We are developing the social individualist meta-context for the future. From the very serious to the extremely frivolous... lets see what is on the mind of the Samizdata people.

Samizdata, derived from Samizdat /n. - a system of clandestine publication of banned literature in the USSR [Russ.,= self-publishing house]

Samizdata quote of the day

The overwhelming majority of theories are rejected because they contain bad explanations, not because they fail experimental tests.

– David Deutsch

16 comments to Samizdata quote of the day

  • dearieme

    I suspect that a bad explanation means one that is inconsistent with general laws, which are accepted as such because they have survived previous confrontations with experiment. So perhaps what he’s saying is that most confrontations with experiment are second-hand. I dare say.

  • How does David Deutsch know anything about the sum total of theories? Let alone the statistics of success or failure thereof?

    Does he have a representative sample?

    [It would help, perhaps, to have a reference to the fuller context of his statement.]

    Best regards

  • Pa Annoyed

    Context is available here. The quote appears towards the end.

  • Nick M

    I thought I’d heard of David Deutsch. Just checked him out and I was right – he is a physicist.

    I agree with him. A similar argument is made by Steven Weinberg at the start of Dreams of a Final Theory. Weinberg gives the example of two postulated cures for scrofula. One is chicken soup, the other the laying on of the hands of a king. He contends that the first potential cure is potentially worth exploring (chicken soup contains a wide variety of complex biochemicals) whereas the second is bunk (there being no particular physiological difference between monarchy and the rest of us).

    The chicken soup theory is a “sensible” one because, if it works, there is the possibility of a scientific explanation as to how it works.

    There is a hell of a difference between correlation and causality and if you don’t even have a hope of finding the latter no amount of the former will advance knowledge one iota. It may be the case that the Denver Broncos alway win at home if my left elbow itches but this remains merely a curious historical fact until a theory which properly explains why is produced.

    As Dirac put it “never trust an experiment until it is backed up by theory”. I paraphrase – but that’s the gist.

  • J

    “There is a hell of a difference between correlation and causality and if you don’t even have a hope of finding the latter no amount of the former will advance knowledge one iota”

    Arrggh!! Noooooo! I couldn’t disagree more 🙂 We have no idea why placebos make people better, but knowing that they do is a rather major advance in knowledge, wouldn’t you say?

    The world of quack science is full of plausible sounding explanations that just so happen NOT to correlate with reality. I admit it’s also full of people citing correlations and then making unreasonable extrapolations or assumptions.

    A very large percentage of modern medicine is entirely based on correlations. We have little idea how many drugs works, even decades after they have been licensed for use. But they are licensed precisely because they can be *observed* to work, and not because someone thinks they can explain *how* they work.

    The aim is to observe truths about the world, and posit general laws that explain the truths. Not to come up with plausible sounding laws and then try to make observations that support them.

    The recent mess over the Lancet Iraq mortality survey has been a fine example of this – people dismissing observations about the world because they don’t seem to fit with general laws.

    When science stops being routed in observation and starts being about sitting in an armchair coming up with rational explanations, it’ll be all over.

  • Were scrofula in fact psychosomatic, then the laying on of the hands could well be effective for those who believe the king is special. Perfectly ration explanation even if the belief itself is not.

  • Paul Marks

    The “study” in the Lancet had nothing to do with “observing the world” – a group of anti American fanatics (some of whom are American – as so many anti American fanatics are) produced a report that said what they decided it should say long before their “study”.

    The Lancet (whose editor is a man who believes that move against Saddam was motivated by an American Imperalist desire for power and conquest) has published “studies” like this before.

    First sanctions were supposed to have killed vast number of people, and now the invasion (the alternative to carrying on with sanctions) is supposed to have killed vast numbers of people. As for the people who really have been killed in Iraq – they have mostly been killed by the very “resistance” the Lancet supports.

    The Lancet is (of course) part of the same medical-social services-academic establisment that demands that medical students study sociology, declares that “inequalty is health issue” and produces statist “solutions” for this inequalty for which it fasely claims the authority of science.

