Proposal:Scientific measurements should have 'Uncertainty bars';
Scientific measurements often have error bars which gives reviewers an indication of the accuracy of the instruments used.
But I also think that 'uncertainty bars' would be useful, and make 'science' more honest and accountable than it is at present.
What do I mean by 'uncertainty bars' ?
As an example, I'll use radiometric dating:
Lets say a sample of basalt is 'dated' by radiometric dating to be 1 billion yrs old by the K-Ar system. Let's say the error bars are 1%.
Biblical creationists often point out that this result relies on unproveable and untestable assumptions, one of which is that radioactive decay has been constant over all this time. The problem is that we've only been measuring the rate of decay for this system for less than a hundred years. Because the decay has been constant over our time of measurement (well, almost anyway!), that certainly doesn't mean that it has been constant for one bill. years! - especially when only God knows what causes the strong and weak nuclear forces to exist.
The problem boils down to this - they are extrapolating outside of a measured dataset. Now I would be the first to admit that science would be useless if extrapolations outside of datasets were not allowed, but any honest person should also acknowledge that the further out from the measured dataset that an extrapolation is stretched, the less chance of it being true there is. Why? There can be unknown effects take over, or even sudden changes.
(Some things in nature have been observed to change suddenly, other things have been changed slowly by previously unforseen effects For sudden changes, think of an earthquake. For gradual changes, think of how quantum machanics takes over from Newtonian mechanics as the scale becomes sub-atomic: or how Einstienian special relativity takes over at speeds approaching c, or how General relativity will take over and cause a black hole through stretching of space).
I propose that the above measurement should be reported as "1 bill. years old with error bars of 1%, and uncertainty bars of 99.99999%". The uncertainty bars are calculated very simply by the ratio of the size of the extrapolation outside the measured dataset (in this case 1 billion less 100 years) to the extrapolation plus the measured dataset (in this case 1 billion years).
i.e. "percent uncertainty" = 999,999,900 yrs/ 1,000,000,000 rys * 100 = 99.99999%
I believe this would make science much more accountable, since very large extrapolations outside the dataset would have this reporting obligation, so thay they could not hide behind a figure that sounds certain to lay people.
What do you all think?
Views: 95
Replies to This Discussion
-
Permalink Reply by Colin Newton on -
I can see your point but I don't think your proposal is workable. It comes pretty close to saying that since we can't prove that decay rates have been constant - and also there are other assumptions about the past involved - the uncertainty is so great that it's not worth making the measurement.
-
Permalink Reply by Donald Smith on -
Different radio metric dating methods give different ages. They should just publish all the dates obtained, old or younger and let everyone make up their own mind if its a valid system. Discrepancies in the dating methods has been pointed out by creationist and evolutionist, but these views don't seem get published to the greater general public via the mainstream media corporations, therefore giving an impression that there are no discrepancies with these systems.
The theory of evolution can't tolerate a young date, because the illusion that it could perform the necessary step by step changes becomes obvious that it couldn't have happened that way.
-
Permalink Reply by Ralph Thurlow on
-
The point is they can still make measurements - just have to admit how uncertain it is.
In some cases it could still be worth while since it is the only guesstimate we have access to. A shaky guesstimate may be better than none at all.
Concerning ages for the earth, anyone who believes that Genesis is historically credible, would say "So what!" to a gurestimate that has a 99.9999% uncertainty.
I personally think it would be just science being honest about it's inherent uncertainties - an aspect of science that lay people never get to look at.
Colin Newton said:I can see your point but I don't think your proposal is workable. It comes pretty close to saying that since we can't prove that decay rates have been constant - and also there are other assumptions about the past involved - the uncertainty is so great that it's not worth making the measurement.
-
Permalink Reply by Ralph Thurlow on
-
Quite right Donald. There are actually numerous loopholes for misleading and dishonest conduct available to scientists.
