What is this series

Discussion in 'Scientific Statistics Math' started by David W. Cantrell, Feb 16, 2006.

  1. +++++++++++++++++++
    I fail to see any productive value out of this thread.

    I have done a lot of testing of random number generators and am familiar
    with the various tests. Neither of the two protagonsits have identified the
    tests that are applicable, nor have they defined what constitutes a
    "acceptable random number generator". My testing was specific to the two
    Excel random number generatores (i.e.2000 and earlier and the 2003 one). I
    did the testing using the Diehard suite and B.D. McCullough did the tests on
    the 2003 version using one of L'Ecuyer's sequences. My paper is on the
    Internet. Look for "Excel Faults and Errors".

    The dialog so far has totally bypassed anything that would represent an
    accepted standard for for those doing Monte Carlo simulations.

    DAH
    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    Non-random behavior is really meaningless unless there is some sort of
    measure as to what non-random means and how it is determined.
    ..
    DAH
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    That's not universally true. I have combined the bad 2000 RNG with a good
    one (using the 2000 RNG as the bad one) with one of Knith's reccommended
    Lagged Fibomacci generators and the results were worse than the Excell RNG
    by itself.

    Combining computer RNG's with electronic noise generators was the vogue
    several years ago. The problem was that the electronic noise was just on the
    borderline of showing significant serial correlations, and the combinations
    had this same problem. I would imagine mechanical devices would have the
    same problem.

    DAH
    ++++++++++++++++++++++++++++
    ++++++++++++++++++++++++++++++
    Most of the RNG's used today generate a new value from the previous value as
    a seed. Marsaglia's current RNGs use a vector rather than a single value.
    (The Excel 2003 RNG uses a previous set of 3 linear congruential RNs to
    generate a different linear conguential value which is output. The 3 values
    are internal and are not under control of the user.) Assuming one is using
    the traditional single seed to generate a new RNG based on 32 bit registers,
    the sequence of (1, 2, ..., 366) would be impossible to generate this
    sequence in the computer, unless there was a hidden program/fault the onlly
    allowed this sequence to appear. Considering the newer RNGs, I would still
    say that this sequence would be impossible to generate. If there was any
    appearance of regularity, I would discand the values and the RNG being used,
    as being faulty. It would be impossible for the Excel 2000 RNG to generate
    this sequence.

    Anybody who is in this area should be familair with Knuth's "The Art Of
    Computer Programming", Volume 2, Chapter 3, "Random Numbers".

    DAH
    +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    +++++++++++++++++++++++++++++++++++++++++++++
    Again, these sequences would be impossible to generate from a valid RNG,
    that has passed Knuth's basic tests.

    DAH
    +++++++++++++++++++++++++++++++++++++
    +++++++++++++++++++++++++++++++++++++++++++
    One of the basic issues not tackled is just what constitutes an acceptable
    RNG for a given task. The next question is just what set of tests must the
    RNG pass to be acceptable. There is some general agreement that the NIST
    suite is inadequate, Diehard # 3 is acceptable, and for serious work, test
    from L'Ecuyer's package are required. An essential problem is the
    arbitrariariness of deciding if a test is passed (based on one sequence or
    on repeated sequences). With Excel 2000 I can get it to pass the 3-D test
    every so oftem. By selection, I can say that the RNG passes the 3-D test.
    However it fails the Monkey tests every time. Is passing the Monkey tests
    essential for Monte Carlo work.?

    DAH
    +++++++++++++++
     
    David A. Heiser, Mar 14, 2006
    #61
    1. Advertisements

  2. David W. Cantrell

    Herman Rubin Guest

    As you will see, I believe it is because you do not see
    all the problems involved.
    The tests which are applicable are the ones which indicate
    whether the random numbers have the appropriate properties
    for the current problem.

    nor have they defined what constitutes a
    See the above. A random number generator should produce
    an independent sequence of bits, each of which is equally
    likely to be 0 or 1. NO pseudo-random generator can do
    this. It is possible to come quite close with physical
    random numbers by using a large amount of condensation.

    My testing was specific to the two
    One of my colleagues had a sign on his door, stating that
    friends do not let friends use Excel.
    I have often stated that I would never use pseudo-random
    numbers alone for "production" work.

    Also, in his early paper, von Neumann stated that someone
    who even thinks of using machine-generated numbers (using
    an algorithm) for random numbers is in a state of sin.
    I do not know how you were combining them, but they are both
    pseudo-random. One has to be physical for my statement to
    even be considered.

