2) Presidential speeches don't matter much, either.The link is to an excellent John Sides post, well worth reading.
Two ways that presidential speeches don't matter: they don't sway entrenched public opinion about the issue at hand, and they don't change public opinion about the president. Those are important, and reporters should know both of them.
Now, that's not to say that presidential speeches don't matter at all (again, Ezra didn't say that; I'm adding stuff, not criticizing).
First, presidential speeches can't sway entrenched public opinion. But they can do some things. On issues for which there is little prior public opinion -- newly emerging issues -- it's likely that supporters of the president will strongly tend to accept the president's arguments, while opponents of the president will be likely to reject those arguments. Moreover, smaller effects such as agenda setting, framing, and priming are all possible. Presidential speeches can raise awareness of an issue; they can affect the way people talk about an issue; and they can affect how presidents are judged on an issue. One helpful concept, I think, is Richard Neustadt's idea of "teaching." Neustadt argued that, at least to some extent, presidents could help themselves by affecting expectations, so that they might avoid blame for bad consequences and get credit for good ones. It's hard to study whether that really can happen, but I think it's likely. My favorite examples are George W. Bush's attempts to sell the Afghanistan and Iraq wars. On the former, he emphasized (in his speech to Congress immediately after the September 11 attacks) that the "war on terror" would be long and difficult, and I think the American people accepted that and did not judge him harshly on his conduct of that struggle. On the other hand, the Iraq war was sold as a cakewalk, and Bush foolishly promised "Mission Accomplished" long before the war was over. Again, it's hard to prove direct effects of this sort of thing, since we don't get to rerun the Iraq war with a president who sets different expectations, but I think the evidence is at least consistent with a teaching effect.
I should add that the reason these effects happen is because they are mediated through the press and other elites. Take agenda setting. One of the ways that reporters "know" which programs a president really cares about is by treating presidential speeches as a good indication of presidential priorities. Therefore, a presidential speech about health care or Afghanistan will produce lots of news stories about those subjects (and fewer about, say, climate and Iraq), and as a result mass publics will tend to believe that health care and Afghanistan are Important Issues. Of course, speeches are not the only indication reporters use -- budgets, allocations of staff, and other markers also count -- but speeches are a pretty useful shortcut.
Also, leaving public opinion aside, Washingtonians act on the information they receive from presidential speeches, using them as cues about presidential priorities and bargaining positions. Indeed, one of the tricks of writing presidential speeches is to address multiple audiences, including both mass publics who pay little attention, and Washingtonians who are apt to carefully parse minute changes in wording (recall the efforts over the summer to deduce Obama's exact position on the public option). Once again, speeches are only one piece of evidence that Washingtonians use, but it's an easy one, so they get used.
So, it's true that presidential speeches don't have a lot of direct, immediate effect on public opinion, but it's not quite true that presidential speeches don't matter.
(Update: Title updated; glitch ate half of it!)