BookmarkSubscribeRSS Feed

Modeling Repeated Measures in Field Trials

Started ‎06-17-2020 by
Modified ‎06-17-2020 by
Views 1,459

Researchers who conduct field trials find time point data (repeated measures) one of the most difficult types of data to wrangle. One common pitfall among enthusiastic amateurs is treating repeated measures as if they were independent. For independence to be true, treatments must be randomly assigned to experimental units, and it’s impossible to randomly assign time points to a single field plot or treatment.

 

One way around this is to apply an integral to each experimental unit, such as AUDPC. A more elaborate way would be to account for the variance caused by time while modeling the impact of the treatments. Block diagonal covariance matrices defray the time-based covariance errors across treatments.

 

PROC MIXED offers a ‘Repeated’ statement to specify the repeated variable. Fisher, 1921 data (the progenitor to the F-Statistic) provides an interesting example, where 70 years of wheat yield data were measured among strips. Source and data citations provided at the end of this post.

proc mixed data=work.fisher plots=all;
 class Alternating Kg_Mg Kg_N Kg_Na Kg_P yr;
 model Yield=Kg_N Kg_Na Kg_P Kg_N*Kg_P Alternating(Kg_N*Kg_P*Kg_Na) Kg_Mg / ddfm=kr2;
 repeated yr;
run;

outp1v2.png

Although individual treatments were nonreplicated, there is enough factorial replication among nutrient composites to isolate and test the elements. Although there was no random component (such as block) in this example, PROC MIXED can certainly facilitate random and repeated statements too.

In terms of the p-values, most nutrients were highly significant.  It goes to show how 70 years of fertilization with the same thing produced a strong effect on the land and the crop.

 

Learn more about PROC MIXED

 

And consider taking this excellent Mixed Model eLearning Course

 

Fisher, R. A. (1921). "Studies in Crop Variation. I. An examination of the yield of dressed grain from Broadbalk". Journal of Agricultural Science. 11 (2): 107–135. doi:10.1017/S0021859600003750. hdl:2440/15170.

 

Data from: Rothamsted Research (2018). “Fisher 1921 Broadbalk wheat grain yields 1852-1918”. Electronic Rothamsted Archive https://doi.org/10.23637/rbk1-data-fisher-1921-01.

Comments

Hello experienced!

I've tried everything to analyze my data. I have data from blood samples taken at 3 times (hours, i.e. T1,T2,T3), 2 blocks (rounds in time), and a double factorial (Zn levels x Zn sources), with 5 pigs in each block sampled 3 times. When I try to analyze as repeated measures in time, the following message appears in Proc Mixed: “An infinite likelihood is assumed in iteration 0 because of a nonpositive definite estimated R matrix in SAS”.
I'd really appreciate support, because I've tried everything. In Glimmix the data runs, but the degrees of freedom don't look right.

Version history
Last update:
‎06-17-2020 08:20 PM
Updated by:
Contributors

sas-innovate-2024.png

Available on demand!

Missed SAS Innovate Las Vegas? Watch all the action for free! View the keynotes, general sessions and 22 breakouts on demand.

 

Register now!

Free course: Data Literacy Essentials

Data Literacy is for all, even absolute beginners. Jump on board with this free e-learning  and boost your career prospects.

Get Started

Article Tags