the negative binomial (it is a built-in function), and for having a random intercept term. The book SAS for Mixed Models, 2nd edition (2006) also has lots of material on this topic. Once you have written and tested some code, we can then give specific advice on problems you run in to,.

lvm · ‎02-01-2016

Forgot the mention above, your SS should not be 0 because your y values are not exactly equal to x**2 (the first y should be 1, not 2, if you want an exact match).

lvm · ‎02-01-2016

Why do you think the results are wrong? I just ran your program with plots turned on, and everything looks fine for the fit. Taking out the weight statement gives a different confidence region, different SE, etc.,as expected. I think you just have a coincidence about the "0.1". For instance, I ran: data b; set startdata; pred = x**2; SS = (y - pred)**2; WSS = z*SS; run; proc means data=b sum; var SS WSS; run; To manually get the sum of squares, weighted and unweighted, based on your parameter estimate. Weighted SS = 0.1, just like the NLIN output. If you run NLIN without the weight, you get SS = 1.0. This is what you get manually (above). I think everything is fine. By the way, put a plots=all option in your NLIN statement for good graphs.

lvm · ‎02-01-2016

You have to be careful in the coding of the data. If rep is corresponding to each unique subject, then there would be confounding of the rep variance and the negative binomial (NB) scale parameter. You could not have both (either rep with Poisson, or NB and no rep). If "rep" corresponds to "blocks", so that there are different groups (e.g., treatments) all with the same rep level value, than it should work (no confounding). You didn't give enough output to identify the problem. Having a lot of zeros can happen when there is a NB or when there is a Poisson with a large random effect variance.

lvm · ‎01-27-2016

You can get this to converge in GLIMMIX (even with that extreme plot that was identified). proc glimmix data=a ; class plot; model cmi = year / ddfm=kr solution ; random int year / type=un subject=plot solution; run; GLIMMIX uses a different optimization method by default than MIXED. Converges in 18 iterations. A nice thing about GLIMMIX, you can use the NLOPTIONS statement to specify other optimization methods (wasn't needed here). To get output, use the OUTPUT statement (the out option on the model statement is not used in GLIMMIX). If you look at the output, you will see that the UN(2,2) variance, the variance of slope, is basically 0. This results in missing values for standard errors of parameter estimates and the EBLUPS. Results suggest an overparameterized model for this response variable. You could reduce the random effect to a random intercept model.

Online Status	Offline
Date Last Visited	‎10-02-2024 05:21 PM

Re: mianalyze of lsmestimate

Re: mianalyze of lsmestimate

Re: TEMPLATE: how to combine the equivalent of LAYOUT LATTICE and LAYO...

TEMPLATE: how to combine the equivalent of LAYOUT LATTICE and LAYOUT D...

Re: SAS code for proc glimmix data - interaction analysis

Re: Mixture of chi square with NLMixed in sas

Re: Stepwise Model Selection for longitudinal binary data using PROc G...

Re: Calculating weight for site effect based on standard error

Re: Estimating treatment effects, 2 Group Pre-Post Matched Analysis

Re: Proc Mix insufficient memory issue

Re: GLIMMIX: order of random variable syntax

Re: mianalyze of lsmestimate

Re: mianalyze of lsmestimate

Re: SAS code for proc glimmix data - interaction analysis

poisson regression goodness of fit stats

Re: Stepwise Model Selection for longitudinal binary data using PROc G...

Re: two sample t-test

Re: Two Way ANOVA fitting Heteroskedasticity - Running out of Memory

Re: Two Way ANOVA fitting Heteroskedasticity - Running out of Memory

Re: 2 fixed effects and clustered std errors

Re: How to include a block factor in ANOVA/ANCOVA model

Re: Quasi-likelihood regressions

Re: negative LRT statistic values

Re: Quasi-likelihood regressions

Re: Quasi-likelihood regressions

Re: Getting confidence interval in PROC MIXED

Re: Need help with proc nlmixed coding

Re: Proc nlin with weights and SSE

Re: Proc nlin with weights and SSE

Re: Analyzing count data in GLIMMIX

Re: Error: "Did not converge" in Proc mixed