Donald Smith said:Different radio metric dating methods give different ages. They should just publish all the dates obtained, old or younger and let everyone make up their own mind if its a valid system. Discrepancies in the dating methods has been pointed out by creationist and evolutionist, but these views don't seem get published to the greater general public via the mainstream media corporations, therefore giving an impression that there are no discrepancies with these systems.
The theory of evolution can't tolerate a young date, because the illusion that it could perform the necessary step by step changes becomes obvious that it couldn't have happened that way.
-
Permalink Reply by Richard A. Meiss on
-
Sometimes I wonder if members of this forum have ever read a scientific paper at all. It is standard practice, and has been for over 75 years, to include an objective measure of uncertainty ("error bars" in the parlance of this discussion) with any published scientific work with a quantitative component.
Computation of confidence limits is an exact science, one supported by the well-developed mathematics and science of statistics. When "error bars" are used in scientific papers, they may take the form of the range, or the standard deviation, or the standard error of the mean, or the 95% (or 99%) confidence interval, or a number of other measures of dispersion. In each case, there is a specific technical meaning to the reported measurement of dispersion, and interpretation of the data requires a knowledge of the way in which the calculation was made and the assumptions on which it rests. The proposal "The uncertainty bars are calculated very simply by the ratio of the size of the extrapolation outside the measured dataset ... to the extrapolation plus the measured dataset ..." is, as far as I can tell, completely without mathematical or statistical foundation.
Extrapolation of data must involve a justifiable mathematical model that describes the nature of the process involved (i.e., whether linear, quadratic, exponential, power law, etc.) and a quantitative measure of the statistical uncertainty in the data set that is the basis for the extrapolation.
It is distressing to read the implication that scientists are being deceptive in failing to report the statistical uncertainty in measured data or its extrapolation, when such reporting is actually the standard practice. The further implication that science needs to be made "more honest and accountable than it is at present." reveals a lack of knowledge and familiarity with the way science is actually done and reported.
-
Permalink Reply by Ralph Thurlow on
-
Richard,
DUH!!!
I think I can justifiably respond with "Sometimes I wonder if [some] members of this forum have ever read a post properly at all
If you'd bothered to read my original post carefully, you would have noiticed that I discussed error bars (as currently used) and pointed to the fact that they are related to the uncertainties of the instrumentation used.
WHAT IS LACKING is any systematic reporting of the ridiculous amounts of extrapolation involved in many of these methodologies. EVERY extrapolation outside a measured dataset introduces the possibility of uncertainty since we know for sure ONLY: the measured dataset - all else is inference.
You don't have to be very bright to realise that the further one extrapolates outside the dataset, the more tenuous is the inference.
For example, in relation to radioactive dating, one (e.g. probably youself - sorry for my presumption) might repsond with: "but radioactive decay rates are extremely constant". Sure that may be true for the time we have measured them, but how do you KNOW that this is true for points outside your measured dataset?
The uncertainty bars I am proposing is unrelated to the existing error bars, and gives the investigative reader an index of the extent of the extrapolation outside the measured dataset that was used to justify the result.
And if you should feel that science is doing just fine and does not need to be made "more honest and accountable than it is at present.", then why are articles like this appearing?
http://www.icr.org/article/6583/
Richard A. Meiss said:Sometimes I wonder if members of this forum have ever read a scientific paper at all. It is standard practice, and has been for over 75 years, to include an objective measure of uncertainty ("error bars" in the parlance of this discussion) with any published scientific work with a quantitative component.
Computation of confidence limits is an exact science, one supported by the well-developed mathematics and science of statistics. When "error bars" are used in scientific papers, they may take the form of the range, or the standard deviation, or the standard error of the mean, or the 95% (or 99%) confidence interval, or a number of other measures of dispersion. In each case, there is a specific technical meaning to the reported measurement of dispersion, and interpretation of the data requires a knowledge of the way in which the calculation was made and the assumptions on which it rests. The proposal "The uncertainty bars are calculated very simply by the ratio of the size of the extrapolation outside the measured dataset ... to the extrapolation plus the measured dataset ..." is, as far as I can tell, completely without mathematical or statistical foundation.
