11-13-2016 09:34 PM
I finished an experiment with two factors: 1) fertilization level and 2) With bacteria or without it.
Fact. A (fertilization) = 100%,70%,30% y 0% (four levels)
Fact. B (Bacteria) = Yes / No (two levels)
I have seven treatmentes, because I only have 100% without the bacteria. I took 4 measures duringin a month, each measure uses 6 plants per treatment.
T1: Withbacteria / 0%fertilization (6 plants/week) x (four weeks)
T2: NObacteria / 0% fert. (6 plants/week) x (four weeks)
T3: Withbacteria / 30% fert (6 plants/week) x (four weeks)
T4: NObacteria / 30% fert (6 plants/week) x (four weeks)
T5: Withbacteria / 70% fert. (6 plants/week) x (four weeks)
T6: NObacteria / 70% fert (6 plants/week) x (four weeks)
T7: NObacteria / 100% fert. (6 plants/week) x (four weeks)
I need help with my design and codes. Thank you.
11-30-2016 10:45 AM - edited 11-30-2016 10:51 AM
If you have access to Milliken and Johnson's Analysis of Messy Data, vol. 1, you could see their approach (termed a "means model") where they fit the response as the highest level interaction, and then create tests and contrasts to fit the research questions. A google search on "means model" turned up a lot of hits.
. Here is a link to an example, which would cover your whole plot analysis:
Now, you have the additional design element of a repeated measures part, so you would need to port the code to PROC MIXED to correctly accommodate the covariance between the repeat measures. Rather than ESTIMATE statments, LSMESTIMATE statements might be more useful.
I would suggest getting a whole plot analysis, and then sharing that code. From there, we could craft PROC MIXED code to correctly handle the repeated part.