Clarify via unit tests model identifiability issues

[larryshamalama]

Because of non-identifiability, it would be then important to investigate situations where we can better identify the treatment effect will converge to its true value. This can either be done using a restrictive data-generating procedure or using informative priors. This can be done in another PR

Originally posted by @larryshamalama in #43 (comment)