Bhupinder,

Well first thing I would ask is what the current litrature says.

Having a Master in Ag. Econ, in which we did these types of analysis on a daily basis Id look for:

Seasonality, watering practices, crop rotation patterns and what crops did you rotate with, how much Fertilizer: when was it applied and what type was it, pesticides: what type how much was applied and when, farming practices (ie. no-till vs. till methods), since you said multiyear did you use the same seed or our you comparing different seeds, Temperatures come into play. What you find is to model something like this well it has to take into account many factors.

Once you have your data, then start with a basic model and look at such things as Adj R^2, residuals, P-Values. Make sure your specification is correct, you have no heteroskadasticity, multicollinearity, etc.... More than likely you will have a non-linear model so you might want to look at regressing your residuals against your variable and variable squared. Another thing we were playing with when I left was a multi-mispecification test to ensure that the model was specified correctly.

On Fri, 5 Feb 2010 07:41:54 -0600, Bhupinder Farmaha <bhupi80singh@YAHOO.CO.IN> wrote:

>-----Original Message----- >From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of Peter Flom >Sent: Friday, February 05, 2010 5:23 AM >To: SAS-L@LISTSERV.UGA.EDU >Subject: Re: Normality problem > >Bhupinder Farmaha <bhupi80singh@YAHOO.CO.IN> wrote >> >> >> >>I need your suggestion about appropriate measure to tackle non-normal data. >>I ran an agronomic experiment for three years. I am modeling it using Proc >>Mixed. I found that data do not meet the assumption of normality. The >>problem I found that response variables values are quite different for one >>year compared to other two years. The part of the reason it is weather >>driven. I have tried log transformation but it didn't work at all. I don't >>know if I have to give less weights to data from that year or what can be >>the other approach. It is clear from the some of the preliminary that data >>have this problem because crop is adversely affected in one particular years >>compared to others. >> >> >> >>I would appreciate any feedback on this. >> >> >> > >What is it that you are trying to model? > >What is your DV and what are your IVs? > >From what you say, it sounds like your DV is something related to crops. Maybe crop yield? > >But a couple things: >1) The fact that the DV is different year to year is not necessarily a problem for PROC MIXED. > >2) The fact that the data (I'm guessing you mean the DV) is not normal is not necessarily a problem either, you need to look at the residuals. > >3) I'm not sure what you mean by "weight the data"; if the problem is outliers, then you might want some form of robust regression, but I don't know if there is a way to do robust mixed regression in SAS. It might be possible in NLMIXED. > >4) Have you looked at PROC GLIMMIX and PROC NLMIXED? > >HTH > >Peter > >Peter L. Flom, PhD >Statistical Consultant >Website: http://www DOT statisticalanalysisconsulting DOT com/ >Writing; http://www.associatedcontent.com/user/582880/peter_flom.html >Twitter: @peterflom > >I am trying to model grain quality parameters like protein and oil content. The dependent variables are fertilizer and management practices. > >I did look at the residuals and it show some kind of pattern. > >I didn't look up at GLIMMMIX and NLMIXED. > >Thanks >Bhupinder