Quick Thoughts on “Pure Reason” vs. Empiricism
I am going to be brief, but there is still much confusion on this grand philosophical dispute, which started when Brad DeLong, as is his wont, said innocently enough about a renowned philosopher: “Thomas Nagel is not smarter than we are–in fact, he seems to me to be distinctly dumber than anybody who is running even an eight-bit virtual David Hume on his wetware.”
I’m not going to recapitulate the whole controversy. But at least some people misunderstood what Steve Landsburg was doing in his response to DeLong. So let me spell out just that subset of the dispute:
(1) In his critique of Nagel, Brad DeLong didn’t merely say, “I find Nagel’s particular argument for the powers of pure reasoning to be a bad example, or to be improperly conducted.” No, he actually said:
Thus Thomas Nagel’s insistence that we need a theory of consciousness that accounts for our reason’s ability to become an instrument of transcendence that grasps objective reality–that insistence falls apart like an undercooked blancmange…Any theory that provided such an account of reason becoming an instrument of transcendence and offering guarantees of grasping objective reality would be hopelessly, terribly, laughably wrong….
And I cannot help but think that only a philosophy professor would believe that our reason gives us direct access to reality. Physicists who encounter quantum mechanics think very differently… [Bold added.]
(2) Steve Landsburg took the above passages to mean that DeLong was claiming “that pure reason can never be a source of knowledge,” and Steve explicitly said that perhaps that wasn’t DeLong’s claim. (However, looking at the block quote above, where DeLong is saying we need to run things by the physicists first, he sure does seem to believe that there’s no such thing as knowledge about objective reality, that you can access via pure reason.) So, if that were indeed DeLong’s claim, Steve refutes him thus, with a list of facts about objective reality that you can only obtain via pure reasoning:
1) The ratio of the circumference of a (euclidean) circle to its radius is greater than 6.28 but less than 6.29.
2) Every natural number can be uniquely factored into primes.
3) Every natural number is the sum of four squares.
4) Zorn’s Lemma is equivalent to the Axiom of Choice (given the other axioms of Zermelo-Frankel set theory).
5) The realization of a normally distributed random variable has probability greater than .95, but less than .96, of lying within two standard deviations of the mean.
…and so on.
Those who are familiar with the methodological works of the Austrian School will recognize that we are here butting up against Immanuel Kant’s categories of knowledge. (Here’s a link that lays out the basics, though I haven’t read it carefully so maybe the writer here is sloppy. But upon a quick glance it looks OK.)
Kant’s system had a two-fold distinction, between analytic/synthetic and a priori/a posteriori. Here are my dumbed-down paraphrases, which a real philosopher in the Kantian tradition may not like: “Analytic” means you can determine the truth or falsity just by analyzing the terms and their definitions. “Synthetic” means the truth value corresponds to something about objective reality, something “out there” and not just coming from human conventions in terminology. “A priori” means you don’t need in your proof to appeal to experience. “A posteriori” means your argument for the bit of knowledge does make an appeal to past sensory experience. Here’s a quick example of how *I* (not necessarily all philosophers and presumably not Brad DeLong) would populate the categories:
Synthetic A Posteriori statements: The sun is hot. (True.) Brad DeLong is very civil. (False.)
Analytic A Posteriori statements: [Empty set. I thought I came up with an example when I was teaching at Hillsdale but David Gordon vetoed it.]
Analytic A Priori statements: A bachelor has no wife. (True.) A bachelor has a wife. (False.)
Synthetic A Priori statements: Every natural number is the sum of four squares. (True, since I trust Landsburg.) All true statements in arithmetic can be proved within an axiomatic system. (False, and I’m sure Landsburg will bite my head off for saying this ungrammatically.)
In Human Action, Mises said economic principles were a priori. However, I don’t think he actually took a stand on whether they were synthetic or analytic. Hans Hoppe in this essay (which I think is really powerful) said they were synthetic, and that Mises’ action axiom solved the mind-body problem. (The guy was productive, what can we say. That’s why they named an Institute after him.)
What is “pure reason”?
Just thinking about the issue. You don’t need any empirical evidence to decide the matter.
No. I’m serious. In case I didn’t make it obvious, I am objecting to the term itself. Reason, as I understand it, is the application of logic to the perceptual concretes provided by our senses or abstractions from concepts thus formed. Perceptual concretes have to be… well….perceived.
