Popper - Falsification and a Life's Struggle

[phank]

There is nothing wrong with ''Smoking causes cancer. Collect data on 1000 people.' It's only when you say after the event - 'See 5% have cancer, therefore smoking causes cancer.' That's an ad-hoc conclusion without merit. It is not a test (in my book).

All you're saying here is, you've never taken a course in statistics and have no clue how this works. And what you've been saying in all the threads you participate in is, you are ineducable because you WILL not learn how this works.

Have you ever read the news? Now and then, they mention a poll, and invariably they give error bars - things like "results are accurate to within 3%" or "these two results are within measurement error and must be considered equal" or some such.

So let's say smoking sometimes causes cancer. How often should be considered "often enough" to draw a conclusion? Statistically, this depends on the sample size. The larger the sample (properly sampled, of course), the more significant the result. What these results are being compared with is the null hypothesis, in this case that smoking doesn't cause cancer. So OK, the idea is, you sample N people at random and determine two factors - do they smoke, and do they have cancer. You might be a bit more complex and say HAVE they smoked in the past, and if so for how long, and how long ago did they stop, etc. If your sample size is small, you really DO need to find an 80% correlation to exceed the (typical) 95% probability that it's not due to random chance. But as the number of people sampled grows, PROVIDED the sample stays random, the smaller the correlation needs to be, to be statistically significant.

Here's an illustration. Say you flip a coin 20 times, and get 11 heads and 9 tails. Is the coin biased? Probably not. The more times you flip it, the more an honest coin should converge on 50% heads and 50% tails (while, paradoxically, the absolute difference between the two will grow!). If you fip a coin a million times, you'd better be very close to 50-50, like maybe 50.001 to 49.999 percent. If, after a million flips, you have a full 5% more heads than tails, this is not an honest coin. That 5% is MUCH too large to have occurred by random chance. If you're looking at a million people and the smokers are 5% more likely to get lung cancer, you have what amounts to statistical certainty.

Some people think in different term to me. I understand that. To me it is simple. 'Y is false because I saw X' means that 'X is true because I saw X.' Some people have different concepts of black and white so the debate will always go on.

But this needs to be put in proper operational terms. And broad generalities are nearly impossible to put into those terms. A good operational definition describes what you expect to see, what you expect NOT to see, and what method you are using to determine this.

[KBertsche]

I agree with all the above.
Einstein proposes an idea and Eddington ventures out and collects data.
I don't know if you call that a test. I'd say a test has other components - criteria and conclusion. Eg - 'Smoking causes cancer. Collect data on 1000 people. Criteria - If 80% have cancer and smoke, then conclude smoking causes cancer.'

There is nothing wrong with ''Smoking causes cancer. Collect data on 1000 people.' It's only when you say after the event - 'See 5% have cancer, therefore smoking causes cancer.' That's an ad-hoc conclusion without merit. It is not a test (in my book).

As Phank said, this is a statistical question. You have to determine whether or not the 5% is statistically significant, and whether or not the correlations are causative or accidental. To do this you need to investigate the mechanisms of how smoking may cause cancer. And as Phank said, you need to understand statistics.

That's what Popper did with Eddington's summary. Eddingtion compiled data and reported back his data . Other people then wondered whether it actually settled anything. Popper was over-the-moon about the 'confirmation.'
I find that approach by Popper contradictory: Any result can confirm the theory, depending on whether I like the theory.

Eddington didn't merely compile and report data. He figured out what sort of experiment to perform, he performed a difficult experiment, and he mathematically analyzed the data.

Modern science is not merely the collection of data (a la Bacon). It also includes the language of mathematics (a la Kepler) and the elements of abstraction and conceptualization (a la Galileo). All three elements are essential for modern science.

Some people think in different term to me. I understand that. To me it is simple. 'Y is false because I saw X' means that 'X is true because I saw X.' Some people have different concepts of black and white so the debate will always go on.

This makes no sense. You are mixing apples and oranges.

In science, we say that theory Y is false because I saw observation X. We are trying to develop over-arching theories that explain how the universe works. These theories must be consistent with observations. To say observation X is true because I saw observation X is a tatutology. This tells us nothing about the principles by which the universe operates.

