What about this kind of bias: every experimental result can be attributed to some kind of bias. And bias is a hot topic, especially given the ascendence of progressive politics in academia and beyond. So there is a strong motivation for experimenters to explain any experimental result in terms of bias, rather than other reasons.
This article is a good summary of the issues of the (mis)use of statistics in science. Having done some research myself, in most cases the statistics don't scream a particular message to you, and it's really hard to understand the data. Even without any external pressures, statistical tools like p-values have limits. If you run 10 different models (all a priori reasonable) and 8 of them seem to say roughly the same thing, does that mean your result is correct?
I don't think that's a fair characterization of the issue. p-hacking is done almost exclusively by people who completely understand the definition and interpretation of p-values. However these people are also under a lot of pressure to produce positive results (and sometimes, results in a specific direction) and this biases their thinking.
The problem is not that people are not aware of the issue of testing multiple hypotheses. The problem is that (1) it's hard to say exactly what your hypothesis is before you've even looked at the data, and (2) it's hard to determine if people are choosing parameters for p-hacking or simply making choices based on their best judgement.
>Furthermore, p-values were designed to deal with experimental data. If you're doing an observational study, perhaps you should use statistical tools designed for that purpose.
This is simply wrong. p-values are equally relevant in both cases. E.g. I can use p-values to reject the hypothesis that consuming saturated fats is uncorrelated with weight gain, amongst the general population. It sounds like you are reaching beyond your actual expertise in statistics.
I'm think disagreement is not about privacy per se, but rather the absolute moral correctness of our legal principals. E.g. other people in this article have commented that people who have affairs deserve to be punched in the face (which is also illegal).
I think engineers tend to place higher values in legal principals, because (a) we tend to prefer rule based solutions and (b) the tendency to allow exceptions to the law also results in school bullying, which nerds have a greater experience with.
People who have affairs arguably deserve to be punched in the face by their spouses (figuratively speaking; I don't advocate actual violence), not by Joe Public.
Open market operations can't really be characterized as lending to banks. Instead, the Fed buys various kinds of debt, especially US govt bonds, thus injecting money into the economy. The fed funds rate is really the rate at which banks lend to each other. Banks are privileged in that they have a special legal mandate to act as banks, but they are not privileged in terms of access to credit (they can borrow at the "discount window" but this is less important than the Fed's open market operations).
But more importantly, QE is more like injecting money into "real things" since it is buying corporate debt that presumably funds real projects. So the (alleged) failure of QE is not very good evidence for your claim that standard monetary policy is bad.
EDIT: and they main reason I and most economists prefer US style monetary policy is that it's very neutral: you have a lever, and that lever is how much bonds you buy. You can choose various flavors of bonds, and various maturities, but they are all fundamentally the same. In contrast, the government directly funding real projects lends itself to corruption and favoritism.
> You can choose various flavors of bonds, and various maturities, but they are all fundamentally the same.
It appears to me that you've just equated a US Treasury bond with a private label subprime RMBS. Is that a fair assessment? If so, how strongly do you feel about that equivalence? I see the private label RMBS's as less of a 'real thing'.
In the context of that sentence, I was referring to the bonds bought by open market operations, which are primarily US government bonds and would not include private label subprime RMBSs.
The kind of close-to-risk-free bonds bought in open market operations are all fundamentally the same because they simply move money from one point in time to another.
Isn't a treasury bond primarily funding social security, medicare, and defense spending (the three biggest US budget non-discretionary spending categories)?
While I agree that that both corruption and favoritism have negative effects on the investment environment and civil society. Wouldn't the money wasted in corruption and favoritism drive inflation even if only in the direct cost of the projects themselves.
The article looked like a load of waffle, but it would be nice to see some basic statistical tests applied, e.g. can we reject the hypothesis that everyone is equally accurate?
One of the main mechanisms by which inflation stimulates the economy, is that it lowers wages (because of sticky wages) hence decreasing unemployment. So intentionally raising wages as a form of stimulus seems counterproductive. Could you explain your argument more?
One of the ideas floating around is that the economy is currently limited on the demand side, and that increased wages will result in more disposable income and higher consumption.
Increasing wages will make employers less inclined to hire people, and thus potentially less total disposable income. Giving ordinary people more money directly seems like a more reasonable sort of stimulus. Where are these ideas floating around? The idea of sticky wages dates back to Keynes.
I tend to side with the argument that employers ultimately hire not because people are cheap, but because they need them to meet demand. If there's no demand, there's no wage low enough to justify a hire. I don't want to speak for Thomas Piketty, but my reading of his data is that generally economic growth is the product of higher wages, not the other way around.
If anyone would like to be "radicalized" in a slightly different political direction, I would suggest visiting the real Bethlehem. Then try to reconcile the experience of the people living there with the politics of writers for the New Yorker (on both US and Israeli politics).
PCA implicitly uses a quadratic form on a vector space. When done the usual way, this is just the identity matrix. But for image data, I think you would get better results using a quadratic form based on a kernel, so that Q_ijkl where (i,j) and (k,l) are pixel coordinates, is given by Q_ijkl = t((i-k)^2 + (j,l)^2) for some kernel function t.
