The Button Collector or When does trend predict future values?

In my humble opinion, whenever anyone compares any scientific statistical argument to the stock market, they’ve lost the plot.
When I take a hot caserole pot from the oven, I use oven gloves so that I don’t burn myself. An hour later, then I take the same caserole pot from the table to sink, I don’t use oven gloves. I have successfully predicted the future. Sure, I can’t know the exact temperature, but I can be pretty sure it will be close to room temperature. I can even take a few temperature readings soon after removing it from the oven, and make a pretty good guess at when it will be comfortable to hold the pot.
Note that unlike the stop market, this is unlikely to be affected news of some CEO having an affair, or by whether the Republicans and the Democrats agree on some particluar financial arrangement. Unlike the stock market, the pot is unlikely to be impacted by whether some trader had a headache this morning. Comparing the two is stupid
In fact, the author of this posting completely agrees. He says:

Armed with this secret formula, which is precise and immutable, we can now adjust our model of this button collecting process.

And here we get to the climatology comparisons. Some climate scientists seem to believe that they know the secret formula – in fact, they go further and claim it isn’t secret, but common knowledge based on physical principles – and using this formula they can predict the future using computer-based models. Other scientists, and much of the WUWT readership, believe that the formula is too simplistic, that it doesn’t allow for a variety of internal and external influences, or that it is tuned incorrectly so that some real influences (such as CO2 forcing) are assigned too much of an influence.
But it isn’t the stock market. And it isn’t some random collection of buttons. So don’t make a silly comparison to ‘prove’ a point.

In the example given, by analysing all past data, you would be able to make a prediction to some level of confidence (say 90%) what the minimum & maximum number of buttons the next day was likely to be. This is done all the time in many fields – for example we don’t know how many particles a radioactive source will emit in the next minute, and we don’t know the underlying mechanism which causes a particular atom to split, but we can estimate a range of values and we can use that knowledge for useful purposes (in a nuclear power station for example). To characterise this nuanced outcome as “We have no idea how many buttons he will have Friday night because we cannot see into the future” is absurd and just plain wrong – we have an “idea” but we don’t know for certain.

On statistics, it is estimated that 50% of all people who have ever lived on earth are alive today. Does this mean I only have a 50% chance of dying? 😉
Of course this post is overly simplified, it’s loosely applying the ad absurdum rhetorical tool. I’m sure most folk frequenting this particular virtual space are perfectly aware that in practice we can attach PROBABILITIES to certain potential outcomes based on HISTORICAL trends derived from hard data.
Nothing wrong with that, and many people base important decisions (investment, policy, research, travel, etc.) on that very concept. In fact we all do this constantly. It’s an integral part of human existence. From the moment we wake up of a morning, we expend a great deal of mental energy throughout the day subjectively assessing risk on the basis of the cognitive models we have constructed over time in accordance with the lessons life has taught us (just take the drive to work, for instance). It’s part of the subjective cost/benefit analyses that life demands of us all continually, and we are all far more adept at it than many people realize. (Indeed, the better we are at this, the more Antifragile we become.)
The point is, there is a huge difference between modeling probable outcomes of physically quantifiable processes, and modeling trends arising from human action. The former entails a higher objectivity quotient, the latter a great deal more subjectivity, thus making it less predictable based on existing trends. And I think that that is an important distinction to make in this thread. (I call this distinction the Light/Life Ratio.)
However, both types of forecasting are at all times subject to Taleb’s now proverbial Black Swan Event (meteor impact, unforeseen geoplitical event, etc.). Obviously, therefore, even the highest probabilities properly assinged to a given trendline can miss the mark completely. And it’s this fact that I think Kip Hansen is getting at here when he asserts we cannot predict the future using past events.
I imagine that the above is as plain as a pikestaff to most folk here, but I suspect that there are a great many others for whom this is not so obvious. And it’s their hearts and minds we seek to win from the prophets of the Great and Terrible Day of Global Thermageddon!
Just remember, “He is no friend that guides not, but ridicules, when one speak from ignorance”.

I would add a caution to the author’s conclusions. No matter how perfectly one understands the underlying causes of a phenomenon, models will still utterly fail if the phenomenon being modelled is inherently nonlinear. As a simple example, consider orbital mechanics. We understand Newtonian mechanics (as modified by relativity) with sufficient precision to be able to predict the ground paths of solar eclipses, the trajectory of Voyager or Cassini, or the return of Haley’s Comet. These are linear systems and thus can be modelled with some predictive accuracy. However, a nonlinear system (like weather) cannot be modelled over any significant duration because it is sensitive to infinitesimal changes in starting conditions. Nonlinearity prevents prediction over anything beyond the shortest timelines. Lorenz recognized this in the 60s, and the IPCC acknowledged it decades ago (and then promptly swept scientific reality under the carpet and spent the next several years desperately clinging to a politically approved but empirically falsified hypothetical linear relationship between anthropogenic CO2 emissions and delta T).

