William M. Briggs

Statistician to the Stars!

The Mysticism Of Randomness: Philosophy of Probability & Statistics Video Series

Nothing can be caused by “chance”. Chance, like randomness, isn’t a thing, it isn’t real, it isn’t physical, therefore it cannot cause anything. Randomness means “unknown.” Therefore “random variables” are only propositions the truth value of which is unknown. Mysticism pervades probability and “randomizing” is like a witch doctor’s incantation.

See also the video Probability Is Logic.

———————-

I’ve improved, ever so slightly, the way this one looks. I can’t figure out how to get my whole body into the frame with my non-adjustable webcam. I need to have my hands (for the coin trick). Well, you can at least hear me, though my voice sounds like it’s coming through a tunnel.

Update Comments wanted on video length. Too long? Too short? Just right?

21 Comments

  1. You own two suits? I think that I heard you say that there is randomness in marriage. Brother, you are sure in trouble now.

    I suppose stepping further back is not an option since you seem to be recording in a broom closet. You could also put a book under the computer, but then I assume that you tried all these options. Some sort of autofocus seems to periodically blur the image. The important thing, however, is that you have a good speaking voice which is much better than I could do. Do you have a teleprompter or is this off the cuff?

    As to the topic, although I understand your reluctance to use chance as a causative mechanism it can still be a useful short hand way of talking. For example, in your book on esp, which I have not read, how did you handle this problem? That is, how did you make a distinction between a demonstration of mystic powers and guessing the Zener cards by chance? How did you state things to avoid the improper use of the word chance?

  2. I just checked and your book is held in the university library. I wonder if it was me who put it there? In any case I suppose that I should read it someday.

    The video length is fine by me.

  3. You could try rotating the camera to give a portrait view but you would need to also rotate the image before placing it online. Alternately, raise the camera; step back; or sit down.

    You could hang some sheets or blankets out of view to reduce the echo but the the doors behind you are still going to produce some echo. Maybe use a different room? Or add a sound absorbing background?

  4. I’ve never heard anybody claim that MCMC needs random values but then I live in a small Universe. What’s needed are a lot of uniformly distributed values to avoid bias.

    Bootstrapping also doesn’t need randomness. It needs a lot of uniform mixtures of the samples at hand. Of course, it can’t fix any inherent bias in those samples. I still think a leave-n-out approach gives a reasonable estimate of the model prediction ranges.

  5. I was wondering about the two doors (like the two hands).

    But fantastic, Briggs. Well done.

  6. Matt,

    A couple of thoughts:

    1. Cut the meta-comments (talking about the medium–“Not sure how to work the webcam,” etc.”

    2. This would probably be more effective as a slide presentation turned into a video with narration. Brainshark.com provides a free service to do this.
    a. You would take pictures of the relevant visuals–your two hands, revealing the dime, etc. and insert those visuals on the slides.
    b. You’d also be able to include slides with bulleted talking points.
    c. You’d also be able to include other visuals–pictures, graphs, illustrations, etc.
    d. Doing a presentation with visuals will force you to focus on your talking points. This will excise what may be less necessary points that come up in an ad hoc “lecture.”

    Hope that helps.

    Kent

  7. Is that a Balthus knot or a full Windsor? Come on mann…relax!

  8. Nullius in Verba

    February 23, 2014 at 2:53 pm

    “I’ve never heard anybody claim that MCMC needs random values but then I live in a small Universe. What’s needed are a lot of uniformly distributed values to avoid bias.”

    Likewise, you don’t need randomness for “randomized” double-blind medical trials – you only need independence. The point is to guarantee that the selection cannot be correlated with any confounding factors.

    And everyone in the trade knows the difference between ‘random’ and ‘pseudorandom’. There are RNGs built in to some computers that rely on physical sources like amplified thermal electron noise.

    And the standard interpretations of quantum mechanics all treat the randomness as intrinsic and ontological, not merely an epistemological “unknown”. The idea that outcomes are merely unknown is related to the ‘hidden variables’ hypothesis, which most quantum mechanics regard as falsified by Aspect’s confirmation of the Bell inequalities.

    But it was a nice video. Don’t listen to the critics – all the talk about the production values is entertaining and gets the audience on side, and the suits look sharp.

