Mostly Harmless Science Blog

Saturday, February 14, 2015

Define. Clarify. Repeat.



Being A New Post for Marco to Hook Comments On

Like most people with an interest in the historical sciences, Alfred Wegener is one of my heroes. He marshalled an overwhelmingly convincing mass of evidence of the need for continental drift, argued his case coherently and courageously against monolithic opposition, and was eventually vindicated long after he disappeared on a macho scientific expedition across the Greenland ice cap.

Like most people who have thought about it seriously, the pillars of the scientific establishment who mocked Alfred Wegener are also my heroes. Because no matter how much evidence there is of the need for a new theory, you can’t throw out the old theory until you have a new theory. And for a new theory to be science, it needs to have a plausible mechanism. And in the case of continental drift, there was no proposal for how it could possibly have happened that was not obviously wrong until the mid-ocean ridges were discovered, long after Wegener’s death.

To generalise: if you are an iconoclast who wants to convince me to change my mind about something scientific, you need to do two things. (1) Present an overwhelming mass of evidence that the existing models are inadequate: there has to be something that needs to be explained. (2) Present some vaguely plausible model, consistent with the other things we know, that explains this stuff that needs to be explained.

 The rest of this post is just going to be me arguing with Marco, so I’ll see the rest of you later. :)

***
From the most recent comment of Marco down on the ‘Yes,  Natural Selection on Random Mutations is Sufficient’ post:

I'm just going to summarise what I believe to be the source of our differences.

It does not make sense to talk about the source of our differences. You have not yet clarified your model sufficiently for it to be meaningful to talk about the differences between us. As the iconoclast, you need to present overwhelming evidence that the existing model needs to be changed, and some plausible mechanism for an alternative model. These are both necessary. Reiterating that you see the need for a change, and advancing very vague mechanisms that are not linked to the known facts of molecular biology, are never going to convince me. Of course there is no need for you to convince me; but if you want to convince anyone in the scientific community, these are the two things that you need to do. 

1) Experimental basis - your mentioning that a synthesis cannot be experimentally derived by definition, is to me an admission that it is not strictly science. You do believe it to be science by a reasoning I do not accept.

The only way I can construe this statement is that the historical sciences in toto are not science. To me, this is an unreasonable limitation of the meaning of ‘science’. It impacts all possible mechanisms of evolution.

2) expanding informatics principles to genetics - genes as a store of information and instructions analogous to a computer algorithm. To me it is obvious and valid - to you (and biologists in general) it is a big no-no.

This is because biologists know more than you do. The relationship of genes to computer algorithms is only an analogy, and it is a very weak analogy.

3) definition of randomness - Ditching a statistically valid definition of random in favour of a statutory functional one makes the synthesis *not falsifiable* in the statistical sense. One should be able to verify that a simulation based on statistical randomness would come to the same probability distribution.
4) Dismissal of trivial non-randomness. You appear to do this equally for biochemical reasons that mutations would happen in certain areas more than others that are not proportional in any way to the function, but also it appears for things like horizontal gene transfer and epigenetics effects. To me it is an admission that random is incorrect even in your functional description. For instance, I think it is as reasonable a synthesis that horizontal gene transfer explains the spread of all beneficial mutations. I do not think that it is the whole story, but the standard evolutionary synthesis crowds out other ideas as if it had been experimentally verified.

I am absolutely sure that a simulation based on statistical randomness  could show natural selection on random variations was sufficient to account for biological change. I am absolutely sure that a simulation based on trival non-randomness could show natural selection on what I call trivially non-random variations was sufficient to account for biological change. Alternatively, I am sure that a simulation based on statistical randomness could show natural selection on random variations was not sufficient to account for biological change. None of these simulations would necessarily tell us anything about reality. The real system that we are trying to simulate is too complicated. Modelling is not experiment. All models are only as good as their assumptions. This quibbling about definitions of randomness is, to me, irrelevant and uninteresting. It does not get us one step closer towards identifying a deficiency in the standard model, nor clarifying a plausible mechanism for directed evolution.

