A problem with this question is that there are a number of possible confounds that have to be controlled for. For example were the babies who were not crying asleep or not asleep when the test is done? You have to control for these. You also have to operationalize your DV. Is it crying/not crying at a certain point? How long did they cry?...some other measure? The test would be a lot easier if they were crying when the rocking started and then you tested if they stoped.
One way to test this, after you control for sleeping somehow, would be to test a DV of cried/not cried which measured if they cried at some point in the process. This would be logistic regression. Than you would test rocking or not rocking influenced this ratio.