As I believe Yogi Berra said (if not, it sounds like something he would say), "You can observe a lot just by watching." I think economists should put more effort into observation and data-gathering and rely less on regression techniques.
William Lewis' book, Power of Productivity is based on painstaking legwork designed to look at specific industries in different countries. Although he speaks proudly on p. xxx that "these conclusions were based on extremely complex, data-intensive 'regression' analyses standard in economics," he persuaded me with simple charts and did not put a single equation in the book. He had me at hello, as it were.
I think that Cox and Alm's classic Myths of Rich and Poor does a lot with just basic data, no fancy econometrics.
It is amazing what you can do with minimal technique but really good data. Think of the Panel Study of Income Dynamics (one of the sources used to good effect by Cox and Alm) or the Medical Expenditure Panel Survey (one of the sources used to good effect in another book that I highly recommend). In fact, speaking of my health care treatise, it cites the work of John Wennberg comparing spending levels in different Medicare regions. Wennberg gets powerful results without using sophisticated regression techniques.
A common thread in what I like is that it tends to look at an issue from a lot of angles. You tend to see that more in books than in journal articles.
If you want to avoid doing junk science, find data that allow you to tell a convincing story without having to estimate deep parameters. In grad school, I remember somebody coming up with a joke about "maximum likelihood estimators using minimum likelihood data," meaning somebody trying to use fancy technique to extract information about data that is not really up to the task.
Also, a commenter asked me to explain my issues with Steve Levitt.
My father used to say that there are some questions that are impossible to answer, such as "What caused World War I?" The problem is that there are too many plausible causes, and you cannot run a controlled experiment to sort them out. The same would go for "What caused the Great Depression?" or "What is causing Global Warming?" or "What caused the drop in crime?" (referring to Levitt's famous abortion/crime piece). Not that you cannot try to answer these questions--it is important to try. But they are the sort of issues that require wisdom, not clever little empirical tricks. Your views should be stated in sentences that end in question marks, not in exclamation points.
The other issue I have with Levitt goes to my point about looking at issues from lots of angles. Instead of taking that approach, his book looks at a whole lot of topics, and it leaves each topic behind long before I feel that I have a satisfactory understanding of the issues.