Pseudoscience

I was recently pointed to a hilariously absurd article called Scientists find Extraterrestrial genes in Human DNA. It saddens me that the site hosting this nonsense is associating itself with my favorite country.

Anyway, I'm sure most people would recognize this particular article for the idiotic drivel that it is (even most of the believer sites that I found referencing it seem to discuss it with an air of incredulity), but I thought it would be fun to analyze it in detail regardless. Let's look at the opening statement:

A group of researchers working at the Human Genome Project indicate that they made an astonishing scientific discovery: They believe so-called 97% non-coding sequences in human DNA is no less than genetic code of extraterrestrial life forms.

The non-coding sequences are common to all living organisms on Earth, from moulds to fish to humans. In human DNA, they constitute larger part of the total genome, says Prof. Sam Chang, the group leader. Non-coding sequences, originally known as "junk DNA", were discovered years ago, and their function remained a mystery. The overwhelming majority of Human DNA is "Off-world" in origin. The apparent "extraterrestrial junk genes" merely "enjoy the ride" with hard working active genes, passed from generation to generation.

Look at the wording of the first paragraph. It's written to imply that the astonishing discovery is that there is alien code in human DNA, but if you actually read what it's saying, the astonishing discovery is that they believe that there is alien code in human DNA. I wonder if that was intentional. Hmmmmmmmm.

They are talking about Junk DNA here. There are a lot of hypotheses as to what it is and how it formed--and I bet they are all a lot more credible than Sam Chang's "discovered" belief that it was planted there by aliens. Speaking of Prof. Chang, I did a search for his name at the Human Genome Project website, and there was nary a Chang to be found. I also checked their list of publications and, surprise, there was nothing there with any mention of extraterrestrials. A few PubMed searches showed many papers attributed to various S Changs, but none that talk about extraterrestrial origins of non-coding DNA.

Let's just think for a minute about what process would be required to scientifically test for the presence of extraterrestrial information in human DNA. The first (seemingly insurmountable) obstacle that must be overcome would be that you must know what extraterrestrial information in the human genome would look like. No problem, right? Just head on down to your local branch of Area 51 and ask the men in black for some sample data. Uh huh. Let's see how Chang approached this issue:

After comprehensive analysis with the assistance of other scientists, computer programmers, mathematicians, and other learned scholars, Professor Chang had wondered if the apparently "junk Human DNA" was created by some kind of "extraterrestrial programmer".

This paints a pretty clear picture for me. I can see the exchange now (doodity doo doodity doo doodity doo (those are the sounds of transition to a fantasy sequence in the TV show of my mind)):

Chang: I have done a comprehensive analysis of the human genome. I have called you all here to assist me.
Scientist: Who are you?
Computer programmer: You said there would be free donuts... But I don't see any...
Mathematician: Why are wee meeting in your mom's basement?
Chang: Hmmm. I bet this "junk DNA" stuff was made by aliens!
Other learned scholars: This is gay. We're leaving.

The article just gets more bizarre from there. Chang's grand hypothesis is that an alien race was working on coding an ultimate human race, the code for which is in our non-coding DNA. Due, however, to time restrictions from the alien programmers' bosses (or whatever), the aliens were forced to scrap the large code base for the ultimate humans, and release a stripped down basic version of their DNA program (ie., our coding DNA sequences). Since our "basic" version of DNA isn't as powerful as the scrapped version stored in our non-coding DNA, we are not immune to cancer (as we would be had the aliens' bosses given them an extension on the "Earth project"). The article also states in various places that this is all "verified fact" and that there is "absolute proof" to support this theory. Right.

So, let's recap:

1. Chang is probably not even a real person, and even if he is, he is grossly misrepresented in the article.


2. The article claims to be about science, but is in reality just a bunch of poorly thought-out childish fantasies.


3. The website hosting it is called The Canadian.


4. The Canadian also has articles titled AIDS is bio-terrorism created to kill billions some doctors and scientists say, Testimony claims crop circles made by "alien allies" of humanity, and Global Warming reveals that Stephen Harper and George Bush are not Christians.

What conclusions can we draw from this information? It's simple: Canadians, myself included, are full of shit, and you should never believe anything we say. Ever. Especially if it's on the Interweb.

What would it mean if The Secret were true, and there really was a Law of Attraction? Let's brainstorm!^*

• Over 800 million starving people around the world simply don't want food badly enough.



Behold the power of The Secret™.



• Millions of people who die from AIDS and cancer every year, didn't think about staying alive hard enough.



If only she knew The Secret™.



• Every two and a half minutes, someone somewhere in America wanted to be sexually assaulted.



He can come for you, too, if you know The Secret™.



• It didn't cross the minds of over six million Jews to think about not being slaughtered during the Holocaust.



Looks like they didn't know The Secret™.



• Hundreds of thousands of people in New Orleans were wishing that their houses and lives would be devastated by Hurricane Katrina.



The Secret™ strikes again!



• Thousands of soldiers in Iraq are dead because they didn't think enough anti-bullet and anti-explosion thoughts.



^** The infidels must never learn The Secret™, or we shall surely perish.



