AES Consulting meeting on 10 Oct 2016

Experimental setup

• Hundreds of individuals
• Randomly assigned to groups of 15
• These groups were assigned to
  • one of 4 treatments, say A, B, C, D
• These treatments
  • half (A and B) stay in the same initial location
  • the other half (C and D) swap initial locations with a group from the other treatment (D and C) half-way through this stage of the experiment.
• Then the individuals are all moved to 2 new locations such that A and C are together and B and D are together.
• Finally, individuals are moved to individual locations with multiple individuals (eventually) being in the same final location.

Suggestions

There is confounding between the treatments and the second location that can not be resolved by statistics. So any statement that is made about differences between AC and BD could be due to the location rather than the treatment. Thus this design has the most information about differences between A and C and differences between B and D.

• Model the group average with the following terms
  • Fixed effects: treatment
• Model the individual with the following terms
  • Fixed effects: treatment
  • Random effects: group, final location, initial location #1, initial location #2

If these two analyses a consistent, then it may be simpler to report the former. If the two analyses are inconsistent, then it may be that final or initial locations are extremely important.