Maximizing the expected log-likelihood given the observed data and current parameter estimate.

Calculating the conditional expectation of the missing data.

Updating the likelihood function itself.

Testing the hypothesis $H_0$.