By observation and testing.

We actually use the scientific method every day in some simple ways. Let's say your phone isn't working. What do you do? You try to find out why it isn't working. Is it plugged in? let's test that idea -- yes it is plugged in, so that isn't the fault. Is there a fault on this line? Let's plug in another phone which is working and see if it works here. It doesn't -- OK, so it's probably a fault on the line. Next step, contact the phone company.

You're making observations, formulating hypotheses, and testing them. What works better -- this, or asking someone if they think they know what's wrong with your phone? At some point, someone is going to have to investigate the physical evidence.

Some types of science lend themselves better to anecdotal evidence than others. Sociology, psychology, anthropology, etc are some of these. Other types are better studied by the methods above. I think it's only logical to keep in mind that people make mistakes and aren't always trustworthy. Ever read Herodotus? He's a cracking good read but he never intended to be completely accurate in his historical accounts. In those days, historians were expected to beef things up to be entertaining. Historians always have biases of some kind, just like journalists do, because this is the nature of the human being. Again, it would be illogical to discount this.