Global Warming and Pirates

Global warming is inversely proportional to the number of pirates

formatting link

Mikek ;-)

--
This email has been checked for viruses by Avast antivirus software. 
http://www.avast.com
Reply to
amdx
Loading thread data ...

Am I suppose to not notice the highly tweaked X-axis scale? Or, was the curve drawn before the data was fabricated?

More of the same:

Did you know that driving causes obesity?

--
Jeff Liebermann     jeffl@cruzio.com 
150 Felker St #D    http://www.LearnByDestroying.com 
 Click to see the full signature
Reply to
Jeff Liebermann

That's Big Data. The more you know, the more you can get wrong.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

No, they got the axes wrong. As temperature rises, the number of pirates decline. If you can't stand the heat, get off the ocean.

Reply to
John S

--
Ergo, restrict the dissemination of knowledge and the error path will 
pinch off to the point where no errors can be made? 
 Click to see the full signature
Reply to
John Fields

Do you know why we have had an increase in pirates from 2000 to 2015?

Mikek :-)

Reply to
amdx

That was fun! Mikek

Reply to
amdx

Arr.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

I grew up on "How to Lie with Statistics" which has served me well over the years: I know very little about big data (except when a non-backed up quadruple redundancy system hiccups, scrambles everything, and I get stuck with putting Humpty Dumpty back together from multiple incremental ancient backups). I guess it's much the same as small data. Some observations:

- The bigger they are, the harder they crash.

- Data is free. Information must be purchased or stolen.

- Garbage in. Exponential quantities of processed garbage out.

- Research moves the decimal point to the left. Funding to the right.

- Because the errors tend to cancel, the desired answer is the average of all the wrong data plus or minus some simplifying assumptions.

- In a research report, the data is owned by the researchers, while the conclusion and summary are owned by the sponsors.

- Data can bets be manipulated to produce the desired conclusion by cherry picking which conveniently leaves no evidence behind.

- Trend lines can be easily manufactured.

The last one requires an explanation. Let us do some plotting comrades. Here's a graph of the past rainfall statistics for my area (San Lorenzo Valley CA): If I use a high odd order polynomial trend line, I get a radically rising trend that predicts increasing rainfall. If I use an even order, the trend is downward predicting a worse drought. The spreadsheet(s) used to create the plot is in: if you want to play with the graphs to see how it works. While it's easy to see that there's something wrong with the trend line in the aforementioned graph, a lower order polynomial can easily be made to simulate a rational projection:

--
Jeff Liebermann     jeffl@cruzio.com 
150 Felker St #D    http://www.LearnByDestroying.com 
 Click to see the full signature
Reply to
Jeff Liebermann

I disagree on that one.

Garbage in, garbage factorial out.

The higher order the polynomial fit, the more radical the curve can get.

I love the climate models that always nail temperatures up to the date of the model run. But come back 5 or 10 years later, and the "future" fits are terrible. Of course, the models are *so* much better now!

--
John Larkin         Highland Technology, Inc 
lunatic fringe electronics 
 Click to see the full signature
Reply to
John Larkin

Of course, if you know what you are doing, you estimate the confidence limi ts on all the parameters you get out of a polynomial fit.

It doesn't take many parameters before the polynomial can fit anything, and the confidence limits go to infinity.

Fitting to sets of orthogonal functions is easier - the higher order functi ons don't interact with the lower-order functions, so the confidence limits on the lower-order parameters aren't compromised by the process of fitting the noise in the data.

Actually, you love thinking that that's what's going on. Climate modellers are a little more more expert than you are in the modelling game, well awar e of the traps that beginners fall into, and inhabit a rather richer univer se than you seem to be able to imagine.

formatting link

That's pretty much the first time that I'd come across a "Madden-Julian osc illation"

formatting link

It's 60-90 day period is way longer than local weather-forecasting can deal with.

--
Bill Sloman, Sydney
Reply to
Bill Sloman

This isn't actually true, or entirely so:

More than a few researchers have been caught falsifying data, causing resignations or better. How? By using the same statistics against them: their data looks "too perfect" and lacks the expected number of outliers (to the tune of several sigma).

There are also funnel plots, which appear in meta-studies. The size of the study should be correlated against the variance of its conclusion: small studies (the most popular kind) give widely varying results (even if they're at p < 0.05, because they're small and there's so many of them), while larger studies (which are less common and more expensive) give better results. It should be a pyramid (or funnel) on the plot. What typically ends up happening is, the unfavorable studies are discarded during analysis, or declined from publishing (voluntarily, or due to any number of actors along the way -- because a false or null conclusion isn't very interesting to a journal, right?), so the funnel diagram gets weighted to one side.

This also works for citations: in a review paper, you can conclude almost anything if you cherry-pick your citations, rather than including all the positive and negative ones related to your subject. But that's detectable by a study of completeness.

Meta humor: apparently, a meta-analysis of meta-analyses showed a lopsided funnel indicating excessive confirmation of publishing bias. (Put another way: few perform such meta-analyses unless they suspect the data?)

Ah, but this involves the most fundamental aspects of modeling: you need to use a function which, over a much larger scale, is a good fit to the subject at hand.

You don't use polynomials to model electronic signals, because signals are finite and bounded for all time; polynomials explode, sooner or later! You can use a Taylor series to make a local best-fit of sin(t), but ignoring the very conditions of that approximation will find you in hot water (or simply too much water, as in your example :) ). We certainly should not expect that rain(t) goes towards +/-infty at t-->infty, or -infty for that matter -- the universe didn't exist yet! Indeed, rain going negative is a completely absurd notion*, so these functions are quite an awful choice, indeed!

(*I suppose one which has a little relevance, if one tracks evaporation as well. But still, it's bounded by physical law.)

So one should wish to use locally finite and nonzero functions, which tend toward finite or zero values at infinity. Some examples include sin/cos, humps and sigmoids like 1/(1+x^2), 1/(1+exp(-x)), exp(-x^2), etc., and perhaps more peculiar functions like sin(x)/x, wavelets (sines weighted by Gaussians), Bessel functions (roughly like sin(x)/x, but aperiodic).

The downside to "locally humped" functions: they don't have any predictive power, because guess what, the humpiness doesn't repeat. And, how could we even begin to estimate it, when the underlying phenomenon is chaotic? For these purposes, a statistical method seems much better suited: rather than plotting a single, absolute prediction, plot a likely *range*. The disappointing thing, I suppose, is that such a statistical solution, despite being the absolute best predictor we can scientifically obtain (in the absence of an extremely advanced model of the overall system), gives us no new insight -- that it seems to be disappointing that our Farmer's Almanac is the best tool, in essence.

But alas, missing these (often extremely subtle) factors is all too common in data analysis, so your point remains valid, of course.

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
 Click to see the full signature
Reply to
Tim Williams

Nice inverted scale, but with a numerical switchback at the left..

Reply to
Robert Baer

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.