Extrapolation of data must involve a justifiable mathematical model that describes the nature of the process involved (i.e., whether linear, quadratic, exponential, power law, etc.) and a quantitative measure of the statistical uncertainty in the data set that is the basis for the extrapolation.
It is distressing to read the implication that scientists are being deceptive in failing to report the statistical uncertainty in measured data or its extrapolation, when such reporting is actually the standard practice. The further implication that science needs to be made "more honest and accountable than it is at present." reveals a lack of knowledge and familiarity with the way science is actually done and reported.
-
Permalink Reply by Richard A. Meiss on
-
Hi, Ralph -
First of all, let me apologize for what may have appeared to be intemperate language. As a scientist, I must confess that I don't react well to blanket implications that scientists are attempting to be deceptive, especially when I suspect that the accuser is not sufficiently familiar with the situation.
Yes, I am aware that uncertainties in instrumentation do need to be accounted for in reporting experimental results. But in many cases, this is only a small portion of the uncertainty that must go into an error analysis. Other factors, such as environmental influences, uncontrolled sources of variation between subjects, etc. can be considerably larger than instrumentation errors. Of course, this depends on the particular discipline, the experimental design, and so forth. But I must re-emphasize that there are well-developed mathematical tools for understanding and accounting for such uncertainty. And proper application of these tools can greatly strengthen confidence in an extrapolation.
You complain about the "ridiculous amounts of extrapolation." Exactly why are they ridiculous, and whom are you ridiculing? For those of us who are not bright enough to realize that the further the extrapolation, the greater the uncertainty, I offer the following. Extrapolation involves much more than simply drawing a line through a data set and extending it at will. Its validity is depends on statistically-based measures of the reliability of the data set and the reliability of the model upon which the extrapolation is based. Some types of data may be extrapolated far into the future (or past), while others may not support large amounts of extrapolation. Competent scientists know the difference, and routinely treat it in the Discussion sections of their papers. The problem that I have with your proposed "uncertainty bars" is that it does not take into account the nature or quality of the data set nor the type of extrapolation (i.e., the mathematical model) that is being used. And, by the way, scientists to not KNOW (your capitalization), for instance, that the rate of radioactive decay has been constant, but they can state with a high degree of confidence, based on exhaustive experimentation, that they can have a high degree of confidence in that assumption.
I could go on with this, but I want to address your comment: "And if you should feel that science is doing just fine and does not need to be made "more honest and accountable than it is at present.", then why are articles like this appearing?" I am quite familiar with that article - it took some restraint on my part to avoid responding to Mr. Thomas, but I will make my response here. As a scientist "in the trenches", I am aware that there is some dishonesty on the part of some scientists, but the actual amount is quite small when compared to the hundreds of thousands of working scientists. How can I say this with confidence? Simply because science is self-correcting. Innocent errors or intentional fraud are discovered by other scientists when they examined published papers or attempt to verify reported results. When one publishes a scientific paper, it is open to the scrutiny of everyone the field of study - and this can be a very critical and unforgiving audience. But here is where I become upset with Mr. Thomas and his assertions. The publications of "creation scientists" are full of errors: errors of fact, of reporting, of interpretation, and some downright distortions. And this persists, because "creation scientists" do not hold themselves accountable to the scientific community at large. If "creation science" were held to the same standards of scrutiny that mainstream science must endure, it would quickly wither away. I am aware of the allegation that the "scientific establishment" conspires to prevent the publication of "creation science." This ignores the fact that it is the quality of the science that is the most important factor in the publication of a scientific paper, and that this is a primary reason that creationist research does not appear in scientific journals. All working scientists, myself included, have had papers rejected by scientific journals - and once the sting of rejection is over, the paper is revised, corrected, and re-submitted. And then it can be chewed apart by other scientists. In short, mainstream scientists are accountable to the whole scientific community, and this is the primary protection against scientific fraud. Creation scientists have no objective accountability (except perhaps to God), and this is why it occupies the "outside looking in" position that it does.