    Also, lagged Fibonacci generators were found to give bad
    results in an Ising model problem. They were then found to
    be a form of multiplicative congruential generators over an
    appropriate finite ring, and so "fall mainly in the planes".
    "Quick" physical generators are not too good, but if one
    uses a fairly good pseudo-random generator, the weaknesses
    will not match up. If it is randomly seeded, the local
    properties will behave randomly, and the global dependences
    of the physical and pseudo are very unlikely to match up.

    The method of combining I suggested is to use exclusive or,
    or integer addition discarding carry. If it is necessary
    to reuse the physical numbers because of size, try to make
    their length as "irrational" as possible compared to powers
    of 2.
    The vector should be as long as the internal size of the
    block needed to generate, and should be under the control
    of the user. To get good results from the beginning, it
    should be as random as the user can make it.
    This is already obsolete. Note that a generator he recommended
    highly is now known to have serious weaknesses.
    If the sequence is impossible, the generator is not random.

    The latest of Marsaglia's generators could do it, at least
    if one used it as a bit sequence. If one uses an efficient
    method of generating a permutation, the number of bits
    needed will be on the order of 2500, or on the order of
    80 32-bit words.
    And as I have stated, this cannot be decided that easily.

    The only way to really decide it is to do that task repeatedly,
    and see if the probability distribution of some of the known
    results match what is found. This clearly cannot be done.

    The next question is just what set of tests must the
    It depends on the Monte Carlo work. If the random numbers
    are used one at a time, or a fixed number at a time, to
    generate terms to be added, I would suggest using quasi-random
    numbers instead; these do not simulate random, but do simulate
    uniform, which is often needed.

    Numerical analysis is an art; it uses science, but requires the
    use of human, not machine, judgment.
    +++++++++++++++
     
    Herman Rubin, Mar 14, 2006
    #62
    1. Advertisements

  3. David W. Cantrell

    Reef Fish Guest

    For one thing, you came to the party when it was ALREADY OVER.
    You merely rehashed some resolved points with your own prejudices
    and ignorance about Monte Carlo methods! In a very REDUNDANT manner.

    The topic of discussion was about "randomization" and how a
    statistician
    who insists on randomization would behave CONDITION UPON a
    "satistifactory"
    random number generator (whatever it is) having generated some patterns
    that appeared to be too "nonrandom". SEE BELOW. It's all there, in
    previous discussions!

    ALL your comments are now irrelevant and impertinent, but for the sake
    of
    backtracking to some points covered, I'll comment briefly on those I
    find
    WRONG.

    Why don't you look at those listed by the National Bureau of Standards
    in
    their handbook of Mathematical Functions, or another few dozens in more
    recent published papers and books on Monte Carlo. From what you
    posted,
    it appeared that you are VERY, VERY ill informed. You seem to know
    one
    of Knuth's books and nothing else.

    Excel is arguably the worst of all software ever WRITTEN or USED for
    statistical computations. You tested it with ONE test and talked as
    if you
    knew something about random number testing or Monte Carlo methods?
    Are you sure you didn't misspell "DUH"? :)

    The above was the statement I referred to as having been the ISSUE
    under discussion all along. NOT the process of random number
    generation
    itself.
    Now THAT is a patently FALSE statement. It is so patently absurd that
    it is laughable!

    If generators that passed Knuth's basic tests CANNOT produce
    certain random permutations, then that generator is, by DEFINITION,
    a defective one!

    I think you are very much CONFUSED about the sequence of random
    digits that are produced by a generator with HOW a random permutation
    can be generated by taking a SEQUENCE of random NUMBERS (of
    arbitrary number of digits) and order them to produce a random
    permutation of N integers.

    Any such test acceptable by a group of discussants OR any test
    acceptable
    by ONE person, such as you (and nobody else) or Herman (and nobody
    else).
    It doesn't matter. The question is what do you do with (1, ..., 366)
    or any
    of the other patterns nearly everyone would NOT use.

    You are being very redundant about your irrelevant point. Ask your
    god
    Knuth what generator he would use, and that's good enough for ME, for
    THIS discussion. It would not, in Monte Carlo methods discussions in
    general.

    DUH, DUH, ..., DUH. <Note no more smileys>\

    -- Bob.
     
    Reef Fish, Mar 18, 2006
    #63
  4. David W. Cantrell

    Jerry Dallal Guest

    There was contest work as part of 2005's Black History month in which
    Art Kane's "Harlem 1958" was displayed.
    http://www.artkane.com/fullsize/harlem/fzharlem.html
    People were given a sketch of the picture with the subjects as outlines
    and each outline numbered. They were also given a list of all of the
    names of the musicians in the picture. The one who could do best at
    attaching the numbers on the outlines to the names on the list would win
    a $50 gift certificate to a local restaurant.