    As for the posting itself.

    Economics (as Ludwig Von Mises shows in such works as “Human Action”) is a matter of reasoning, “empirical evidence” can be found, for example, to suggest that minimum wage laws increase unemployment and that they do not increase unemployment (indeed that raising the minimum wage level reduces unemployment – a New Jersey study claimed to show just this). In economics “studies” can be used to show many things AND THE EXACT OPPOSITE OF THESE THINGS to be the truth.

    It is only through hard reasoning (in accordance with the laws of logic and human action) that an economic theory can be judged – piling up “empirical evidence” is useless (as it can be done on both sides). This is not easy and a mistake in reasoning is always possible (which is why even the most careful thinker needs to expose his ideas to critical examination), but it is the only way to advance understanding in economics.

    In the physical sciences the experimental method (rather than the logical deductive method) is normally considered correct (indeed this method has become known as “the scientific method” – meaning the method of the natural, rather than the human, sciences – especially physics).

    However, as many thinkers (Karl Popper and so many others) have pointed out, of a theory does not make sense (if it is not an explination of WHY somthing happens) then it is not a scientific theory – indeed a correlation is not causation (there is such a thing as coincidence in the world). [Milton Friedman’s ultra positivist writings on the question of scientific theory (1953, if my memory serves) are profoundly wrong.]

    And even if X really has been caused by Y it is only a scientific theory if there is some effort to explain why this is so.

    Many efforts at explination will be wrong (indeed one can never tell for certain that a scientific theory is a full and correct explination of what one observes in the natural world – one can just say that it has not been refuted yet). And one of the ways to weeding out a bad hypothesis is to spot an internal contradiction in it.

    Even in physics it does not make much sense (at least not normally) to spend money on experiments if a theory contradicts itself. Although (of course) there may be some truth even in a confused and contradictory “theory” (the quote marks are because a theory that does not make sense, that has a basic logical contradiction in its very structure, is not strictly speaking a scientfic theory at all).

    Even the ravings of a madman, where each few words contradict the last few, may contain an element of truth (and truth that no one else has seen). So I hesitate to say that a contradictory “theory” should never be investigated. But resources are limited, so it would seem to make sense to concentrate on theories that do not contradict themselves.

  • Nick M

    J, Triticale,

    I don’t consider many aspects of medicine to be scientific. The fact that it might work is neither here nor there. But do bear in mind that a lot of medical procedures had been standard treatments in the past for a long time (and therefore must have “worked” to a certain degree) which are now seen as being hopeless if not counter-productive. A prime example is the standard battlefield treatment for a gunshot wound in the early-modern period – bleeding.

    I was going to say more on correlation and causality but Mr Marks did it better so I won’t bother. Paul’s comments on economics amused me. I think the lack of a consensus as to basic concepts amongst researchers is the reason why the “social sciences” have never made the advances that the physical sciences have. Certainly that’s the argument that Tom Kuhn makes in The Structure of Scientific Revolutions. For a particle physicist or an astrophysicist “energy” means the same thing. A classical economist and a Marxist one probably have very different ideas about correspondingly “basic” concepts in economics. (They also, of course, have essentially political axes to grind). Alas, my knowledge of economics is too poor to furnish a concrete example.

  • Midwesterner

    The aim is to observe truths about the world, and posit general laws that explain the truths. Not to come up with plausible sounding laws and then try to make observations that support them.

    J, my son. (daughter?) There is a third way. That is to posit plausible (or not) sounding laws and then see if they predict the known data. If they do not, discard them.

    Since the data contains the ultimate truth and the theories are mental constructs, room for the unknown must be given to the data. To allow for unknown data, one conditions on the theory and throws it out if it doesn’t predict the known data. Conditioning on the data (perceived reality) denies full reality the opportunity to be something more than what we already perceive.

    I realize this epistemology is very unpopular with ‘enlightened’ folk, but if one would place reality over our perceptions, then our theories must reach beyond present perception.

