My answer is likely biased by the fact that we work with Hierarchical Bayesian (HB) Analytics so frequently (mainly using choice data such as that created by conjoint). After all, HB requires a starting hypothesis. But the reality is that even if we don't use HB, a hypothesis is a useful thing.
First, understanding what our clients EXPECT to find is a great way to understand what they NEED to find. They need to validate or reject their prior thinking so the more we understand their thought processes the more we know where to focus. In addition, this understanding often leads to insight into their firm's business decision making. This helps us to present results that tell a story that resonates with them. This is true even if the findings contradict their thinking.
Second, by presenting results in this way we help our clients to do more than meet the objectives of the current study, but to walk away with a better understanding of what to expect in the future. Flaws in logic will help them to avoid those flaws when similar issues come up.
Of course purists will point to the risk that starting with a hypothesis may bias our results. We might be inclined to design our research and reporting to match the narrative we expected to find. We might also be tempted to avoid the "kill the messenger" problem by sugar coating the truth.
These are fair points and well worth guarding against. They do not, however, undercut the premise that having a starting hypothesis makes for better market research and likely better use of results.
 |
 |
 |