I just don’t see how the term “pure reason” has any meaning at all and, further, how this 4-fold categorisation makes any sense either. So do you accept or reject the analytic-synthetic dichotomy?
“Perceptual concretes have to be… well….perceived. ”
This is not relevant to the nature of logic. “Pure reason” is such that you take a statement and can analyze its truth value without taking measurements or making observations.
“and, further, how this 4-fold categorisation makes any sense either.”
An analytic statement is true by definition. A synthetic statement is true because of the nature of things, but the predicate is not implied in the subject.
An a priori statement does not require verification by checking with measurements or observations. An a posteriori statement requires verification – its truth is contingent about something empirical.
“An analytic statement is true by definition. A synthetic statement is true because of the nature of things, but the predicate is not implied in the subject.”
OK. The challenge starts!! 🙂
Proposition 1: Man has 2 eyes
Analytic or synthetic?
Proposition 2: Man is rational
Analytic or synthetic?
Another question on your statement. You said
“An analytic statement is true by definition. A synthetic statement is true because of the nature of things, but the predicate is not implied in the subject.”
So are you saying that the definition has nothing to do with the nature of a “thing” (material or otherwise)?
This article is helpful (if perhaps loaded …):
Science: Philosophy’s Handmaiden
http://www.str.org/site/News2?page=NewsArticle&id=5285
Thanks. Not that I ever disagreed with the contents, but this was definitely a good read. A very nice way of putting it.
Sorry I missed this. You said
“Just thinking about the issue.”
How did “the issue” get into your mind for you to think about it? Prior to any experience at all? Don’t think so.
“How did ‘the issue’ get into your mind for you to think about it? Prior to any experience at all? Don’t think so.”
A priori statements can be proved (or refuted) without appeal to empirical evidence. This does not mean that someone who has never experienced any sensory data can know them; e.g., you might need experience to understand some a priori statements. (I think Hoppe mentions this in A Theory of Socialism and Capitalism.)
You’re missing the point.
Yes, you have to experience something in order to gain knowledge of facts, but then analysis of those facts can, depending on the issue, be done completely in your head.
To be sure, the capacity for learning must also exist along with your physical self, in order for you to learn; Your body may send signals to your brain, but then something that is not deterministic must be in place in order to analyze those signals.
My evidence.
http://en.wikipedia.org/wiki/Victor_of_Aveyron
Hm, does the output of my internal dialogue count as empirical evidence?
I think “pure reason” in this context means things that are *conceptually* deduced. We may need to see many real-world instances (examples of how various numbers are written as the sum of four squares) before we can grasp the *idea* of the theorem, but we can’t determine its *truth* in this way.
It’s a good idea to contrast this with scientific research, where once a property is verified on lots of instances, it becomes elevated to “law.” If we verified a mathematical proposition on lots of instances, we’d similarly elevate it to a *conjecture*, but it would not be a *theorem* until we had a *proof*. The proof would rely solely on properties of the *class* of natural numbers. [Obviously, the properties of this class would rely on how we defined the set of natural numbers, which is why lots of people think all a priori truth is analytic].
Interestingly, a disproof (showing an instance where it does not hold) is very similar to hypothesis invalidation in scientific research (constructing an experiment in which some law does not hold).
Thanks everyone. So are we effectively saying that “pure reason” refers to deductive reasoning? But then how can deductive reasoning not tell us anything about reality when the subject matter that deductive reasoning deals with is concepts formed inductively or through deductive reasoning applied to other more fundamental concepts formed inductively?
When we say something to the effect of “deductive reasoning does not tell us anything about reality”, aren’t we in effect saying that in reality, it is possible that A is non-A?
Just asking these questions because I see people doing this rather commonly (I see shades in the extracts Murphy has shown).
“Physicists who encounter quantum mechanics think very differently”
This sort of thing always bugged me. The physicists interpretation of what the statistical nature of QM *means* is…. well, pure reason. And, in my opinion, wrong. But there is no empirical way to determine whether, for instance, the resolution of statistical probabilities of events to determined phenomenon is a result of multiple universes, or – as I hold – actually always determined at a level we cannot examine without screwing up (I mean, really, this seems the most obvious thing to me – that QM is deterministic as is classical mechanics, but there is no way to measure in this manner as you are measuring with something you don’t know itself. Like trying to measure the position of baseballs with other baseballs…. that ain’t gonna work well.