• Not enough of the thousands of people who died on 9/11 were concentrating strongly enough on the buildings not collapsing.



If only the movie about The Secret™ had come out a few years earlier.



• Let's not forget about Steve Irwin.



The Secret™: Not just for humans any more!

It's called blame the victims, folks. Oprah does it, so it can't be wrong! If you'll excuse me now, I'm going to go visualize me up some french fries and gravy. Yum!

^* Yes, I realize that this post is in incredibly poor taste. Hopefully it conveys how amazingly fucking stupid I think The Secret is.
Special thanks to Tombstone Generator and The Newspaper Clipping Generator.

An idea struck me after reading the article at CSI by David F. Marks and John Colwell that provides a critical analysis of Rupert Sheldrake's staring experiments (see my previous post on the subject here). It occurred to me that the gambler's fallacy might be able to explain their results (as discussed in another previous blog entry here), and that I might be able to verify their conclusions for myself by means of a simple computer simulation. Being the complete uber-geek that I am, I immediately set to work.

The Program

I spent a few hours hammering out a fairly simple Perl script that could be used to simulate a few different kinds of trials. The source code can be viewed here. I've tried to make the code as easy to read as possible, with commenting where appropriate (an unbelievably ginormous leap from my normal coding practices). The program simulates staring experiments using either Sheldrake's trial sequences, or random sequences (generated using Perl's built-in rand() function). The virtual-test-subjects' guesses can be configured in any one of three ways:

1. completely random (using Perl's built-in rand() function),


2. using a gambler's fallacy algorithm applied to their own guesses, or


3. using a gambler's fallacy algorithm applied to the actual trial types.

It is possible to configure a yes-bias for the subjects (baseline increased likelihood that a user will choose yes instead of no)--this is set to 5% for the purposes of this article, since this is the number that seems to come up most frequently in the relevant literature.

The gambler's fallacy algorithm I used in the simulations essentially increases the subject's likelihood (configurable in the program, set to 33% for this article) of choosing the opposite answer on their next guess, if the previous trial or guess was the same as the current one. For example, using method #3 from above, the subject's chance of guessing yes on trial 1 is 55%. Say trial 1 is a yes. On trial 2, their chance of picking yes is still 55%. Say trial 2 is a yes. Now, since the previous trial (1) and the current trial (2) were both yes, the subject thinks it is less likely to be a yes on the next trial. On trial 3, the chance that he will guess yes is now 22%. If trial 3 were also a yes, then his chance on trial 4 would drop to 0% and stay there until a no trial came up which would reset him back to 55%. This is also applied conversely in the event of multiple no trials in a row. Using method #2 above, the numbers work the same, but instead of looking at the actual trial types to adjust the values, the subject bases the fallacy on his own guesses (essentially ensuring that the subject will never guess yes or no too many times in a row--instead spreading guesses out more evenly over a number of trials, as a real human might be more likely to do). Method #3 assumes that the subject is given feedback after each guess about whether he was right or wrong, method #2 assumes that he is not.

Got all that? Me neither, but that's the best I could explain it. Just learn Perl and look at the code if you're unsure. Let's move on.

When generating random trial sequences instead of using Sheldrake's sequences, it is possible to specify the number of subjects and trials to simulate. In either case, the gambler's fallacy modifier and "memory" of the previous trial is reset for each subject.

The source code linked above is in the configuration to run 20 experiments using Sheldrake's trial sequences, guessing method #3 from above, with a yes-bias of 5% and a gambler's fallacy modifier of 33%. These values and methods are all easily modified by adjusting the values of variables in the first few lines of code. Since Sheldrake provides 20 sequences of 20 trials each, the total number of trials that will be simulated when the program is executed is 8,000.

Upon completion of the simulated trials, the program outputs the following results for each experiment, and for the total run of all experiments:

1. total number of trials run across all subjects,


2. the total number of trials that were staring trials,


3. the total number of staring trials that were correctly guessed by the subject,


4. the total number of non-staring trials that were correctly guessed by the subject,


5. the total number of trials that were correctly guessed (#3 + #4 from above), and


6. the breakdown of subjects into Sheldrake's "plus or minus" categories, in which subjects who guessed correctly more often than incorrectly are labeled "plus" subjects, subjects who guessed incorrectly more often than correctly are "minus" subjects, and subjects who guessed correctly and incorrectly an equal number of times are ignored.

The Method

Using my program, I generated four sets of results (Set I, Set II, and Set III, and Set IV--roman numerals are t3h aw3s0me). Each set consisted of the 20 simulated experiments, each consisting of 20 trials on 20 different subjects, for a total of 8,000 trials each.

Set I

For the first set, I used randomly generated trial sequences, and guessing method #1 (no fallacy algorithm) from above. This was primarily to test the accuracy of my program under truly random circumstances. The results I expected to see were values that were statistically insignificant and no better than chance.