Monday, September 15, 2014

The Irish Elk is not a good role model for a Higher Education sector

Professor Young said international rankings that show no Australian university in the top 100 for research citations demonstrate the need for dramatic higher education reform.”


No, international rankings in research citations demonstrate nothing.


High research citations are the result of working in important areas that lots of other people are working on. 

The most highly cited results in these important areas that lots of people are working will come from the scientists with the best toys.

We cannot afford the best toys.


I just did a bit of exploring on the internet, and in 2012 the Australian Commonwealth government spent about $850 million dollars on the Australian Research Council, which is the main funding source of curiosity-driven research in this country. In the same year, Samsung spent $42 billion dollars on research and development. There is no way you can expect Australian researchers who are interested in making conducting polymers for flat screen displays, for instance, to compete with that sort of funding. There is no point in the Australian Commonwealth funding work that is trying to compete with the work Samsung is doing, or with the funding priorities of the Max Planck Institute, or of the strategic goals of the California State University system. There are lots more foreigners than us and they have a lot more money. They can afford better toys.

We should spend our limited resources on research that is important to Australia, but less-so to the rest of the world. With things that are of value to the citizens of the Commonwealth of Australia, but that the big fish in the pond can’t be bothered with. There is no reason to suppose that the results of this work will be in topics that are flavour-of-the-decade overseas. If the funding mechanisms of public universities are doing their work properly, they won’t be.

An Australian university making the top 100 list would truly be a great tragedy.  It will mean that Australian problems are being ignored and universities working on problems relevant to our country are being starved of funding. If it ever happens, I will burn Professor Young in effigy and drown my sorrows in a whole lot of soju.

Wednesday, September 10, 2014

I know what you want to see: some half-arsed Climate Modelling!

The graph I showed in the last post wasn’t very good evidence for anthropogenic global warming. If I wanted to scare you, I would show you this graph instead.

This shows the correlation between carbon dioxide and temperature found in a brilliant set of data collected from ice cores at Vostok, Antarctica, where ‘0’ on the temperature axis is the average temperature for the last century or so. [Attribution to text files of raw data: J. R. Petit, J.M. Barnola, D. Raynaud, C. Lorius, Laboratoire de Glaciologie et de Geophysique de l'Environnement 38402 Saint Martin d'Heres Cedex, France; N. I. Barkov, Arctic and Antarctic Research Institute, Beringa Street 38, St. Petersburg 199226, Russia; J. Jouzel , G. Delaygue, Laboratoire des Sciences du Climat et de l'Environment (UMR CEA/CNRS 1572) 91191 Gif-sur-Yvette Cedex, France; V. M. Kotlyakov, Institute of Geography, Staromonetny, per 29, Moscow 109017, Russia]

I came to this data because I wanted to have a closer look at an assertion I have come across a number of times, that changes in carbon dioxide lag changes in temperature in ice core measurements. And yes, it does seem to, but it is a very unwise thing to base a full-blooded skepticism to global warming on. Because the lag is smaller than the uncertainty in the data. The age of the ice, and the age of the air trapped in the ice, is not the same: there is a difference of about 3000 years between the age of the trapped air and the age of the ice, which isn’t known with absolute accuracy, because it takes time for the snow and ice above a little bubble of air to be compact and impermeable enough to trap it there for good. The carbon dioxide content is obviously calculated from the air, while the temperature is calculated from the isotopic ratio of deuterium to hydrogen in the ice molecules. And the imprecision in aligning the exact times of these two sets of data is larger than the lag values that have been reported. It would be nice if this data gave a definitive answer as to how closely carbon dioxide and temperature changes track one another, but all we can say is that on a time scale of +/- 1000 years or so they move simultaneously.  I could just let you draw a line through the data extrapolating to the 400 ppm of carbon dioxide we have today, but I will do it myself.

This is a fit to the data assuming that all the change in temperature is due to radiative forcing by carbon dioxide, fixing T = 0 as 286 K and 284 ppm CO2, with the log of the concentration change giving a change in absorption which has to be compensated by increasing the temperature of a black body radiator, with one adjustable parameter (an invariant non-CO2 radiative forcing) adjusted to minimise the sum of the square of the differences between the fitted curve and the experimental data.

