...no, no, no, no, no.
Krugman has a post up plotting change in government spending against change in real GDP and getting a positive relationship. The only sentence I really agree with is "it's a bit striking, isn't it". Yes, it is. It's first-step towards proof that should encourage people to look at the question more.
But this is obviously very bad empirical reasoning, and the worst part is - Krugman knows that. Indeed he lists several reasons not to take this scatterplot seriously. I'll add another reason - take a look at the scattered points and ignore Poland, Latvia, and Ireland (you can even keep Greece in if you want). Do you see a clear story anymore? I don't.
And none of that is to say that I disagree with Krugman on austerity. You all know I agree. But it's this kind of low-quality latching onto whatever buttresses your point attitude that gets sophomoric libertarians yammering on about the Romer-Bernstein chart. This is exactly the wrong way to approach macroeconomic empirics.
I'm not saying you always need a fancy model when you discuss these things. Tell a compelling narrative that makes your case about a compelling counterfactual or justification. Krugman is often quite good this, and that approach is not only sound - it's well suited to a blog. Point being - identify your damn models, people. You aren't doing your viewpoint any favors by defending it with sloppy evidence.