1) DeLong “Did Not Do the Research” – he appealed to Hume as an authority, as if Hume had the last word on the question, and no philosophers (or anyone else) have grappled with the question and found Hume to be not quite right.
2) DeLong appeals to the idea that the mind is just the wetwork – well, as Ed Feser pointed out in several places, materialists have been shoving phenomenon under the carpet of the mind that they could not explain through eliminative materialism, and it will be impossible for them to do that with the mind itself. (He explains this much better than I can in a comment, so i recommend checking out what Feser has said on this).
3) DeLong himself in his snarky, dismissive, but not very thoughtful post says “Physicists who encounter quantum mechanics think very differently – thinking being, of course, an action, and presumably DeLong believes Physicists could think very differently if they learned new facts. Therefore their thoughts on the matter are not mechanically determined, and so on.
4) the scientific method does not generate itself, empiricism does not generate itself. As in the mathmatic examples Landsberg used, the scientific method was formulated through reasoning about how to inquire about the world, and updated with new reasoning, successively. The scientific method itself depends upon “pre-scientific” axioms and principles, which were not discoverable through science itself, but formulated as a necessary precondition of doing empirical, scientific research in the first place.
Therefore we must conclude that Brad DeLong “is not smarter than we are–in fact, he seems to me to be distinctly dumber than anybody who is running even an eight-bit virtual Nagel or Mises on his wetware.”
1) The ratio of the circumference of a (euclidean) circle to its radius is greater than 6.28 but less than 6.29.
=============
Hmmm. How about a piece of string and a ruler?
Or ‘Euclidean.’ I mean the claim is that we can learn something about the real world, synthetic a priori. But this looks like we have ourselves an analytic truth, dependent on definitions doesn’t it?
Hmmm. How about a piece of string and a ruler?
And an infinite amount of time, that’s another necessary input for your plan. Landsburg can use pure reason to prove that for every conceivable Euclidean circle his statement is true.
So, draw me a perfect circle.
I blame Base 10 for the difficulty of drawing a perfect circle.
“In Human Action, Mises said economic principles were a priori. However, I don’t think he actually took a stand on whether they were synthetic or analytic.”
Mises:
“Every theorem of praxeology is deduced by logical reasoning from the category of action. It partakes of the apodictic certainty provided by logical reasoning that starts from an a priori category. Into the chain of praxeological reasoning the praxeologist introduces certain assumptions concerning the conditions of the environment in which an action takes place. Then he tries to find out how these special conditions affect the result to which his reasoning must lead. The question whether or not the real conditions of the external world correspond to these assumptions is to be answered by experience. But if the answer is in the affirmative, all the conclusions drawn by logically correct praxeological reasoning strictly describe what is going on in reality” (Mises, L. 1978 [1962]. The Ultimate Foundation of Economic Science: An Essay on Method (2nd edn, Sheed Andrews & McMeel, Kansas City. p. 44).
Cleary even Mises himself admitted that synthetic propositions as auxiliary hypotheses would sometimes enter into his praxeological deductive reasoning. If such assumptions do not correspond to the “real conditions of the external world,” then his inferences are unsound and untrue.
George J. Schuller long ago identified the problem with Human Action:
“Acceptance of Mises’ stated axioms does not necessarily imply acceptance of the “principles” or “applications to reality” which he has drawn from them, even though his logic may be impeccable. When a logical chain grows beyond the limits set by stated assumptions, it uses unstated assumptions. The number of unstated assumptions (axioms, postulates, or other) in Human Action is enormous. If Mises denies this, let him try to rewrite his book as a set of numbered axioms, postulates, and syllogistic inferences using, say, Russell’s Principia or, closer home, Von Neumann’s Theory of Games as a model” (Schuller, G. J. 1951. “Mises’ ‘Human Action’: Rejoinder,” American Economic Review 41.1: 185–190 at p. 188).
Unstated assumptions are present in every economics treatises ever written.
Let those authors rewrite all their treatises into 10,000 page long logical syllogisms.
Should be a hoot.
“Cleary even Mises himself admitted that synthetic propositions as auxiliary hypotheses would sometimes enter into his praxeological deductive reasoning.”
Assumptions of axioms are not synthetic propositions.