Set II

The second set used Sheldrake's sequences, and guessing method #3 (gambler's fallacy when feedback is provided). This would test whether or not Colwell's conclusions--that Sheldrake's sequences are sufficiently non-random enough to allow for statistically significant results without any paranormal effects (assuming that the subject is given feedback)--could be illustrated using my simple gambler's fallacy model. The results I expected were values that would show the subjects guessed correctly significantly more often than chance would dictate.

Set III

The third set used Sheldrake's sequences, and guessing method #2 (gambler's fallacy when no feedback is provided). This was purely for curiosity's sake, to see if the gambler's fallacy has any effect when applied even without feeback. Based on common sense and the conclusions drawn by Colwell et al in their experiments, I expected no significant results above chance to be shown in this set.

Set IV

For the final set I used randomly generated trial sequences, and guessing method #3 (gambler's fallacy when feeback is provided). I wanted to see if gambler's fallacy also could have a significant positive effect on truly random sequences, and if so how did it compare to Sheldrake's pseudo-random sequences. I expected a possible significant effect here, but not as strong as in Sheldrake's sequences.

The Results

First, I will provide a table of overall results from all four Sets, then discuss and further analyze each individually. For ease of comparison, I will use the same table layouts and graphing methods that Rupert Sheldrake presents in his paper The Sense of Being Stared At - Part 1: Is it Real or Illusory? (pdf).

Set	Trials	Right	% right	N	+	-	p	Raw data
I	8,000	4,013	50.16	400	160	167	0.6987	set1data.txt
II	8,000	4,143	51.79	400	187	140	0.0093	set2data.txt
III	8,000	4,040	50.5	400	165	162	0.8682	set3data.txt
IV	8,000	4,029	50.36	400	175	166	0.6260	set4data.txt

The sign scores are tallied the same way Sheldrake did in the paper mentioned above. Subjects who were more often right than wrong were scored +, subjects who were more often wrong than right were scored -, and subjects who were right the same number of times they were wrong are excluded. The p values are also calculated the same way Sheldrake did, using the chi-square test where the null hypothesis is that + and - should be equal.

As you can see, Sets I and III have high p values, indicating no significance beyond chance, as I had predicted. I expected a possible significance in Set IV, but this did not pan out in the actual trials. The fact that these three sets did not produce any significant results essentially means the following:

• (Set I) when the trials are completely random, and the guesses are completely random, the results are dictated by chance (duh!);


• (Set III) when Sheldrake's sequences are used, and no feedback is available to base the guesses on (but the guesses themselves are affected by the gambler's fallacy), the results are dictated by chance; and,


• (Set IV) when the trials are completely random, feedback is given to the subject, and the guesses are affected by the gambler's fallacy, then the results are dictated by chance.

The interesting set was Set II, the one for which I predicted that the subjects would guess right significantly better than chance would dictate. This was the set that used Sheldrake's trial sequences, provided feedback to the subject whether each trial was stare or non-stare, and had guesses that were affected by my gambler's fallacy algorithm. As you can see, the very low p value tells us that, shockingly, I was right--the results are significantly above simple chance. Here are some charts made from my Set 2 data, also modeled after the ones Sheldrake produces in his paper:



Chart A gives the percentages of correct and incorrect guesses in staring trials, non-staring trials, and total trials. The findings that people did much better on the staring trials than non-staring trials is consistent with Sheldrake's findings.

Chart B gives the number of subjects who guessed correctly more often than incorrectly and vice versa in staring trials, non-staring trials, and total trials. Again, subjects did much better in staring trials than in non-staring trials (which is also consistent with Sheldrake's findings).

Conclusions

This little personal exercise has solidified for me that any experiments that use the trial sequences that Sheldrake provides on his website should be disregarded. Indeed, this is true for any experiments at all that use pseudo-random trial sequences arrived at using similar methods to those used by Sheldrake. The fact that a computer simulation implementing a simple gambler's fallacy algorithm can clearly produce statistically significant positive results in-line with Sheldrake's findings makes it an undeniable fact that this can happen even in the complete absence of any paranormal psi effects.

If you're feeling particularly sadistic, you should download and run my script with the settings from Set II, but with a way higher experiment number (for example 50 or 100). It is 100% consistent in showing huge statistical significance using the p value as calculated by Sheldrake, even with ridiculously huge trial numbers. (Just in case someone gets the idea that I cherry picked the best run to fit my hypothesis out of a billion runs, or something. The results I used were all first-runs with the specified settings.)

I do recognize the fact that the effect shown by my program is not as strong as the one that Sheldrake reports (mainly in the "% right" column from the results table). The human mind is a complex thing though, much more complex and clever than about 30 lines of code in a Perl script (most human minds, at any rate). It is no stretch of the imagination to presume that the subjects involved in real human experiments employ much more complex and effective strategies for guessing the correct answers (such as some kind of subtle pattern learning as suggested by Marks and Colwell) which could account for the slightly better hit rates. It might even be possible to tweak some of the variables in my program to improve the numbers. The ultimate fact remains that, when using pseudo-random trial sequences, such as the ones provided by Sheldrake, it is entirely possible to get statistically significant positive results in staring detection experiments due to reasons totally unrelated to psi or any other paranormal effect.

Q E (Oh man did I really waste two whole days on this?) D.