    I don't know much about modern jazz or what was going in in Harlem in
    the 50s. However, I know a reasonable amount about popular music of the
    20s and 30s, where many of the performers in Kane's picture got their
    start. As, I started filling in the easy half dozen, I began to notice
    a pattern--the silhouettes appeared to be numbered in the same order as
    the names on the sheet contains the names of the musicians!

    I took the organizer aside and told her of my suspicion, which she
    confirmed. I decided not to submit my answer sheet and the prize went
    to someone based on her ability to identify the musicians.

    As far as everyone else was concerned, the names might as well have been
    listed in random order. However, for me their order had the potential
    to undermine the game.

    Anyone who understands the reasons we randomize and sees the point of
    this (true!) story knows what I would do with the random permutation
    1(1)366 and why. Even more, they know why there is a trick in the
    previous sentence!
     
    Jerry Dallal, Mar 19, 2006
    #64
  5. David W. Cantrell

    Reef Fish Guest

    Jerry, your amusing anecdotal account stood untouched for 3 days --
    which prompted me to at least comment on the fact that it had (to me)

    ABSOLUTELY NOTHING to do with the question of "randomization"
    being discussed, other than an unrelated similarity. :)

    When I started the SUBTHREAD of "randomization" 3 weeks ago, I wrote

    RF> I am now refreshed from a few dives down below 200 feet
    RF> during the Mardi Gras festival in Cozumel. Perhaps the
    RF> example above about a random permutation turning out to
    RF> be 1, 2, ..., 365, 366 is an easy starting point to discuss the
    RF> subject of randomization -- that is, what do statisticians DO
    RF> with random outcomes which "look" nonrandom?

    I am now refreshed from a few days (22 miles NORTH of North
    Pole <G>) where as recent as a week ago the temp was minus 40F
    (without wind) to discuss the same problem, having resolved with
    others that the question was CONDITIONED on the generating
    process being a RANDOM one (see proir discussion), in my
    attempt to relate the discussion to your anecdotal account.

    But THEY WEREN'T! It was the carelessness of the organizer who
    posed the question in the form posed.

    HAD the names been RANDOMLY ordered, and turned out the way
    you noticed (from the non-randomly assigned pattern), it would have
    been a different issue altogether!

    Thus, IMO, you made TWO technical errors. :)

    1. You should have asked the organizer if the order of the names were
    RANDOMLY assigned. If the answer was "yes", then you may (or
    may NOT) take your chance that the OTHER names would follow
    the same pattern!!! (if they were randomly assigned -- you have
    only looked at a SMALL subset of the entire list) and submit your
    "solution" to see what happened.

    2. You used the anecdote in a manner that is NOT particularly
    pertinent to either the equal-probability on all 366! permutations
    theory, nor the conditional INFERENCE based on a small
    subset of matching pattern to draw inference on the ENTIRE
    sequence, CONDITIONED on the (secondary) KNOWN fact
    that the permutation was indeed randomly generated!

    I believe I DO.
    I am afraid I DON'T.

    I DON'T but eagerly look forward to your answer, especially the "why".

    Since I've known you to write tricky sentences without trying, on
    several
    previous occasions <G>, I equally look forward to the enlightment of
    your connudrum.

    -- Bob.
     
    Reef Fish, Mar 21, 2006
    #65
  6. David W. Cantrell

    Jerry Dallal Guest

    Au contraire! It has EVERYTHING to do with randomization.

    I didn't want to win that way and I didn't want to have her go through
    the bother of having to name a second winner after I passed on the prize.
    I have no reason to doubt it.
    Think about why we randomize.
    If you know what I'll do, then you don't know what I'll do, unless we
    are talking about a specific problem.
     
    Jerry Dallal, Mar 21, 2006
    #66
  7. David W. Cantrell

    Reef Fish Guest

    I am all ears. But your subsequent rhetoric failed to relate it to
    the original
    question of what you would do if an agreed upon random process
    generated
    (1,2, ..., 366). <the original problem was in a draft lottery, but
    that's
    immaterial on what one would do in the randomization theory/practice>
    Jerry, you not only write ambiguously, you don't read well. I said
    "IF the
    names were RANDOMLY assigned, than you should submit your entry,
    because you have only observed the "coincidence" of match order in a
    small subset.

    Of course the actual answer you got that they were NOT randomly
    assigned, then your action was proper.

    In the RANDOM assignment case, how would YOU know that the rest
    of them would follow the same pattern? That was my SECOND point.
    I have already thought about it even before you were born. :)
    You rhetorical response is not shedding ANY light on your peculiar
    rhetoric
    in the first place.