  • Andreas

    It is hard to guess what the author of the post wanted to say with the qoutation. But the statement quoted is not hard to believe since there are always infinitely many possible explanations conforming with any certain set of data. Some of these explanations (theories) will be thought up in the scientific process and most of them will be rejected by parsimony. (So conforming with data alone is obviously not a good truth marker for a theory).

    Midwesterner: What you are actually saying is that (logical) positivism is an anti-realism, incompatible with objectivity. And I believe that you are right.

  • guy herbert


    a man who believes that move against Saddam was motivated by an American Imperalist desire for power and conquest

    A theory that is not obviously untrue, even if you couch it in sneering language. Setting out to topple dominoes in the region by displacing the alienated tyrant holding a strategic keystone state with a friendly regime is an imperialist plan of conquest. And as such was approved by most writers here, because many people believed: that it was legitimate as defensive grand-strategy and – mistakenly – that it would be liberalising in its effects on the Arab world, and would be executed somewhat in accordance with those goals.

    The WMD excuse was transparent tosh: a causus belli as adventitious as Captain Jenkins’s ear. An d those under he impression Saddam was removed because he was a very very bad man need to explain why it is that those who removed him couldn’t care less about Robert Mugabe, and find criticism of the Uzbek regime unreasonable, considering it is such a friendly, enlightened, provider of interrogation transcripts.

  • guy herbert

    On the main point, what Midwesterner said. And never forget William of Ockham.

    It is significant that Deutsch is a physicist. Quite a lot of theoretical physics escaped the possibility of being called to explain the real world some time ago, and has become like theology with equations.

  • Concerning correlation and causation, there is more to be said.

    Firstly, as is continually repeated, correlation does not imply causation.

    If A and B are correlated, that information gives us no knowledge on whether A causes B, or whether B causes A. Either or neither might be true (and even both, to some extent, through feedback mechanisms).

    Another possibility is that the observed correlation is a matter of chance, especially with small statistical samples. However, statistics does give us various tests as to how likely this is: generally called “significance tests”.

    Where chance is not thought to be the explanation, of course we have to accept that the correlation is present. Then there will be causation, though it might not be from A to B or from B to A. So it might be better to seek refinement of that old “correlation does not imply causation”.

    The final possibility, for causation, is that both A and B are caused by some common precursor, and thus there is causation, but not directly linking from A to B or vice versa. It is ‘common causation’.

    The classic and often quoted case of correlation without causation is that of the increase in crime following the introduction of television. The classic refutation is that crime is positively correlated with the number of fridges too. As much as some people may want to associate increasing crime with TV watching, those people do not want to associate said increase in crime with fridge ownership; they are (most likely) against television, not fridges (nor even, perhaps, crime). So they seek further information.

    Watching TV is not the cause of the increase in crime (or rather it is neither the primary cause, nor the sole cause). Both TVs and crime have increased over time; passage of time is the common cause, or the linking of other ‘common causation’ into this particular correlation. However, there is nothing to support the view that TV ownership and crime rates will continue to rise with the passage of time. [Of course, there are other intervening factors in this particular case, concerning level of economic activity, extent of knowledge of the amount of crime, technological advances (which affect both of the former and have other causative effects), changes in law enforcement and societal views of itself, etc.]

    A big problem is correlation of the observed effects with time, in the same sense (ie both positive or both negative) with some continuity over time that is coincidental, and may not continue forever. This sort of causation is problematic, and perhaps I’m wrong in labelling it as “causation”. In many ways, the effect is a statistical one of measuring the correlation over too short a time period, in which low-pass filtering effects (present in many physical phenomena and their measurement) is interfering with a true measurement of the correlation.

    The next problem with correlation and causality is the belief (currently highly prevalent concerning Anthropogenic Global Warming: AGW) that if there is both correlation and a valid theory of causation, then the extent of causation is either total, or that desired by the proponent’s own case. This is irrespective of other (partial and perhaps dominating causes) that have not been properly investigated (or are carefully omitted from highly partisan and irrational rhetoric).