“If such assumptions do not correspond to the “real conditions of the external world,” then his inferences are unsound and untrue.”
I don’t see what bearing this has on the synthetic-analytic positions. The only thing you are saying here is that if the axioms are not true, then the logical implications of them are not true, either. Duh.
“George J. Schuller long ago identified the problem with Human Action: ‘The number of unstated assumptions (axioms, postulates, or other) in Human Action is enormous.'”
Like? It’s been a few months since I read Human Action, but I recall that every time an assumption was made, Mises clearly stated what he was assuming, and why. Thus you have an explanation for why he brings in the disutility of labor, and a detailed argument for the use of the ERE as a mental tool and thought experiment, or the use of ceteris paribus clauses.
What Schuller has done is to make a claim about logical errors in Human Action – unstated and potentially false assumptions – and then shift the burden of proof to Mises. He only discredits himself with such fallacious and erroneous reasoning. If there are such errors in Mises’ writing, then they should be pointed out and refuted – not merely alluded to.
What you just quoted is essentially, in academic prose, the equivalent of “your argument is wrong, and if you deny this, prove it by rewriting your argument with a formal system of logic point-by-point.”
Point is that axioms are always not-true. That is to say, they never fully and exactly capture the real world.
Of course, the empiricists can’t do any better because their measurements can never fully and exactly measure the world either. In economics the measurements are sloppier than most attempts to apply empiricism and the subject is more complex.
“Point is that axioms are always not-true. That is to say, they never fully and exactly capture the real world.”
The action axiom stands as a testament to how wrong you are here. Unless you meant that any set of axioms will not explain any and all facts about reality, which is fairly obvious and inconsequential.
I must admit that I haven’t read “Human Action”, so probably I should do, but the Mises “action axiom” is unfortunately a bit poorly defined.
If you refer to any goal-seeking behaviour then it is not unique to humans. Cybernetics has covered much more detailed and useful studies of goal seeking behaviour than economics (and that’s from both a theoretical and empirical standpoint).
Most Austrian economists don’t appreciate the cybernetic perspective on goal seeking because they want something that implies conscious decisions made in a human-like manner… but humans act for both conscious and unconscious reasons, and psychologists are only scratching the surface of how it works. No one can build a conscious machine, so no one really knows what goes into one.
Humans sometimes base decisions on prejudice, or on ideology, or because they are confused, or have been lied to, or just happen to be having a bad day and blundered, or in some cases out of genuine stupidity. What’s worse, a given human won’t necessarily do the same thing under the same circumstances a second time!
What sort of axiom is that?
At best the “action axiom” implies that some sort of decisions are being made for some purpose, and humans do it. That’s way to much hand waving for my liking. We might choose to define an archetype “rational economic man” but that’s just a model again, not detailed, and not real.
People say you cannot deny the action axiom because to deny it is itself an action, but I can just as easily program my computer to deny it, or teach my parrot to deny it. Then someone would deny the axiom, but not by demonstrating a purposeful decision.
With a bit of help from the Board of Education I could teach a whole generation of human kids to deny it as well, and most of them wouldn’t even stop to question why they are denying it. Proof of human action, or proof of obedience?
These are examples of what Murphy is talking about:
Logic Problems
http://www.puzzles.com/Projects/LogicProblemsArchive.html
DeLong may not be as correct as he thinks. I’ve noticed physicists often don’t ask the right questions and that makes them susceptible to errors in judgement and renders some theories erroneous.
If you examine some theories closely and ask the right questions,you’ll discover those theories break down. Likewise Keynesians make errors because they don’t ask the right questions and if you examine their theories closely they also break down. Many of the “right questions” in physics are related to fundamental principle of physics.
If there are such errors in Mises’ writing, then they should be pointed out and refuted – not merely alluded to.
This is all that Keynesians and assorted anti-Austrians can ever do. Mises starts at the very beginning, humans and their actions, and develops economics from that obvious starting point. He starts with “human act purposefully” because he explains that we cannot read peoples’ minds. We really do not know what motivates them and the only real evidence we have of the economic wants and desires of the billions of strangers on earth are the terms of their exchanges, prices.
