AES Consulting meeting on 21 March 2017

Split-plot design

Consider Figure 2A where the whole plot is mouse and randomly receives drug (light or dark), and a subplot that is the different tissue (A or B) in the mouse.

QUESTION: When the split-plot is not randomized, can we call this a split-plot design?

Comments

• It is not clear there is a standard definition of split-plot design that requires the subplot to be randomized.
• When analyzing data collected in this manner a split-plot design seems reasonable, although we may be concerned about a negative correlation between the two samples within a mouse.
• We could also calculate differences within each mouse and analyze the differences as a completely randomized design.

Hills Criteria

We also discussed non-statistical methods for determining cause-and-effect relationships. Specifically, we discussed Hills Criteria for causation and its application to Zika.