It always results in higher bias.

It typically leads to larger standard errors due to the design effect.

It is always more expensive to implement.

It requires a complete list of the population.