Table 1. WAIC score differences relative to the score for model M3.

Model	Effects	Δ WAIC
M0	random	531
M1	random + control + experimental (time only)	336
M2	random + control + experimental + pairwise interactions	33
M3	random + control + experimental + pairwise interactions + 3-way interaction	0

Model M3 has the lowest WAIC score and is therefore preferred for out-of-sample prediction—the standard for model quality measured by WAIC. Larger WAIC score differences indicate less preferred models relative to M3. The models increase in complexity reading down the table.