How should we interpret the evidence about head start