On the other hand, the Keynesian Hoax begins in the middle of the story in the middle of the depression. As Daniel “Strangelove” Kuehn has shown, the 1920 depression was caused by price distortions that had resulted from the Fed’s underwriting of WWI. Others have shown that the Great Depression was caused by further credit expansion in the 1920s plus British wages that were too high as a result of attempting to repair the distortions that began with WWI.
Keynes and the Keynesians ignored the entire Misean analysis. However, these days Keynesians can do nothing other than take potshots at the philosophical terms applied to Mises’ basic insights and thus allude to alleged but unexamined “errors”. The insights themselves are never analyzed or refuted. The central concept of economic [mis]calculation is never engaged.
BTW, the other day Tom Hickey, MMT blogger stated that the Austrian analysis of problems caused by distorted prices had been rejected by most economists as too simple [really?? When? Where?]. He did not further explain. But as I recall, economic statistics themselves are comprised of aggregates of (surprise!!!) prices. So, where is this other source of information that the simple Austrians have failed to identify?
“MMT blogger stated that the Austrian analysis of problems caused by distorted prices …
You mean “distorted” away from the mythical, non-existent price vector that supposedly clears all markets?
So, you’re challenging the laws of supply and demand? No wonder I rarely ever read your commentary.
Natural numbers (despite their friendly sounding name) don’t appear in nature.
Sure you can have two apples, but since no two apples are exactly alike, you don’t have two apples, you have those particular two apples. If I take one apple away, then it is not the same to take the left apple or the right apple.
For that matter, Euclidean geometry doesn’t exist in nature either.
Abstract reasoning is fine for abstract systems. Generally if you want to represent a real system, you need an abstract model to work with, in which case you are pretty much stuck with the implications of whatever abstract methodology you choose for that (or invent something new if you can). It’s a model though, it will never be real. Maybe good enough, but never the real thing.
Is there no two of anything exactly alike?
Is there no two of anything exactly alike?
Is the fact that your comment is positioned slightly lower than my comment, sufficient to making it different and thus not an example of two somethings exactly alike?
Or are they indeed exactly alike because “exactly alike” excludes location, and maybe other predicates as well?
>For that matter, Euclidean geometry doesn’t exist in nature either.
I suppose all that engineering I learned was false. Tear all the buildings down, guys. Triangles apparently don’t actually exist like we thought they did.
“Sure you can have two apples, but since no two apples are exactly alike, you don’t have two apples, you have those particular two apples.”
Sure, but two protons are exactly alike. Two hard drives are virtually identical, if of the same size, brand, and capacity. Two apples might even be so similar their differences are inconsequential. An inch is the same as another inch. A pound, the same as another pound. And so on.
Different protons have different internal energies, do they not?
What is the meaning of “another inch”, that distinguishes it from “an inch”?
The proton by itself (i.e. H+) has velocity, thus kinetic energy. I seem to remember that the velocity can be any continuous value, not a discrete value.
If the proton is in a nucleus with other protons, there’s another energy for the protons in the nucleus, and I think they are still arguing over that one, but beyond anything I can calculate. They are discrete levels because one some combinations are stable.
For that matter, suppose you have an atom that is not stable, but might be good for an hour or two. So from a natural number point of view, do we count that, or do we wait a little while and count the bits that it breaks down into?
It you were taught that buildings are made out of triangles, then I’d argue you were taught wrong. Maybe a useful simplification, and possibly a reasonable place to start learning, but triangles don’t have melting points, triangles don’t have welded corners, triangles don’t have hairline cracks running across the bar… etc.
Triangles are a lot easier to type into your FEM modelling software, and still get a plausible set of equations… but that isn’t a building.
As for two protons being exactly alike, well that is the theory as I have heard it, but protons exist at a position in space, and they don’t share the same position, so how do you intend to use natural numbers to describe the position that the proton sits at? Or its velocity? Or mass?
I guess you could argue that natural numbers exist in chemistry where you are counting the number of protons and electrons in each atom, but even then they are not fully discrete concepts because the electrons in a metal don’t belong to any atom in particular, they just belong to the metal.
You could argue that the mass of a piece of matter must be equivalent to just adding up the number of each type of particle in that block, but it still would not be EXACTLY right because there’s a tiny mass in the bond energy, and in the magnetic field (if the block happens to be magnetic) etc.
Tel,
“Nature” here means “as in the nature of things that exist”.
“Sure you can have two apples, but since no two apples are exactly alike, you don’t have two apples, you have those particular two apples.”