So pardon my long response. It is an honest expression of my views, not an attempt to make trouble in this forum.
-
Permalink Reply by Ralph Thurlow on
-
Thanks for the nice post, Richard, but it seems to me that you (and your collegues) are not enthusiastic about my proposal simply because you have something to hide.
Consider the example I spoke of for radiometric dating (K-Ar method, 1 bill yrs calculated age. The simple fact is that if we look at the entire 1 bill yr timeline, 99.99999% of that timeline is inferred. Only 0.00001% of that timeline consists of actual measured data points. The proposed uncertainty bars simply tell what proportion of that timeline is inferred and what proportion is actually measured data, so why not let that information be clearly available to laymen???????
Because 99.99999% of that timeline is inferred, that doesn't mean that it must be wrong. But there is an awful lot of space there for unknown effects, sudden changes, etc., It could well be wrong (just as any inference can be wrong) and laymen should have the right to know exactly how much inference is involved. Inference works only SOMETIMES and it is the uncertainty concerning inference that the public should know about (and by the way, I'm not dealing at this stage with the problem of Ar in the pre-basaltic lava - we'll overlook that problem for the moment).
I believe that the science community will vigorously resist my proposal because the size of extrapolations used is an embarrassment to their exact sounding results. They are, after all in a self serving industry, no different in that respect to cornflakes companies. Sometimes the truth hurts, yet the public are entitled to know the truth and the whole truth.
-
Permalink Reply by Richard A. Meiss on
-
Hi, Ralph -
In your opening sentence, you suggest that your proposal would not be accepted because scientists (myself included) have "something to hide." If you would care to look, you will find most of my scientific papers on line (try Google Scholar). I invite you to determine just what I might have tried to hide. As I have said before, the reason that your proposal would be rejected is that it has no basis in mathematical or statistical theory, and you would be applying your method to data sets with widely different properties. In any case, any layman should be capable of recognizing when an extrapolation has taken place. The problem comes when that person lacks the background to make a critical and informed analysis of the reported scientific findings.
But lets get on with the problem of the constancy of radioactive decay rates (which is the heart of your objection to extrapolation) and the creationist claim that the rates may have changed radically in the past, How can scientists be sure that the rates have been essentially constant? First of all, there are the modern measurements of decay rates (by the way, do you have any idea how these measurements are made?) that show that environmental factors have only a miniscule effect. Secondly, there are strong theoretical reasons as to why the rates would be expected to remain constant. Anyone wishing to challenge this theoretical basis must have an alternative theory or mechanism to present, and must have data to support the challenge; i.e., the burden of proof lies with the challenger. Thirdly, the agreement of multiple methods of radioactive dating applied to the same sample (such as the Apollo moon rocks) indicates that all of the isotope systems employed have had the same constancy in their decay rates.
This whole question leaves creationists in a bind. For their chronology of the earth to be valid, as derived from Biblical "begat-counting", they must arrive at a figure of around 6,000 years. Do you wonder why scientists, for their part, have settled on the specific figure of 4.6 billion years as the age of the earth? Why not 2.7 billion, or 6.9 billion years? The answer is simple - the age of 4.6 billion years is what the analysis of the data yields. If it had been 2.7 or 6.9 billion years, scientists would not have been upset - they have no a priori reason to want a particular value. But there is a wealth of scientific support for "their exact sounding results", and the precision that has been achieved is hardly an "embarrassment."
There is another problem for creationists in this regard. If I suggested that you read a detailed book (there are several excellent sources) on radioactive dating by a working (non-creationist) geologist, you would object that the book was a fabrication designed to hide the truth from the public. So creationists are placed in the awkward position of considering that those most likely to understand the science behind radioactive dating are at the same time engaged in a conspiracy to hide the truth from the public.