[pancreasman]

As usual, I suspect complete misunderstanding and mixing of the terms observation, interpretation, hypothesis and theory.

[magellan004]

As Phank said, this is a statistical question. You have to determine whether or not the 5% is statistically significant, and whether or not the correlations are causative or accidental. To do this you need to investigate the mechanisms of how smoking may cause cancer. And as Phank said, you need to understand statistics.

Please explain how Eddington making a few measurements in adverse conditions has anything whatsoever to do with 'statistics'. Perhaps you could start by telling us how many measurements of light from stars during eclipses Eddington would have needed to make to get statistically significant data.

Then we can talk about what Eddington did actually do.

* Eddington didn't merely compile and report data. He figured out what sort of experiment to perform, he performed a difficult experiment, and he mathematically analyzed the data.

Reference please. Note that in my Opening Post I specifically referred to two articles. Maybe those articles were wrong. In which case Popper and Peter Coles are wrong. (And you are right. They are wrong.)

* This makes no sense. You are mixing apples and oranges.

In science, we say that theory Y is false because I saw observation X. We are trying to develop over-arching theories that explain how the universe works. These theories must be consistent with observations. To say observation X is true because I saw observation X is a tatutology. This tells us nothing about the principles by which the universe operates.

Why is 'Theory Y is false because of an observation' not a tautology when 'Theory Y is true because of an observation' is a tautology?

If an observation can establish a true/false state then an observation can establish 1. True and 2. False.

Magellan

[KBertsche]

Please explain how Eddington making a few measurements in adverse conditions has anything whatsoever to do with 'statistics'. Perhaps you could start by telling us how many measurements of light from stars during eclipses Eddington would have needed to make to get statistically significant data.

Then we can talk about what Eddington did actually do.

My comments on statistics were in response to your "smoking causes cancer" example. Not in relation to Eddington's tests.

Why is 'Theory Y is false because of an observation' not a tautology when 'Theory Y is true because of an observation' is a tautology?

??? I have no idea how to respond to this, since it's not at all what I claimed. Please re-read what I said previously.

[magellan004]

My comments on statistics were in response to your "smoking causes cancer" example. Not in relation to Eddington's tests.


??? I have no idea how to respond to this, since it's not at all what I claimed. Please re-read what I said previously.

OK, so we don't understand each other very well.
May I suggest step by step?

Do you think that, as far a science goes, if someone reports seeing something then that observation can be used to refute an idea? I am not talking about a test or part of a test, just a simple report of an observation. By refute I mean 'Show false' .

So for example if someone hears a report of Mr smith observed a black swan in Sydney, Australia. - does that report refute the idea of All swans are White. ?

Magellan

[phank]

OK, so we don't understand each other very well.
May I suggest step by step?

Do you think that, as far a science goes, if someone reports seeing something then that observation can be used to refute an idea? I am not talking about a test or part of a test, just a simple report of an observation. By refute I mean 'Show false' .

So for example if someone hears a report of Mr smith observed a black swan in Sydney, Australia. - does that report refute the idea of All swans are White. ?

Magellan

Not incontrovertibly. Few observations are, in and of themselves, sufficient to do more than lend support to one conclusion or another. At the cutting edge of science, where theory is lacking and tools are inappropriate, nearly all observations are tenuous at best, almost always incomplete, and not fully understood. This inherent ambiguity is what makes scientific discovery fun and exciting. Generally, refutation is more than a matter of observations, and also consists of the advances of theory, in light of plenty of experiments, lots of predictions, lots of testing. Eventually, wrong ideas fall by the wayside because they become incompatible with too many other observations and the advancement of theory into which wrong ideas just don't fit.

So, what's your point?

(And alas, reports of the ivory-billed woodpecker don't seem to have been verified, despite a lot of effort. They're probably gone forever.)

