Consider a regularized growth rate estimator

[eric-czech]

Currently, growth rates are calculated as the linear regression log(cell_count + 1) ~ a + b * time, but this is a very high variance estimate when the sample count is low. A prior or regularization of some kind on the coefficients would make global analysis simpler.

It may also be worth using pooled/hierarchical models across apartments/chips, but that would likely depend on publication plans (otherwise, a slightly more biased single apartment growth rate estimator would probably suffice).