Likely it would be the first two options. A "valid inference" in this case is essentially a conclusion you come to, a decision you make about a large population based on a small sample.
In order to have a sample that would produce a valid inference, you need it to be completely random so that it really represents the whole population. Calling the first 50 names in the phone book isn't very random—you would get all "A" last names, and probably some people that are related to each other. That wouldn't be random, and it doesn't represent everyone very accurately.
I hope this helps.