  9. Ye Olde Statisician

    February 23, 2014 at 4:07 pm

    Ed Schrock, my first boss, used to denigrate random samples. “Trouble comes in bunches,” he said, “so you should take your samples in bunches.” In pulling wax pillows from a carton, it was more important to pull some from the top, middle, bottom, left, right, front, back, than to pull them at random. The purpose of randomization was to get some assurance that the samples do not happen to coincide with some shift or trend in the process. In a study of batch yields versus ingredient ratio, the production supervisor wanted to run four batches at the normal ratio, then four at the high ratio, then four at the low ratio. But it is easy to see how a shift in the process might coincidentally coincide with such a strategy. So we took the projected 12 batches and broke them up into four segments of three batches each. In each segment there would be one N, one H, and one L batch. But to alternate them regularly ran the risk of coinciding with a regular cycle and so accidentally assigning to the ingredient ratio the effect of some extraneous variable. So within each set of three batches, we pulled slips of paper labeled N, H, and L from a hat. The purpose of the randomization is to obtain some modest assurance that the sequence of treatments would not coincide with a shift, trend, or cycle due to some other factor.

    In the test, the yield crashed after the fourth batch. If we had run NNNNHHHHLLLL as the supervisor had initially wanted, it would have appeared as if the yields crashed because we had changed the ingredient ratio. But because we had blocked and randomized, it was clear that ingredient ratio had nothing to do with it. Further investigation revealed that one of the ingredients had been poisoned by water condensate in the RR tank car, and this led ultimately to a solution of the yield problems.

  10. Your personal appearance and presentation are generally excellent. No need to change them much.

    Improvements:
    (a) get to the point much more quickly and definitely — e.g. start with a question;
    (b) have a fairly complete script planned before you start;
    (c) every time you say “Umm”, half your audience will be distracted away by a cat video;
    (d) every time you pause for longer than 1.5 seconds, another cat wins.

    Technically:
    1) use a movable camera (an actual video camera that e.g. focuses properly, never a webcam);
    2) use a tripod;
    3a) never use the mic on the camera;
    3b) use a separate mic, probably a lavelier (the audio could even record into a separate device — leave the video and audio running continually, then sync them later);
    4) pick a nice (but not too distracting) background. E.g sitting at a table with a cup of coffee and a pastry in front of you — the audience is then implicitly just across the table from you, and this helps to draw them in. (And if they start to get bored, they will look at the pastry until you say something interesting again).
    5) have more than one kind of picture. E.g cut between a general view of you and a zoomed-in-to-head view, or cut to some helpful graphic or diagram. Each purposeful cut helps maintain attention.
    6) use video-editing software;
    7) use natural lighting (in your video here, you have a wonderful light source just behind your left shoulder — exactly where it shouldn’t be).

  11. Briggs

    February 23, 2014 at 5:46 pm

    All,

    How about the timing?

    The secret to the suits—and I hesitate to say it—is thrift stores. But you have to be assiduous about it. At the Salvation Army two years ago I found two perfect condition Paul Stewart suits which had to be altered not at all for $20. Total.

    YOS,

    Control. Exactly so. “Some at random” means “Make sure we’re seeing all parts of the box, in case something changes.” But notice that it’s not “random” in the traditional sense, then. It’s controlled.

    Actually, of course, the proof of the non-necessity of randomness is very well known in Bayesian circles.

    NIV,

    It’s not so that all standard interpretations treat quantum unpredictability as ontological and not epistemological. In fact, there is quite a debate about this (see the guest post by Bob Kurland recently, who discussed this; if I remember to do it, I’ll come back and link this). Obviously, I come down on the side that it is at least epistemological; i.e. it is our knowledge of the outcome of experiments which is a matter of (logical) probability. What goes on, what makes this state and not that appear, we not only don’t know, but can’t know, so the ontology is hidden from us.

    And the MCMC algorithms do not require just “independence”. Take the proposition, p = “George the Martian wears a hat” which is not probative of anything, meaning adding it to the evidence does not change any probabilities; i.e. it is “independent”. But adding that to the MCMC soup does not a thing. What’s needed are a bunch of numbers from 0 to 1, spaced uniformly.

    DAV,

    The phrase “uniformly distributed values”, as it is commonly used, means “random numbers from 0 to 1.” Same thing. See comment to NIV.

    Hwan,

    I am perfectly relaxed, hence the (obvious) half-Windsor knot.

    Kent,

    I will not participate in anything to do with slides. Slides with words on them can be replaced by words. Perhaps if I need a picture, then I can cut a picture in. But no slides. The world already has too many! I tweeted once, “I remember when slides used to be things you enjoyed.”

