turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Multi year data analysis (procedure other than pro...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

04-13-2016 01:51 PM

Hi,

I have a 5-year experiment, with four crops in 4 rotations. There were a total of 48 plots, 12 plots per block. I tested the datasets for each of the year for normality and only one year out of the 5 years had a data that is not normally distributed. When I tested the 5-yr data (multi-yr) for normality, it was normally distributed. I ran the analysis on effect of rotations on yield using proc mixed, but none was significant.

1. Can someone please help me with a code for other SAS procedure apart from proc mixed for comaprison of the results from proc mixed?

2. I read about proc corresp and multivariate analysis, which of these is better for my type of analysis

3. Do I have to transform (normalize) the year that is not normally distribute before performing a multi year analysis?

Thanks.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

04-16-2016 04:57 PM

This forum may not be the best place for such general questions. You didn't provide any code, so we have no idea if you did the analysis correctly (or reasonably) with MIXED. How are you testing for normality? From the residuals after a model fit (that would be the reasonable way)? Are you trying to do the analysis across all 5 years in one model fit? There are MANY statistical and methodological issues when modelig crop rotation experiments over years. Not always clear what to use for the model. You should read the papers in the agronomy literature by H.P. Piepho and others.

It is bad statistical practice to look for another method of analysis (another PROC) to get the result you *want*. That is not the reason to choose a method. (Rotation may not be affecting yield). Your second question indicates that you really need a lot of advice to properly frame the questions you are trying to address, and how to go about testing the questions. I highly recommend that you find a good applied statistician who can talk things through with you. I think you need more than the advice on coding that we can give here.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

04-18-2016 10:36 AM - edited 04-18-2016 10:38 AM

Hi Ivm, Thanks for your observation and advice. First I conducted the analysis across and then within the 5 years because I am interested in both. Please find below the code used

proc mixed data= AllYield2015T;

class Year Rep T_System;

model &var = T_System;

random Rep(year)year year*T_System/solution;

repeated year/subject=rep*T_System type=ar(1) RCORR;

Lsmeans T_System/diff=all;

run;

After this code, I had my contrast statements.

And also below is the code I used to test for Normality:

proc print data = Pred;

var year resid Pred;

proc plot data = Pred;

by year;

plot resid*Pred resid*T_System/vref=0;

proc univariate normal data = Pred;

by year;

var resid;

ods output moments = getsd;

proc univariate data=Pred;

by year;

var resid;

histogram resid / normal;

QQplot resid;

run;

2. I am not trying to work from the answer (looking for how to get exactly what I thought the results should be), I was only trying to run another analysis (MAYBE MULTIVARIATE WHICH I AM NOTFAMILIAR WITH YET) to be sure of the results I have. I have always believed in 2nd or 3rd opinion in whatever I do. I will go through the manuscripts by H.P. Piepho as you suggested. AT THE SAME TIME ANY HELP / SUGGESTION WILL BE APPRECIATED

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

04-25-2016 11:04 AM

Mixed model methodology, independent of the procedure selected, is the only way to generate the appropriate standard errors needed for mean comparisons in designs of this sort. NO OTHER PROCEDURES IN SAS WILL ALLOW FOR THIS. Even PROC MCMC, which takes a Bayesian approach, applies a mixed model framework.

The only other approach you might look at would be to write your own maximum likelihood estimation routine in IML--but it ought to give you what PROC MIXED/GLIMMIX/HPLMIXED/HPMIXED?MCMC?NLMIXED do.

Steve Denham

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

07-05-2016 10:55 AM

Thanks a lot Steve.