    Thus, the existence of causal mechanism (in addition to correlation) does not, necessarily, justify the attribution of total (or dominant) causation. Nor does the “quality” (ie extent) of a single correlation, alone, give evidence of the extent of causation. There are, however, statistical techniques to apportion the share of correlation (and possibly causation) between various known factors; for example Analysis of Variance and extensions thereof (which are so very necessary for many real-world problems). By including, in this sort of analysis, all known possible causations, plus time, a somewhat more useful analysis is obtained; furthermore, unknown or unused possible causes with time correlation get shoved into the “time” category in the analysis. And, of course, if you don’t have the data on known possible causations (eg actual solar irradiance, pre-1978, in the global warming contention), one’s analysis is b*****ed. [That is to the extent that (including the period over which) the data is missing.]

    Best regards

  • Paul Marks

    No Guy – there was no desire of imperialist conquest. Mr Bush and Mr Blair really do mean what they say about spreading democracy when they can (it might be better if they did not). As for dealing with Comrade Bob – that would be rather difficult when one considers the attitude of South Africa. North Korea would be an even more difficult problem (given that it is right next to the People’s Republic of China).

    Of course you could reply (and you would be right) “what about Syria and IRAN being right next to Iraq” – yes this was not really considered (apart from vague hopes that liberal democracy in Iraq would spread to Syria and Iran)

    As for W.M.D. – actually some old chemical stuff was found (only Fox bothered to report it – even though several Senators had a news conference about it) a lot of stuff was taken to Syria (due to the super slow preperations for the campaign). And plenty of documents (which even the New York Times does not dispute) were found showing that the atomic weapons project was ongoing.

    So the “tosh” was not tosh.

    Of course this does not mean that the judgement to in was the right one, at the time I though it was wrong one. But what is done is done – now it is a matter of winning the war (if this is possible).

    As for Jenkins ear. Like you I wish that Sir Robert Walpole had won that political struggle (which he might have done had he still had the strength he had some years before) and had not been forced into war (it was a classic example of the government of the day being bounced by the House of Commons into a war it did not want).

    If the Iraq war had been a similar case then Mr Bush and Mr Blair would have been OPPOSED to war – and forced into it by the Congress and Parliament.

    “Our aircraft are being fired upon in the north and south no-fly-zones and the Right Honourable Gentleman just sits there and does nothing”. Of course, without the no-fly-zones the Kurds and some southern groups would have ben exterminated (the Shia were hit very hard in 1991 after they revolted in response to George Herbert Walker Bush’s appeal – they will never forgive us for calling on them to revolt and then allowing many thousands of them to be killed).

    Or (in the American case) “The President has no more backbone than a chocolate eclair” – as you will remember, the President being bounced (by Congress, the press and others) into the war with Spain in 1898.

    However, as I type this I remember that there IS an argument on your side Guy.

    In 2000 President Bush campaigned on a being LESS interventionist than President Clinton – it was only after September 11th 2001 that more interventionist forces got his ear.

    I can remember a time when the “Weekly Standard” (and the rest of them) did not like George Walker Bush at all (their man was John McCain).

    So, actually, you may be correct on that point Guy.

  • Experiments only address a specific hypothesis, not some overall unified theory. BIG difference in scale here.

    I know of the null hypothesis, but not the null theory.

    Theory really is a step removed from experiment and observation.

    I’ve always seen the outliers on the scatterplot as being as true as the statistically significant clusters. But I’m a divergent thinker and refuse to be cowed by statistics. I know too much to be fooled very easily.

  • Midwesterner


    I drifted away for a bit and I’m not sure I understand the context or intended point of your first statement re experiments.

    I also think outliers are significant. For one thing, they suggest the need for a larger sample size. The second data point is always an outlier unless it is identical to the first. And if a claim is made that they represent sampling error, then that should be should itself be demonstrable and reproducable. No?