    However, you’re quite right that the production still stinks. I’m working on it.

    Pedro,

    I’m working on finding something more cinematic.

    Scotian,

    Good grief, your poor library!

    Actually, the Psychic book was written before I discovered logical probability, but it’s not terribly far off. I can’t say for sure, but I probably did abuse the language and say “caused by chance.” Been a long time since I looked at it.

    Paul,

    Thanks for the tips. I’m working on ideas of how to improve the technology side.

  12. The timing was fine, in my opinion. The video seemed reasonably well organized. It got to the point quickly enough, without sacrificing important points. In general, it seems 10 – 15 minutes and one important concept, well illustrated, are suitable per video. I await the next one eagerly.

    Prior comments have covered many of the points I wanted to address; I was in church, so I’m a little late to the party. To reinforce a couple of points:

    1. Cheap and easy way to improve the sound immediately (regardless of microphone type or placement): place a rug and/or heavy blanket on the floor under you, and extending all the way to the camera/mic. That will kill a lot of the reflection that is causing the hollowness in the sound.

    2. Try placing a lamp at either the 10 o’clock or 2 o’clock position with respect to the camera lens, which shines in your direction from behind the camera position. A nice, moderately bright, preferably soft light source. A Chinese lantern would be ideal.

    3. If you don’t have a suitable lamp, see if you can set things up so a nice, bright window is in the same position – but one with skylight, and not direct sunlight streaming through it. This kind of light is regarded by many a portrait photographer as nearly ideal.

    4. Not sure if you are able to fine-tune the framing, but a wee bit more head room would lend it a better composition.

    Thank you for the videos, and your blog! They are most illuminating, and I enjoy learning from them.

  13. Don’t change a thing Dr. Briggs. I read you loud and clear and to me that is what matters most.

  14. Waste of space doing stuff like this until you can get decent audio quality (it sux) and at least a vaguely professional video setup.

    What you say might be great, but no-one will be viewing…

  15. You kept looking at your right hand as you presented the problem. That made me guess the coin to be a “1”.

    Like the rest of the folks here, I’m enjoying these weekly videos as well.

  16. The phrase “uniformly distributed values”, as it is commonly used, means “random numbers from 0 to 1.”

    Yeah, it does, but I meant it literally. I think of it as laying down a mesh and keeping it from wrinkling or bunching up. The only reason behind “random” is because it would be nice to not have to use all of the mesh every time and to avoid the same problem YOS mentioned (for Monte Carlo integration, anyway, MCMC samples are correlated).

  17. Nullius in Verba

    February 24, 2014 at 3:32 pm

    “Yeah, it does, but I meant it literally.”

    It doesn’t, actually. A discrete uniform distribution can be defined on any finite set: it requires that every element of the set has equal probability.

    (There is also a continuous version with the probability of the outcome being in any subset proportional to the subset’s volume, but I know the professor doesn’t like continuous distributions to be mentioned on his blog, so we’ll say no more, eh? 😉 )

    It so happens that a lot of computer languages implement an approximation to U(0,1), because it’s relatively easy to transform that into whatever other distribution you want. But uniform distributions are a little more general.

  18. Briggs

    February 24, 2014 at 3:43 pm

    NIV,

    Oh, I love continuity; nothing like it for spending infinite amounts of time. But how one approaches it matters. Simply plopping down distributions (for use in some analysis) as most do, when their objects of interest cannot be measured or acted on to infinite precision, is to be greatly discouraged.

  19. The presentation content was very good. There was some repetition, but that may be OK for some people. You were dressed very well. Some may say you were overdressed. Love the blue suit. Was that French cuffs?

    There are a number of solutions to the video framing, i.e., not cutting off your head. You could step back some, or sit on a stool. Perhaps a new camera with zoom capability would solve that problem.

    You might think of a different background, i.e., curtains or drapes. The doors are OK, but lend a bit of a homemade feel to the video.

    I am not an expert on videos, but in my career I made hundreds of presentations. You look great and deliver you message very well.

  20. Oh, one more thing. You need to do something about the lighting. You are backlit in this video.

  21. I know some are poo-pooing the presentation, but I don’t care much about the presentation. The content is great. I’m finding these much easier to understand than your writing on the probability. I find this stuff fascinating and there’s really nowhere else to get it.

Leave a Reply

Your email address will not be published.

*

© 2016 William M. Briggs

Theme by Anders NorenUp ↑