Thank you for your response! I at least feel like I am on the right track, even if some of the finer details still need to be worked out. I wanted to add a couple of thoughts in response, if not for anything to help others who have stumbled across this post in search of getting their own related questions answered - browsing through forum discussion is where I have learned a great deal of what I know about GLMMs...   After I changed the covariate from NO3cov to log(NO3cov), the improvement in the model was appreciable. The standard errors for my estimated lsmeans decreased, and the AICC decreased to a value about equal to the baseline model without the covariate. I understand that using a metric like AICC is tricky in situations when pseudo-likelihood is involved and that the Laplace or Quad methods would be most appropriate. When I have enough computing downtime, I'll run one of these models because at this point the runtime is just too long for this model + data set... I also wanted to add that I moved away from the AR(1) + CS to an ANTE(1) R-side covariance structure. After doing some additional exploratory and diagnostic work, I informally found this structure to be the best fit.   Addressing some of the broader questions you pointed out at the end of the post, here are my thoughts on why I chose the approach that I did. Historically, this type of data (Concentration_NO3) has been used as an intermediate value in computing a response variable of agronomic interest (but which we do not directly measure) - NO3 leaching load. Concentrations are in units of mg/L while loads are in units of kg/ha. In short, NO3 load is calculated by multiplying a daily estimated NO3 concentration by a daily estimated flux of water downward through the soil (this is a whole other problem of estimate, fraught with its own issues...). We measure NO3 concentration on only a handful of sampling dates, due to logistical constraints. In previous analyses, the raw data for each experimental unit was interpolated across the growing season to obtain daily estimates of concentration for each experimental unit. In this manner, an estimate of NO3 leaching load was obtained for each experimental unit by multiplying the interpolated NO3 concentration by the daily estimated percolation below the root zone (i.e. downward water flux), and more standard split-plot mixed model analysis was appropriate.   However, serious red flags were raised for this method which questioned the reliability of this method of analysis. We found in some cases that our numerical estimates of NO3 load for the control N_Trt was higher than the high rate N_Trt, which was very likely an artifact of either a failed experiment or failed analysis. In other cases, we found no significant differences between any of the N_Trt or Irr_Trt which suggests that the power of this method was too low to be suitable.   We decided to try a new approach to this analysis, focusing on the measured NO3_Concentration as the response variable of primary importance rather than the calculated NO3 load. The thought was simply that we should make our statistical inferences before proceeding with calculations (more on this in a bit). And then an interpolation-type approach could proceed using the estimated means for Irr_Trt*N_Trt. Using a non-linear regression approach is possible, but the shape of the response curve is problematic. Within a given growing season, the response curve has a similar form for all treatments; however, the shape of the response curve is not similar across growing seasons. I have 5 years of similar data to analyze, and each has a distinct and non-linear curve for NO3_Concentration as a function of time. An analysis with time as a categorical, rather than continuous, variable seemed appropriate in this respect.   The goal is then to compare the treatment means on each Sample_Date, or at least to compare a given set of non-orthogonal contrast statements on each Sample_Date. I also sliced my analysis of the lsmeans on the Sample_Date effect to evaluate the N_Trt, Irr_Trt, and their interaction on each date. I didn't include this in my original post, but here is a partial snippet of code for reference:   lsmeans Irr_Trt N_Trt Irr_Trt*N_Trt Irr_Trt*Sample_Date N_Trt*Sample_Date Irr_Trt*N_Trt*Sample_Date/CL alpha=0.1 slice=Sample_Date;   contrast 'N1' N_Trt -5 1 1 1 1 1; contrast 'N2' N_Trt 0 -1 -1 1 1 0; contrast 'N3' N_Trt 0 -1 1 -1 1 0; contrast 'N4' N_Trt 0 0 0 -1 -1 2;   contrast 'I*N1' Irr_Trt*N_Trt 5 -1 -1 -1 -1 -1 -5 1 1 1 1 1; contrast 'I*N2' Irr_Trt*N_Trt 0 1 1 -1 -1 0 0 -1 -1 1 1 0; contrast 'I*N3' Irr_Trt*N_Trt 0 1 -1 1 -1 0 0 -1 1 -1 1 0; contrast 'I*N4' Irr_Trt*N_Trt 0 0 0 1 1 -2 0 0 0 -1 -1 2;   contrast 'N1 @ D1' N_Trt -5 1 1 1 1 1 N_Trt*Sample_Date [-5,1 1][1,2 1][1,3 1][1,4 1][1,5 1][1,6 1]; contrast 'N2 @ D1' N_Trt 0 -1 -1 1 1 0 N_Trt*Sample_Date [0,1 1][-1,2 1][-1,3 1][1,4 1][1,5 1][0,6 1]; contrast 'N3 @ D1' N_Trt 0 -1 1 -1 1 0 N_Trt*Sample_Date [0,1 1][-1,2 1][1,3 1][-1,4 1][1,5 1][0,6 1]; contrast 'N4 @ D1' N_Trt 0 0 0 -1 -1 2 N_Trt*Sample_Date [0,1 1][0,2 1][0,3 1][-1,4 1][-1,5 1][2,6 1];   The contrasts I chose reflect the a priori hypotheses that the experiment was designed with, relating to different nitrogen rate, source, and timing treatments. I am not interested, necessarily, in a full blown multiple comparisons of my treatment means on a given date, but rather if any of the specified contrasts are significant. Becuase of the high CV on this data, I wanted to be especially conservative when interpreting my results and before using the concentration data to make estimates of N-leaching load and to prevent a situation in which calculated N-leaching loads have numerical differences without any significant statistical differences. The mean comparisons of the contrast statements on each Sample_Date is effectively a protected version of the original computational procedure.   The next steps are to use the results from this analysis to compute NO3 load, which is the real response variable of interest. Although I will not be able to conduct a formal statistical analysis of the computed data, I should end up with a set of values for NO3 load where numerical differences represent true and significant differences between treatments. In the end, an analysis of the main effect of Sample_Date is just an intermediate procedure that matters for computational purposes, but by itself is not all that interesting.   If anyone else has thoughts on this approach or ways in which I could improve it, feel free to contribute! ... View more