But how do you even get to say “apple” without the concept “one”? And how do you form a concept “apple” to use in your proposition without the concept of measurement omission? Didn’t you, at the time of forming the concept “apple” choose to ignore the particular measurements that distinguish individual referents carrying the label “apple” while including those measurements that are necessary to distinguish them from all other existents as instances of the concept “apple”?
In a philosophical debate, the first person to mention quantum mechanics loses.
+1
+1 is the spin of a photon.
You lost the debate.
Photons aren’t quantum mechanics.
You lost.
Oh, but a spin is…
I guess I lost…
🙂
What? I thought pointing to people who made up their philosophy on the spot as whatever sounded good at the time to help them explain what they think they see was the best way to understand philosophy. Just like the best way to understand astronomy is to ask a 12th century cleric!
Bob,
This is from the link you gave.
“Kant defines a priori knowledge as that kind of knowledge which is held independently of all experience”
How can there be any “knowledge” independent of all experience? Would such a thing actually be called “knowledge”? It raises the question “What is knowledge”?
“He further qualifies a priori with the adjective pure when the proposition in question contains no empirical elements”
How can there be a proposition with no empirical elements when every word denotes a concept that is either a concept (formed inductively) of something that exists or an abstraction thereof? And if there are words with no referents in the real world, then isn’t the proposition itself a flight of fancy with no grounding in reality?
Bala wrote:
And if there are words with no referents in the real world…
Bala, you are assuming that the “real world” consists entirely of matter. Yes, if that’s how you think of reality, then there are no synthetic a priori statements. But if instead you agree that mathematics exists, that prime numbers actually are, then there are synthetic a priori statements.
“Bala, you are assuming that the “real world” consists entirely of matter.”
No. I am not. I am just saying that something exists of which I am aware. I am saying that reason enables me to form concepts. Any proposition necessarily works with such concepts.
“Yes, if that’s how you think of reality, then there are no synthetic a priori statements.”
I am just saying that this entire classification makes no sense.
“But if instead you agree that mathematics exists,”
Mathematics is a set of abstractions. Abstractions do not exist except as a set of concepts we hold in our minds.
“that prime numbers actually are”
“numbers” are an abstraction and “prime” is an attribute of these abstractions.
Bob,
Sorry to bother you with more from the article you linked, but here’s what it says.
“As examples, Kant suggests, “all bodies are extended” as an analytic judgment, and “all bodies are heavy” as a synthetic judgment.”
Doesn’t the very concept “body” include existing with some mass? If it didn’t have any mass, would it be called a body in the first place? And doesn’t “heavy” by definition mean “existing with some mass”? Therefore does the concept “body” not include “existing with some mass” as an attribute? Doesn’t the proposition just identify a property of the concept “body”? How does this become “synthetic” while “all bodies are extended” is “analytic”? What is it about “extended” that makes it an attribute of the concept “body” and what is it about “heavy” that tells us that it is not an attribute of the concept “body”?
Bala I don’t know what Kant meant by the term “body” so I don’t want to get hip deep defending his examples. I am prepared to argue for the examples I chose in my original post.
Fair enough. I am just trying to say that time spent using and studying a Kantian framework is time wasted.
Bala,
Kant explains this in the introduction of his Critique of Pure Reason as well as his Transcendental Aesthetic. So you could check that out to get an answer.
I think it is a bit uncharitable to say that studying a Kantian Framework is time wasted given the simple fact that Kant is considered the most influential philosopher ever, whether a person agrees or disagrees with Kantian philosophy. Also, his thought has influenced many of the most important thinkers of the twentieth century; two of which are Ludwig Wittgenstein and Donald Davidson.
i think those mathematical examples are a priori analytic. they dont tell us anything about the universe. isnt space actually non euclidean?
I am curious how Mises solved the mind-body problem; given that it seems as though his metaphysic is one that is grounded in naturalism, but moves to argue for dualism in his epistemology. That seems problematic in itself, and does not seem to get us closer to solving the mind-body problem.
Did you follow the link?
Why is the text for all comments on this post in italics?
Sustain, to find before most people, {Actually, a couple of modest pal is actually anyone {which|that|whom|exactly who|who seem to|just who|what individuals|so. vines eventually reduce for it.
outlet http://dqoreen8.hatenablog.com/entry/2014/03/15/235329