We've pretty much been acting like population and density effects will be additive, however I'm not so sure about environment.  We've been treating that, and its interaction with population and density, as a source of error in the random statement.  Since we already have the logistic curve parameters from NLIN, we can pretty much cross that off the list.  We have a total of 42 different GD10 points.  While they should technically contain each population/density combination, things like bad weather and poor growth made it so that in several of the points we're missing one or more of these combinations.  Time constraints also made it so that some days had less samples taken than others, leading to somewhat unbalanced data.  Additionally, only one or two environments is represented at each GD10 mark since different environments had different temperatures and therefore different GDD values.  Also, since the last two GD10 values were final harvest values, we have vastly more data for those two points than any of the other ones, if that matters. ... View more

Well, it's becoming increasingly clear that I have very little idea of what I'm actually doing, but what better purpose in life than to serve as a warning to others.  GD10 (growing degree days divided by 10) has replaced days since it is a more reliable measure of time for growth data, and is supposed to be continuous as a covariate.  Density really should be set up as a class statement variable, so I went ahead and did that.  However, running the proc only led to it not converging, even with iterations increased to 100, so looks like we're back to square one no matter what. Poisson distribution is used for counts, but would gamma distribution or some other one work here?  The NLOPTIONS page states that most models fit with the GLIMMIX procedure typically have one or more nonlinear parameters, so does it have any use with my conundrum?  The data does take a logistic shape over time, with the shape of that logistic curve being affected by density and population, so is there any way to treat the length-GD10 relationship as logistic rather than linear?   I would like to stay away from trying to utilize A, B, and C as dependent variables to be analyzed because that takes us into the cold, dark pit that is multivariate statistics.  So let's see what we can do with NLMIXED or GLIMMIX.  I've been looking through the SAS-L archives like you've said, but everything there that has to do with NLMIXED seems to be about people wanting to find the values of parameters to model their equations, while I simply want to find type III fixed effects for density, population, and density*population. Thanks for being so patient with me, I know it seems like we're taking one step forward and two steps back. ... View more

Let's just assume that we did want to go with GLIMMIX because the other bit sounds really complicated.  Would this code work?  I'm really the first person in our lab to majorly have to work with nonlinear data, so I'm kind of the guinea pig for NLIN, NLMIXED, and GLIMMIX. proc glimmix data=short; class breed rep environment; model length = breed|density|GD10 / ddfm=satterth dist=poisson link=log; random rep(environment) breed*density*rep(environment); run; I'm pretty new to the whole ideas of dist and link, so would poisson and log cause GLIMMIX to treat it as a logistic rather than linear model?  Also, I know you are a fan of kr2 for ddfm, but I was recommended satterthwaite for this. ... View more

I've been looking through SAS-L and the SAS books and online resources, and I think I may have made a terrible mistake.  Is it even possible to run a test for type III fixed effects using NLMIXED?  If not, I think I may be barking up the wrong tree and need to go with either GENMOD or GLIMMIX. ... View more

Is the logistic equation flexible enough to work like that? I tried the new equation and just got a whole lot of this: Execution Errors for OBS 1: NOTE: Missing values were generated as a result of performing an operation on missing       values.       Each place is given by (number of times) AT       (statement)/(line):(column). ... View more

Here is a basic version of my NLIN code for one of the populations at all 4 densities.  I've done this for all 16 combinations.  As you can see, I really haven't done it where fixed effects are parameters, I've instead broken them all up.  It seems I've given us a bit more pain to deal with. proc nlin data=bscb13; parms A=232.6 B=1332000 C=0.2065; model length = A/(1+(B*EXP(-C*days))); run; proc nlin data=bscb16; parms A=185.2 B=563354 C=0.1951; model length = A/(1+(B*EXP(-C*days))); run; proc nlin data=bscb19; parms A=234.3 B=1738760 C=0.2020; model length = A/(1+(B*EXP(-C*days))); run; proc nlin data=bscb112; parms A=189.7 B=1686900000 C=0.3048; model length = A/(1+(B*EXP(-C*days))); run; As you can see, they're all fairly different when it comes to the parameters. We've been thinking of both density and population as fixed effects because with only 4 different levels for each, you can't have a very good estimate of variance.  Additionally, we're mainly focusing on the data we have, so we're not looking to make giant inferences about all of the species.    Again, with only 4 environments, we don't have a good idea of the variance, but with our 64 population*density*environment classes, we have a better idea.  Each data point in this set is an entirely different plant (we had to hack them down to get the data), so we can't really use individual plants as the subject.  However, each field we used had 9 replicates so would population*density*rep(environment) be a good subject while the random effect would be population*density*environment?  This is the closest you're probably going to get to the individual trees in the example. ... View more

I know there is the NLOPTIONS statement for GLIMMIX, but we're pretty in the dark about it (I'm the first person in the lab to work with nonlinear data).  Do you need to do anything to specify a logistic curve as opposed to something like a gompertz or a monomolecular curve?  ... View more

I should probably add in that we have already used PROC NLIN and have our three parameters for the logistic equation for each pedigree/density match. ... View more