Help on Analysis: Repeated measures? Split-Plot? Something else?
May 13, 2009 9:33 AM(1150 views)
I have a set of data that seemed straight forward, but now I am sure how to proceed. The experiment consisted of a single factor (fertilizer) with four treatments (fertilizer levels) and eight replicates (plants) of each treatment. Each replicate then had 5 data points collected on it (ie, 5 leaves from each plant were measured for area at a single time point). I am not sure what the proper way to handle this data is. Do I consider this a repeated measures, since each replicate had multiple measurements? Can I consider the repeat observations as subplots and run a split-plot analysis? Or can I just average the data points for each replicate, and run an ANOVA?
You could use custom designer and set up a 2 factor categorical design. One at 4 levels (fertilizer amount) and the other at 8 levels 9 (plant type). I assume that you have used the same fertilizer and just different amounts of that fertilizer. After you generate the 32 treatment combinations you could then recode your fertilizer level by modifying the column properties as a continuous factor and code it accordingly and input the low and high levels. The responses for the plant measurements could be added as individual columns, Y1 through Y5 and a sixth column that represents the average of the 5 measurements could be calculated. In addition an additional column response of standard deviation could be generated. You could then model the average growth and standard deviation back to the continuous fertilizer amount and the 8 categorical plant types. Hope this helps.
Other than your four fertilizer levels, this is a standard nested design. You have 8 plants nested within fertilizer level, and then five different leaves on each plant nested within plant. This is not repeated measures (as you are not repeating the measurement on any particular sampling unit), and this is not a split plot experiment. Nor should you analyze this as a crossed factorial design (you could set it up as a crossed factorial design, 4x8x5, just don't analyze it that way). Both plants and leaves are random effects.
Following up on your response, do you know if it is possible to include random effects (like in this fertilizer/plant/leave study), but use the Generalized Linear Model personality rather than Standard Least Squares? Or even to use the Stepwise personality? I really am looking for AICc values, but can't get those in the Standard Least Squares personality, which is the only way that seems to accept random effects!