Science is not easy, and reading scientific papers is not easy. It takes effort, persistence, and at least a modicum of science in one's background, and I do not fault those who follow science in the popular press (which, however, does a rather bad job of presenting science). But please be aware that what is reported third-hand is likely to lose something in the translation, and don't leap to the conclusion that scientists are engaged in some sort of dark conspiracy to delude the public.
-
Permalink Reply by Ralph Thurlow on
-
Richard Richard Richard Richard!!!!!
You've printed a lot of words, but come nowhere near adressing the issues I presented.
At the heart of the issue is inference - it may work and it may not.
Thus while it may help us on occasions, on others it may lead us down the garden path.........
....and that's where the uincertainty arises.
Let me give you two other simple examples to illustrate "uncertainty bars":-
Imagine I should look up all the data on the measurements for gravity strengh at sea level ever since it was measured.
The records go back say 500yrs.
They all agree (within error bars) to 9.8000000 m/sec^2 say.
(If the figures or the years are not in exact accord with reality, don't get your knickers in a knot....its just an example!)
The question I ask is what will the measurement for gravity be tomorrow?
Answer:9.8000000m/sec^2 (same error bars as the measuremnents up to now) AND uncertainty bars of 0.00055% (i.e.1/365*500 x 100).
In other words the answer is obtained thru 99.99945% data and only 0.00055% inference.
BUT
If I change the question to: What will the measurement for gravity be in fifty years time?
The answer is the same but with different uncertainty bars.....in this case 10% (i.e. 50/500 x 100)
That is because in this case the answer is obtained with 90% data and 10 % inference...a much larger amount of inference.
I think this is a far system which shows clearly just how much inference is used in an extrapolation.
And people SHOULD know because there is ALWAYS uncertainty in inference.
-
Permalink Reply by Richard A. Meiss on
-
Hi, Ralph -
Yes, I did print a lot of words, and they seem to have had little effect. Here is the essence of what I have been saying (on the original topic). No one disagrees that long extrapolations are subject to error, which increases with the length of the extrapolation. BUT, your proposed method will not work for the reasons that I have been discussing. It does not take into account the nature of the variation in the data set, that is, its statistical description. This must be taken into account when assessing the reliability of any extrapolation. Your method ignores this critical factor, except for a vague reference to "error bars", without any specification as to what they are and how they were derived from the data set. Secondly, the quality of an extrapolation, regardless of its distance, is highly dependent on the reliability and validity of the theoretical and mathematical model that describes the data (and which is the basis of the extrapolation). Likewise, it ignores ancillary information from other related sources that could support the validity of the extrapolation. By the way, please don't confuse extrapolation (a mathematical process) with inference (a logical and deductive process). They overlap, but they are not the same.
I understand that you wish readers to be aware of the general pitfalls of extrapolating data. This is all well and good, although it doesn't relieve a reader of the responsibility for careful and critical reading. But scientists really do wish to communicate their findings in an accurate and timely manner, and they will not adopt a method of describing their data which is mathematically unjustified, logically misleading, and will often obscure the point they are making.
-
Permalink Reply by Ralph Thurlow on
-
Yes, the word "inference" as I've confirmed from various sources does have a rather broad meaning which includes logical deduction.
But we are not talking about that are we!
In the case of mathematical extrapolations, we are talking about INDUCTIVE inference. Induction essentially works by a GENERALISATION or ANALOGY. We all know that while generalisations and analogies can be used tentatively, they are also VERY capable of leading us up the garden path, BTW.
Which ever way you paint this, Richard, the example 'date' of 1 bill yrs. consists of (by proportion) 0.00001% DATA and 99.99999% (inductive, if you must) INFERENCE!
That is the facts (uncomfortable little beasts they can be at times!)
Publishing 'uncertainty' bars based on the relative ratio of INFERENCE to ACTUAL DATA sounds like a fantastic idea!
Resources
© 2012 Created by Creation Conversations.
Powered by
