In many ways I think that a consulting relationship is similar to that which might occur between someone who designs his or her own house and then hires a builder to build it. There are many pitfalls that can menace such an undertaking, and I believe that many of these same pitfalls can cause problems in a statistical consulting situation. The following are some of the things which have struck me as being important.
1. It helps for both people to understand the problem as a whole. While it is entirely possible for a builder to follow the designer's requests board by board, this is most likely not the most effective way. The builder has many tools and techniques at his disposal that the designer may not know about. It is entire possible that the builder can use some of these different techniques to create a better end result, but this is only true if the builder has an understanding of what final result is desired. Similarly, the statistician has many statistical techniques that can be applied to various problems. In many cases one problem can be approached a number of different ways. Choosing the best method may not be something that the researcher is able to do, given the level of statistical training he or she has had. It is the responsibility of the consultant to ask questions and try to make sure the ultimate purpose of the data analysis is understood. I like to think in terms of "What questions do you want to answer?"
2. It's a good idea to be in communication from the very beginning of the project to the very end. If a designer designs the house and then buys materials all before talking to the builder, he may discover that the materials he bought will not work for the house he wants. He is then stuck with all of the material and is forced to go out and get all new materials. If the builder starts to build and fails to keep in touch with the designer he may find at the end of the project that there has been a fundamental problem in his work like he was holding the drawing upside down and built the house facing the wrong direction. Similar things can happen between a statistician and a researcher. If the researcher fails to consider the statistical methods that are available and what is required to apply those methods, he may find that he has a stack of worthless data that took years to collect. Likewise, if the statistician fails to communicate what he is doing to the researcher, he may find that when all is said and done, he has provided answers to questions that are meaningless to the researcher.
3. If a designer has designed a house in order to sell it later, it is important that the designer be able to talk about what went into the house. The best way for this to happen is to have the builder show the designer what is going on in the building process at each step of the way, possibly having the designer help in the building. Similarly the ability to justify the statistical conclusions is also important for a researcher. It is the responsibility of the statistician to try and make sure that the researcher understands at least the basic idea of the statistical analyses that are being used. It would also probably be a good idea for the researcher to be aware of any particular strengths and weaknesses in the analyses as well.
The three points I have mentioned can all probably be put under the heading of "good communication." The most important thing (other than having a sufficient knowledge of statistics) the consulting statistician needs to have is an ability and/or desire to make sure that effective communication happens throughout the course of the consulting relationship.