Scary, eh?

If this is the correct way to extrapolate the data, then we are about 6 degrees cooler than we should be, and are just in some sort of lag period - of some unknown length, but definitely less than a thousand years - waiting for this to happen.

I was on the brink of converting myself to global warming alarmism, but I thought I should have a look at the original papers first. Here are some great graphs from Petit et al., 'Climate and atmospheric history of the past 420,000 years from the Vostok ice core', Antarctica. Nature 399: 429-436.


Carbon dioxide is not the only thing that is correlated with temperature changes. Methane, another greenhouse gas, is correlated with temperature changes. (They did the maths in the paper, and r2 is 0.71 for CO2 and 0.73 for CH4). The temperature changes are also closely correlated with the predicted insolation – the amount of sunlight incident on the Earth, varying according to irregularities in its orbit. Dust and sodium (a proxy for aerosols, which we know are cooling) are negatively correlated with temperature changes (r2 is 0.70 for sodium). Ice volume (which is a proxy for water vapour, another powerful greenhouse gas) is positively correlated with temperature.

While insolation can only be a cause of warming, all of these other correlating things can be both a cause and an effect of increasing global temperature. We do not know, just by looking at this data, what is what. A sudden fall in dust and sodium, an increase in ice volume, and a sudden rise in CO2 and CH4 characterises the onset of all of the interglacial warm periods covered in this data. In the graph below I’ve fit the data again, but this time instead of adding an invariant radiative forcing by other things term have multiplied the carbon dioxide radiative forcing term by an adjustable constant to approximate the effect of all the other variables that are changing in synch with carbon dioxide. This constant turned out to be ‘41’ for the best fit, shown. So using this very crude fit, I can extrapolate the effects of *just* changing carbon dioxide concentration to 400 ppm, without any of those other things changing.(That's the line of green triangles hugging the axis from 300 to 410 ppm). This result seems absurdly Pollyanna-ish, even to me, and I'm sure I could make it looks scarier with a model for the experimental data with more adjustable parameters: but that's what the 'suck it and see' model gives me.
I've also put in the observed changes in modern times on this graph. It does make sense to attribute these to CO2 with a little help from the other greenhouse gases we've been putting into the atmosphere. And because we're extrapolating beyond the bounds of the historical data, we may be in a strange and uncharted perturbation of the global climate system. So maybe there is still a significant lag for us to catch up with. Maybe.

But there is one other thing that emerges from this ice core data that suggests very strongly that carbon dioxide concentrations are much more an effect than a cause of global warming. Have another look at this figure:

At the end of each interglacial period, the temperature drops before the carbon dioxide concentration does. This is not a minor effect lost in the uncertainty, like the possible lag in carbon dioxide concentration at the beginning of warming periods; it is a big lag of many thousands of years. Insolation and methane don't behave like that: they rise and fall in lockstep with temperature. What this tells me is that carbon dioxide has historically not been sufficient, by itself, to maintain a warming trend. So we can completely discount any panic-mongering positive feedbacks.

Thursday, September 4, 2014

Slicing Idea Space



The other day I became aware of a publication attacking the group Friends of Science in Medicine – of which friends I am one – published in the minor journal ‘Creative Approaches to Research’, taking Friends of Science in Medicine to task for unfairly attacking complementary medicine. As the publication originated in my own institution, I felt an obligation to respond. So I wrote a vituperative response. If it is not accepted by the journal, you can expect it to appear here in good time; if it is, I will provide a link. So I won’t go into any further detail now.