This issue was treated once very well by Karl Popper, e.g. in The poverty of historicism. Is has nothing to do with statistics as it can be summarized by his statement that regularities or not laws of nature. Even the sun-rise tomorrow is not a law of nature and cannot be predicted. We do not even know the probability of this event.

Hi, Kip. I’m afraid I agree with Fred (the Statistician) above, and others who have made similar comments. I think you have gone quite overboard by saying we have “no idea” what will happen next. This is the sort of black and white thinking that gets a lot of people into rhetorical trouble. Statistics is not about black and white – it is about quantifying the grey.
On a related point, you said “Models, not trends, can predict, project, or inform about possible futures, to some sort of accuracy.” This is true, about models, but it is (of course!) also true about trends. You then add, “Models must include all of the causative agents involved which must be modeled correctly for relative effects.” This is absolutely incorrect.
Models *never* include “all of the causative agents involved.” Never. A model is a simplification of the real world. By definition, some things are left out. We hope, when building our model, that only trivial causes (or trends) are left out, and what we have is a pretty good approximation. But models are always a simplification of the real world.
Further, it is entirely reasonable to sometimes leave out ALL causative agents. This would be an empirical model – i.e., a model based only on data, and one that tries to capture or describe the trends found in our data set. Absent any other knowledge, an empirical model will be the best predictor of future values – i.e., better than random guessing (which – absent any other knowledge – would be our only other option.)
Finally I would submit that empirical modeling is always the first step of any attempt to understand a phenomenon. That is, it is the first step in the scientific process. You first notice a pattern in nature or society. You try to describe that pattern, and – if it persists to some degree – you then investigate the pattern to look for causal agents – the underlying mechanisms that produce the pattern. But you have to notice the pattern first, and “noticing a pattern” is another way of saying “building an empirical model.”
The Theory of Gravity is the classic example of this. The empirical results are so thoroughly robust and understood that we even call it a Law. But we don’t know what “causes” gravity … we can model it very, very well, but we don’t know the cause. (Yes, there are a few hypotheses of late; but these are as yet uncertain.)
Cheers.

If every event is causally related to every other event, past or present, forming a compete model is a forbiddingly large enterprise. Worse if the causal relations are non-linear. In brief – trends and models are just masks to conceal our ignorance. However all is not lost – we can and do form very accurate predictions – so accurate that they are ALWAYS correct.
If I am x% (it does not matter what value ‘x’ has – this does not affect the accuracy of our prediction) certain that the sun will or will not rise tomorrow. So long as I do not claim perfect knowledge or absolute certainty, I shall always produce correct predictions; even better, I can never produce wrong ones!
This is why IPCC and all the other quacks playing with models/ and trends to produce forecasts/predictions can continue to syphon money from tax payers – they are never wrong! Even if their predictions never correspond to the events, they can (and do) claim increased confidence in the skill of their models and they can justify whatever causal relations they may fancy – yes – even that Great Greenhouse in the Sky!

@acementhead
Point taken, sorry for being so sloppy! However, it was only meant as joke about how statistics can be used to assert any nonsense (hence the winking smiley). But I consider myself suitably chastised…
Here’s a good link to support acementhead’s correction of my spurious nonsense:
http://www.prb.org/Publications/Articles/2002/HowManyPeopleHaveEverLivedonEarth.aspx
Do you agree though, acementhead, that statistics can and are used to support all manner of claptrap?

mhx says:
October 18, 2013 at 2:06 am
“Suppose that a car at a distance of 500 meters starts driving in your direction and for 490 meters it drives in a straight line at constant velocity. What would you do? “
—————
I would predict that the right-front tire would “blow-out” at the 492 meter mark and the car would veer off the road and crash. That is unless my earlier prediction proved to be correct that it would run out of fuel at the 491 meter mark. 🙂
Predicting earth’s climate for the next 100 years is akin to predicting Super Bowl or NBA winners for the next 10 years. Iffen you think you can, …. go for it, ……. place your “bets” today.
If one could magically remove all of the CO2 from the atmosphere there would be no measurable effect/change in the near-surface air temperatures and that is because 398 molecules of CO2 are irrelevant when intermixed with 20,000 to 40,000 molecules of H2O vapor.
Thermal energy in the near-surface atmosphere is NOT cumulative from one (1) week to the next and surely not one (1) year to the next.