    Your silence is deafening.
    Quite beating around your rhetorical bush. Suppose Bush decides to
    use a random permutation for an Iraq draft lottery, and he has chosen
    a random generation device that meets Jerry Dallal's approval. Out
    comes the perverbial (1, 2, 3, ... 366), Do you think that outcome
    should be kept or thrown out?

    NOW you have a specific problem <though immaterial>, how do you
    answer the same questions you evaded -- and my first POINT which
    you misread?

    -- Bob.
     
    Reef Fish, Mar 22, 2006
    #67
  8. David W. Cantrell

    Jerry Dallal Guest

    If those in charge decided on a procedure, then the outcome is what the
    outcome is. One would hope that the sequence would be generated in
    public. If one could demonstrate that 1(1)366 was the result of the
    agreed upon procedure and not due to chicanery, then that's the outcome
    because it would be chaos otherwise.

    However, this is more a legal problem than a statistical problem and in
    some ways a distraction.
    I *did* read too quickly, but I asked if the silhouettes and the list of
    names were in the same order because I expected a straight answer to
    that question. I'm not so sure what answer I'd've gotten in response to
    whether the names were ordered at random if, in fact, they were as I
    suspected. Therefore, if I were told that they were in random order, I
    would still not have submitted my answer sheet. If I could have been
    guaranteed a truthful answer, I would have submitted it.

    However, were I running the contest and my random sequence came out this
    way, I would have rerandomized. I am under no legal obligation to stick
    with a list ordered in the same way as the silhouettes and, in some
    ways, the principles behind randomization demand that I rerandomize.

    This gets back to why we randomize to begin with.
     
    Jerry Dallal, Mar 22, 2006
    #68
  9. David W. Cantrell

    Reef Fish Guest

    Kept. You sure are wordy. Then we are the only two (so far) who
    said we would keep that outcome, given the random generation agreement.

    I am still waiting for someone to explain WHY they would throw it out.
    Then they would throw zillions of others out too. Complete chaos,
    but it seems to be the state of chicanery in the randomizers.
    Not a legal problem at all. Once a random generation mechanism is
    agreed upon, it's a no brainer. If ONE is thrown out, then there is
    no
    end to the argument that others are thrown out whenever anyone
    finds some pattern that looks "non random", and THEN you have
    lots of legal problems.
    That makes no sense whatsoever, logically, ethically, and otherwise!
    Why should you have to second guess everything? In fact, I might
    go so far as to say that if you HADN'T asked any question, it would
    have been perfectly ethical to turn in the systematic answer -- as
    someone who had noticed the pattern caused by the stupidity of the
    problem poser -- everyone ELSE has the same opportunity to
    recognize the same pattern.
    WHAT "principles behind randomization" that demand you to throw out
    a perfectly legit randomized outcome?

    WHY????? You have already randomized the FIRST time. You are
    now speaking from both sides of your mouth! Earlier you said you
    would keep the outcome that looked completely systematic; and now
    you are talking trash that is completely counter to your earlier
    position.

    No, that gets back to you either (a) believe in randomization (in which

    case you are obligated to keep your randomized outcome no matter
    what) or (b) you don't believe in randomization -- which is what you
    did when you did not keep what you generated, by a random process.
    There is no end to THAT chaos. That's NOT randomization at
    all -- to keep or throw out a randomized outcome as a function of
    your whim!

    So, the score stands I am still the ONLY one who would keep the
    non-random looking pattern that is generated from an agreed-upon
    random process.

    I simply don't understand the reason behind your straddling the fence
    and speak out from both sides of your mouth.

    -- Bob.
     
    Reef Fish, Mar 22, 2006
    #69
  10. David W. Cantrell

    Jerry Dallal Guest

    Bob,

    Why do *you* randomize? If you don't randomize(*), why do you think
    others randomize?

    --Jerry

    (*) If you don't randomize, what would you do instead of a randomized,
    double-blind, controlled trial?
     
    Jerry Dallal, Mar 22, 2006
    #70
  11. David W. Cantrell

    Jerry Dallal Guest

    Bob,

    It's not necessary to respond to this. I *do* have a tendency to read
    Usenet too quickly, especially if I try to dash something off before a
    full day of meetings. (I look forward to joining you retired guys in a
    few years.) I reread it and see your "out", or maybe it's the first
    time you said it exactly that way, I don't have time to go back.

    You wrote: "So, the score stands I am still the ONLY one who would keep
    the non-random looking pattern that is generated from an agreed-upon
    random process."