However, the process of writing the response has crystallised something in my mind which I think is valuable. This something is the distinction between evidence-based and science-based policy. The main coherent argument of the published attack on Friends of Science in Medicine is that complementary medicine is evidence-based, so it is unfair to deny it a place with other evidence-based treatments at the public trough. And this argument is valid. Complementary medicine is evidence-based. There is a vast literature of studies giving positive evidence for complementary medicine. It is published in professional-looking journals. It is peer-reviewed, for what that is worth. The evidence is apparently there that all manner of weird treatments work. But if you look more carefully at this body of literature, it looks much less impressive. The design of the studies is flawed. Controls are missing. Data is cherry-picked to support a preconceived conclusion. Alternative hypotheses that could explain the results observed are not considered. The overwhelming bulk of the great mass of evidence is just not very good evidence. 

And the same is true for many other things that are not complementary medicine: any observation that can be selected from the overwhelming deluge of data that eternally gushes out at us is evidence. 

This picture is evidence for the Loch Ness monster. 

 











This graph is evidence for anthropogenic global warming.








What we need to do is test evidence. The process of testing evidence that has been proven to work is called "science". You might remember my amendment of a Richard Dawkins quote so that it made sense:


We see some evidence, and create a model that explains the evidence. A hypothesis that the evidence means, if we do action X, we will see result Y. We do action X, and see if we see result Y. We don’t do this just once. We think about what the implications of our model are, and what new things it predicts: if it predicts we will see result Y1 if we do never-before-attempted action X1, we try action X1, and see if our model has correctly predicted this new outcome. As described, this procedure seems flimsy, because obviously an infinite number of explanations are possible for anything we observed. So we apply one simple additional requirement: our model has to be consistent with all the other models that have been tested a lot, and not shown to fail yet. Our overall model of the universe has to be consistent. This process of testing the evidence is science, and policy that is based on models describing the evidence that have passed all of these tests is science-based policy.


Clearly, evidence-based policy is better than policy based on the things on the edges of the figure below. And clearly evidence-based will shade off into a penumbra of flimsier and flimsier evidence. If we don’t have science to base our decisions on, we should base them on whatever evidence is available. But, if we have science-based understanding of a phenomenon, we should preferably base our actions on the science, not simply the evidence per se. And if we are spending other people’s money, we should spend it in the most effective way we can. Which means a science-based way where we can, instead of an evidence-based way. Is that clear? 

To recapitulate:

Science-based policy is based on a model that explains the evidence;

It is based on a model that is testable;

It is based on a model that has been tested, and not found to fail;

It is based on a model that does not contradict all the models for different phenomenon that have been tested, and not found to fail.

When we are distributing resources, we should wherever we can distribute them in a science-based way. 

I reckon this distinction between science-based and evidence-based anything is a distinction that is underutilised and valuable, and we should make it, loudly, whenever it relevant.

Thursday, April 3, 2014

Challenge Accepted

Not that long ago I was reading Sir Arthur Eddington's 'The Nature of the Physical World' (which I can heartily recommend) and came across the following claim :

'But no one can deny that mind is the first and most direct thing in our experience, and all else is remote inference.’ 

There is nothing more galling than a statement like 'no one can deny X' to someone who is a vehement denialist of X. I will now demonstrate the falsity of Eddington's assertion that no one can deny the mind is the first and most direct thing in our experience:

I deny that mind is the first and most direct thing in our experience. 

I should probably expand on that.

I assert, contrary to Eddington, that sense impressions are the first and most direct things in our experience. From these sense impressions, we deduce by a process of non-verbal inference both an external world that generates these sense impressions and an ‘I’ that receives them. These two things are cogenerated simultaneously as the world and ‘I’ are disentangled and cannot be separated from one another as things in our experience. Mind and the external world are equal in primacy and directness in our experience.

I will now go beyond the mere fact of denial to rationalise my denial.

I find that very often the ‘I’ that is thinking is a mere passenger on a more fundamental ‘I’ that is acting on the basis of sense impressions without the intervention of mind. I ride a bicycle, for instance, without thinking about what I am doing; I can catch a ball – so long as I do not think about it. As I sit here typing, I do not think about where my fingers are going; if I do, they stutter and fail to go in the right places. I have driven a manual car at some speed on a very complicated path, slowing at certain pre-determined places to throw newspapers in pre-determined directions, without the slightest conscious thought: my mind was entirely consumed in discussing the nature of consciousness with a passenger, and it was another more fundamental ‘I’, the ‘I’ of being and doing, that carried out those complicated actions.