[magellan004]

Not incontrovertibly. Few observations are, in and of themselves, sufficient to do more than lend support to one conclusion or another. At the cutting edge of science, where theory is lacking and tools are inappropriate, nearly all observations are tenuous at best, almost always incomplete, and not fully understood. This inherent ambiguity is what makes scientific discovery fun and exciting. Generally, refutation is more than a matter of observations, and also consists of the advances of theory, in light of plenty of experiments, lots of predictions, lots of testing. Eventually, wrong ideas fall by the wayside because they become incompatible with too many other observations and the advancement of theory into which wrong ideas just don't fit.

So, what's your point?

(And alas, reports of the ivory-billed woodpecker don't seem to have been verified, despite a lot of effort. They're probably gone forever.)

Well, we agee. That is one thing. It is good to arrive at agreement when we both have different 'attitudes' to many issues.

That leaves the question of what Popper meant when he uses the terms 'falsification' and 'refute'. He was a smart man. He knew the meaning of various words such as refute, false, confirm, disconfirm etc.

I have alwasy said that if people understand that Falsification does not mean 'to show false' then I have no problem with Falsifiability.
What has happened in the past is that some people actually do think Falsification is about 'showing false'. Why not just dump the misleading term 'Falsifiability' and use 'Able to be tested'?

Those people haven't thought through the matter. You have , so that's good.

Magellan

[phank]

Well, we agee. That is one thing. It is good to arrive at agreement when we both have different 'attitudes' to many issues.

I doubt it, somehow, but...

That leaves the question of what Popper meant when he uses the terms 'falsification' and 'refute'. He was a smart man. He knew the meaning of various words such as refute, false, confirm, disconfirm etc.

Hasn't this been covered? He was referring to a proposition that is capable of being tested, and capable of failing a valid test. Now, consider your single, unconfirmed report from some anonymous source far far away, that a black swan was sighted. I equated this with an area of new scientific discovery - hazy, unreliable but tantalizing indications of something pending better observation. Popper would say that the existence of a black swan is capable of being tested.

Does your single unconfirmable report constitute a test? This is a key issue, because of course science rests on the notion that nothing can be known absolutely, and the goal is to gradually banish doubt with stronger support. So if a dozen respected ornithologists were to capture and investigate the black swan in detail, make the bird available to anyone else who wished to do so, publish their methods and their findings for peer review, etc. THEN how trustworthy is the report of a black swan?

I have alwasy said that if people understand that Falsification does not mean 'to show false' then I have no problem with Falsifiability.

I just hope you are not confusing something being falsifiable, with something being false! All scientific theories are forever hostage to tomorrow's new observations, and all are therefore forever subject to improvement, extension, modification, replacement, etc.

What has happened in the past is that some people actually do think Falsification is about 'showing false'. Why not just dump the misleading term 'Falsifiability' and use 'Able to be tested'?

I think this is what Popper did, after criticism much like yours to his earlier ideas. He moved very much in the direction of "testable". The whole idea of the null hypothesis is to avoid assuming conclusions no findings can dislodge. So a researcher starts by assuming his ideas are wrong (the null hypothesis says his prediction is false) , and that means the prediction can't be the default if the results are ambiguous.

[magellan004]

'What has happened in the past is that some people actually do think Falsification is about 'showing false'. Why not just dump the misleading term 'Falsifiability' and use 'Able to be tested'? '

I think this is what Popper did, after criticism much like yours to his earlier ideas. He moved very much in the direction of "testable". The whole idea of the null hypothesis is to avoid assuming conclusions no findings can dislodge. So a researcher starts by assuming his ideas are wrong (the null hypothesis says his prediction is false) , and that means the prediction can't be the default if the results are ambiguous.

Excellent.
Any reference for this change of mind for Popper?
You'd have to agree that in 1963, at the age of 61, he still had this black and white notion of Falsification.

Anyway - the thing is, and I know it irritates you, but we do agree.

Magellan

[shunyadragon]

'What has happened in the past is that some people actually do think Falsification is about 'showing false'. Why not just dump the misleading term 'Falsifiability' and use 'Able to be tested'? '

I'll bite. I think your splitting frog hairs here with terminology, but let's us 'Able to be tested.' The Theory of Evolution is 'Able to be tested.' The biblical description of Creation is NOT'Able to be tested.'