The best argument for emphasizing evidence in educational policy and practice is what happens when evidence plays no role: Practice and policy swing like a pendulum from one enthusiasm to the opposite, and then back again, but no progress is made.
This is typical in fashion, where hemlines rise and fall and ties get wider or narrower just because people get tired of the previously prevailing hemline or tie width, or whatever. Similar swings are common in any field where taste, rather than evidence, is what drives change: Art, architecture, cooking, and so on. There are also fads and fashions in evidence-based fields, such as medicine and agriculture, but in those fields a given solution ultimately wins out based on evidence, and the field ultimately moves forward, not sideways.
The education pendulum is frustrating not only because it puts our field in such evidence-free company, but also because the pendulum makes irrelevant a lot of the research that does take place. A huge proportion of research funding in education goes to evaluating government policies, for example. Because research takes time, it is quite often the case that by the time the findings appear, the policy is already gone. For example, evaluations of various components of No Child Left Behind are now appearing, long after anyone cares. Politicians always support the doing of such studies, because they need to appear to be accountable, but they rarely read the reports, as things have invariably shifted (politically) since the evaluation began.
The solution to the pendulum problem is to have a wide array of research going on at all times that is creating and evaluating promising solutions to longstanding problems, including teaching methods as well as policy options. Then both practice and policy can begin to learn from the evidence and move forward together toward a better future for children.