I feel that practice and experience go into improving this more fundamental ‘I’ on which the conscious ‘I’ is but a passenger. My thoughts before I get up to talk seem the same halting, bumbling things they were when I was a dreadful public speaker; but the ‘I’ that does, rather than thinks, now does a much better job of carrying out the task.

I find that ‘mind’ is not associated with being or doing, but with change: with the necessity of doing something different. My sense of consciousness does not flow smoothly; it is strong when I am receiving new sense impressions and need to do something different about them; when I am receiving familiar sense impressions and need only do things I have done before with them, it is much weaker.

When I was young, and much more in my life was novel, I was uncommonly bad at reacting on sense impressions. I could not ride a bicycle; I could not catch a ball. At the same time, my sense of consciousness was considerably stronger than it is now – both mind and the external world were more direct and vivid to me. I was less being, and more becoming. My consciousness did not fade all at once, but neither was it a simple linear process of dulling: it happened in many discrete steps, the first few of which were terrifying, before I became inured to the increasing sense of unreality of myself and the world. The majority of this happened in the few years just before puberty, with additional steps at longer and longer intervals ever since.

When I observe the world around me, I see that I am not the only thing in it that behaves like I do: I am surrounded by animals that react on sense impressions and that are certainly not conscious, that certainly have no mind; by other, ‘higher’ animals that react on sense impressions and may or may not have ‘mind’. I am surrounded by other people who are generally better at reacting to sense impressions than I am; and who, statistically, are rather worse at conscious reasoning than I am. So when I have to chose between the relative importance and primacy of my inferred mind and the inferred external world, it is obvious to me that the inferred external world is the logical starting point for all my more remote inferences.

Thursday, January 16, 2014

Five Myths about MOOCs

Thanks to @Brian_UNE tweeting a link to this article, titled 'Five Myths about MOOCs', I was compelled to drop all the important scientific stuff I was writing and immediately write a screed. So, read the article and then scroll down.



















Five (?) myths about MOOCs in higher education 


1. the idea that “content is free” in education

The idea that ‘content is free’ is not a myth. It is a fact. It is the one unmistakeable fact that is driving all the changes that we see. It is why universities are flailing around like headless chooks. Denial is not an option. Content is free. Accept it and move on.


2. students can support each other

Prof L argues that this won’t work, because universities have always taught with an effective staff student ration of 1:25.

Okay, so we have always taught with a model where a relatively small number of students relied on a tutor to assist them. Where is it written on golden tablets brought down from heaven by the archangel Michael that the way we have always done something is the only way to do it? Consider how our students learn how to play World of Warcraft. One high-level tutor does not shepherd 25 n00bs through instances explaining how to keep aggro off the healer. They learn in small groups, consulting forums and wikis that embody the accumulated wisdom of the community. You can learn anything this way. This is a natural, bottom-up, human way of learning things.


3. MOOCs solve the problem of expensive undergraduate education or educational scarcity in emerging economies

Prof L presents no evidence whatsoever against this so-called ‘myth’. So what if 60% of people enrolled in MOOCs at this moment already have degrees. At one time 60% of the people who owned personal computers were white male uni drop-outs working in their parents’ garages. At one time 60% of the world’s motor cars were made in Germany. Emerging technologies are going to be localised. Early adopters are not the same cohort of people as late adopters. 


4. Education is a mass customer industry

Now, here I can agree with Prof L. Education isn’t a mass customer industry. It isn’t an industry at all. It is a human activity as natural as eating or playing sport, and a fundamental human right. If the education ‘industry’ as constituted currently is getting in the way of changes that are making it more accessible and affordable, it needs to die in a fire.

Sunday, November 3, 2013

Ow. Ow. Ow. Ow.



I don’t think I can do this. I really can’t. 

I’ve been asked to apply for promotion next year, and one of the mandatory things is to submit at least three ‘Student Evaluation of Teaching’ reports. These are evaluations, not of the unit, but of the lecturer, and they are not compulsory for the students to fill out. 

