The middle class and "Screwflation"....

[Infohawk]

You definitely didn't read the study then. You appear to have glanced over their main section on their findings (that did include political predictions), and completely missed the section where they talked about possible confounds and the methods they took to control them, which is where I quoted from. You will notice they ran the regressions both with, and without political predictions, and still found a statistically significant effect. Before you start LOLing, you should probably know what you're talking about.

What do you mean it doesn't prove it? I don't even know what you're trying to argue. In your previous post you said it appeared that Republicans were being penalized for predicting their side would win. The authors of the study say that even when political predictions are removed, the statistical significance remained. Are you claiming they are lying? Are you seriously looking to personally go through every prediction they analyzed to see if they meet your personal standard for what constitutes a political prediction? Also for your 'apples to oranges' comparison, are you attempting to argue that people are more likely to be correct on economic issues than political ones? What are you basing this idea on?

This is a measure of pundits in general, not of political ones or of economic ones. Oddly enough, you read the part where they specifically mentioned his economic credentials probably helped him, and then drew somehow a negative conclusion about the study from that. (this is baffling) As to how he compares to other economists, he outperformed the Chairman of the Federal Reserve.

Even moreso, who cares? People said Krugman was wrong a lot. Some people looked at his predictions over a year period and found during this time he was right a whole lot more than he was wrong. Since we don't seem to have any other actual studies, just people hand waving, it seems relevant.

Yes I do want to see their data and how they interpreted it? Is that too much to ask? Especially since in some cases there were only 17 predictions. That is how social science would normally be done. The data is more important than the analysis. The fact that you can't understand that tells me it's not worth discussing this further. You can keep thinking this study somehow proves Krugman is a good source of authority...

 

[nonlnear]

Nobody's trying to gain some sort of huge universal truth from that thing, and their scoring system was fine.

Yes, it's fine to have such a scale. It's not fine to waste so many words explaining it. This paper was written by morons for morons. (Yes, they actually do say "subtracting by three", and it's not a typo.)

They gave one point for every correct prediction, subtracted a point for every wrong prediction, and gave no points either way for wishy-washy hedged ones. How would you have improved upon it?

By having a clue how to test for predictive accuracy instead of blindly calculating statistics that give absolutely no support for the assertions I'm making:

The first analysis tested correlation with a correct prediction. This was done by generating a new variable, PredABS, that was the absolute value of the difference between the prediction's truth (PredTrue) and the predicted outcome (PredProb). This PredABS could be between 0 (guessed correctly) and 4 (guessed the absolute opposite of what really occurred). Because a 0 represents a correct guess, variables with negative coefficients improve a prediction's accuracy.

An R-squared value of .156 was obtained for the regression. This means that about 16 percent of the total variance in prediction outcomes is explained by the variables in our equation.

Okay that's enough laughter for one sitting. (And it's annoying typing the quotes out manually because they use a "clever" font obfuscation trick to make copy-paste inconvenient.)

First, the way to test for the predictive power of a set of variables is to withhold a portion of your data to test your fitted model. Then again, they used such a pittance of data to begin with it would almost be a waste to use proper methodology after making the first grievous error.

Second, an R-squared value says absolutely nothing about how much of the variance is explained by the variables in the equation. To be fair, in one strict (completely non-causal) sense it kind of does, but they are clearly asserting a causal relationship when they say "explained by" so they don't get a pass on this one. There is a nasty problem called overfitting (which one runs an especially high risk of when one's data set is of insufficient size), which is why one generally withholds a portion of the data for testing, but...

Seriously, this whole paper's "analysis" is complete shit. It's barely enough to snow an incompetent public policy prof. A good policy prof would fail this paper.

Either way, who gives a shit?

You do. You posted this article to broadcast to the interwebz how much you care about this very point.

The only reason it was mentioned was to specifically rebut that people said Paul Krugman is often wrong. In at least this year period that was analyzed, he was often right. That's the beginning and end of why this was even brought up.

If you don't understand why the "study" (it's really quite a shame to sully the term by using it here) is worth less than toilet paper then I guess nobody is going to dissuade you from believing that it's an effective rebuttal.

Now none of what I've posted here is meant to say anything about whether the assertions of the study are actually true or not. It may well be true that more "liberal" (as defined by the study) prognosticators are more accurate, but this study does absolutely nothing to bolster such a claim. In fact, it reads like a Fox News snow job. Even a stopped clock...

 

[fskimospy]

Right. I've said that repeatedly. The only reason it was brought up to begin with was to show a record of Krugman being accurate. Whether or not you agree with their regressions is pretty meaningless (as I've also said repeatedly), because the absolute only reason it was brought up was to show that he was almost always right.

Krugmans predictions are data points, your complaints about their other methodology are arguing against something nobody is arguing for, so you're wasting your time. (Unless you believe they incorrectly interpreted what should be considered correct.) The only time I even mentioned the papers analysis was to correct infohawk when he obviously hadn't read it.

I see this happen a lot on here, where some data is brought up, people raise moronic objections like infohawks, and then when those are corrected its taken as some sort of endorsement of the entire product. I have no idea why.