While there is a process for getting us to do 'Unit Evaluation' surveys as a matter of course, 'Student Evaluation of Teaching' surveys aren’t done automatically: instead, you request the teaching-management-minions-that-be to do them your behalf - by the simple expedient of sending them (the minions) an email. 

I have gotten away without doing any of these for the past nine-and-a-bit years. It isn’t actually because of the excuse I gave my colleagues the other day, that it is too much bother (after all, I just have to send someone an email). I don’t like the whole idea of them. The thought of using them in a promotion application makes me twitchy in a way people who knew me in high school will remember.

Why, you might ask? 


#1. They don’t measure anything relevant. 

With all respect to my students – who are uniformly great people, eminently deserving of HDs and free beer – a student who has just completed a unit is not yet in any position to evaluate the unit or the lecturers who have helped them through it. They don’t know if the skills and knowledge they obtained from it will be useful to them in their career, they don’t know how it fits into the whole body of knowledge and skills they will obtain in their degree, and they can’t judge whether it will have a permanent impact on how they view the world or was just an entertaining intellectual cul-de-sac. They can't judge whether their lecturer has given them a fatally flawed and bogus take on the topic, or has set them up with a solid basis for an ever deepening life-long understanding of it. The immediate impact of the unit or the teacher on the student is not relevant to the desired educational outcome.

Okay, so they don’t measure anything relevant. But I can just about put up with all the rigmarole about citation counts and impact factors – which also aren’t measures of anything relevant. Why can I swallow irrelevant measures of value in my research, but not in my teaching?


#2. They measure the irrelevant thing badly.

With research, the irrelevant indicators are at least reasonably transparent and quantitative measures of something. Okay, forget the goal of measuring how I helped the unit to meet its true educational outcome. How well did I help the students pass tests and keep them entertained in the process?  This is also something that student evaluations of teaching can’t really tell me.

You can’t step in to the same river twice. So a student can judge how they did in my part of a unit compared to how they did in other parts of the unit, or how entertaining my part of the unit was compared to other parts of the unit, but they can only encounter my material for the first time once. The material and the lecturer are inextricably entwined, so on the more modest goal of judging how good I was at getting them to know topic X, or entertaining them while I did it, a student survey is also flawed. They can only compare me with other lecturers teaching topics Y and Z – topics which might be intrinsically easier or harder and more or less entertaining.

And, since these evaluations are not mandatory, the proportion of students who fill them out is always woefully unacceptable by the standards of a poll or any peer-reviewed work in the social sciences. The only students who will be bothered to answer them will be the students who want to drive a stake through my heart and bury me in the crossroads at midnight, and those who want to have my baby. Normal middle-of-the-road representative worked-off-their-feet students will not bother.  

Those first two complaints are almost equally applicable to 'Unit Evaluation' Surveys. Which I don't like doing either, but I do when I have to.

There are two other  irritating things that only apply to these ‘Student Evaluation of Teaching’ reports:


#3. They are open to abuse.

With research, I can’t pick and chose what part of my ouevre to display, unless I want to cut my own throat and look unproductive by leaving a whole bunch of papers out. They are all out there in the public domain anyway, with quasi-empirical quantitative variables attached to them telling you how popular they are. 

But the rules for the promotion application are practically begging me to cherry-pick the very best teaching evaluations I can, with no oversight. That is just bad. Bad! No peer-reviewed journal in the social sciences would accept a methodology where researchers conducted ten surveys and reported on the three that gave the results supporting their theory.


#4. They are an imposition on the students.

I know these surveys don’t measure anything relevant. And any qualitatively useful information about things I might have done badly, or compliments that make me feel warm and fuzzy that I am on the right track, show up on the Unit Evaluation surveys anyway. So I don’t need these Teaching Evaluation surveys to learn anything that might be useful for current students. Or future students. They are only useful for me. I don’t want students to waste their time doing something that is only useful for me. I would rather they spent their time creating new Chemistry Cat memes.