    I disagree that you are the "ONLY one". I'd say that it could apply to
    almost everyone, if only they'd think about it. The out is
    "agreed-upon". I would argue that, when pushed, most people are
    actually doing some form of restricted randomization, so, in many cases,
    1(1)366, is not part of the restricted set. That's why I agree with
    Herman that keeping 1(1)366 depends on the trial (and whether 1(1)366 is
    in the set). That's also why if you know what I'd do and why (It depends
    on the problem at hand!) means you don't know what I do (Because it
    varies according to my set of restrictions) except in specific cases
    (Because then I must specify my set, if only in broad terms). Once the
    restricted set is conceptualized, then we MUST do what you propose,
    namely, stick with what we get.

    This describes what we do, but still avoids what for me is the main
    issue--*Why* do we randomize?--but that's another topic.

    --Jerry
     
    Jerry Dallal, Mar 22, 2006
    #71
  12. David W. Cantrell

    Reef Fish Guest

    It goes without saying. But I think it's good to clear the air on any

    point of misunderstanding that could be clarified, easily or not.

    Comprendo. But keep in mind that there's no urgency in this or
    any other discussion. Bob Dole waited three days before he
    was the first to plunge in. I waited three days before I responded
    to your post, not having seen anyone else doing so.

    I can't say that I look forward to your new excuses. :) "Retarred"
    (as we say in the South) or not, there's no excuse for sloppy
    reading and citing remarks out of context!

    No. It was the n-th time, where n is an integer greater than 6. :)

    That was no "out". It was assumed to be implicitly understood at the
    beginning; but then when Herman and Heiser went on the tangent
    of arguing about what constitutes a satisfactory or "agreed upon"
    random number generator, i stated EXPLICITLY, many times, as
    I stated to you, that ANY pseudorandom/random number generator
    that is agreeable to Herman (even if he is the only one), or Heiser
    (even if he is the only one), or you, would do, for the present
    discussion of what one would do IF such a generator generated
    a non-random looking sequence.

    If was always the CONDITION upon which one's action (as one
    who insists on "randomization') was discussed or taken into
    consideration.

    Otherwise, we could spend the rest of our lives arguing about what
    random mechanism to use, while leaving the REAL issue untouched.

    Another quote out of context. I DID spell out that I was the ONLY
    one
    (among those who responded to the question in this thread). You
    should know that I am not quite THAT dumb to make any global
    statement like that. As a matter of fact, I would HOPE that there
    are many NOT in these ngs or in this discussion who hold the same
    view as I do, on randomization, that you either believe it or you
    don't!

    Bob Dole was the first to admit that he would not keep the
    deterministic-looking sequence. Herman was next, after I pinned
    him down to the generator HE prefered. There were another one
    or two not in the extended discussion who did the same.
    As I explained above, BOTH the "ONLY one" and the "out" were
    the result of your own misreading.

    That's where Fisher (remember him <G>) come into this discussion,
    and why randomization is a non-issue in Bayesian inference a la
    Savage and others.

    -- Bob.
     
    Reef Fish, Mar 22, 2006
    #72
  13. David W. Cantrell

    Jerry Dallal Guest

    I've decided I'll snip no more, since it often causes problems. For
    example, I view posts in the context of their threads. I took and
    intended "ONLY one" to mean only one in this thread. Unfortunately,
    since you didn't write say "in this thread" yourself, I couldn't quote
    you that way. I suppose I could have written "[in this thread]", but
    it's easier to not snip.

    I believe that most posters in this thread would agree with you IF they
    took the trouble to see the (proper) importance you place upon the
    descriptor "agreed upon". Who's going to say, "I agree that this is
    what we'll do" and then not do it? Who was holding a gun to their
    heads? That's why I'm disappointed that you snipped my comments about
    restricted randomization. Give me a problem and I'll tell you what I'll
    accept and why, before I generate the random sequence.

    I believe the reason people would reject 1(1)366 in some contexts and
    not others is because they are really doing restricted randomization,
    but can't articulate it. Sometimes there are good a priori reasons for
    excluding 1(1)366; other times, there aren't. However, it's hard to
    find generators that do that sort of thing. So, they use generators that
    generate anything, recognizing that they may have to step in if
    something "bad" happens.

    Perhaps you'd like to start another thread with a different subject
    header, because I really am curious to know why you randomize or, if you
    don't, why you think others randomize.

    --Jerry
     
    Jerry Dallal, Mar 22, 2006
    #73
  14. David W. Cantrell

    Reef Fish Guest

    Jerry, it's more than "snipping" -- it has to do with reading a post in

    the CONTEXT of the thread, of what should have been obvious to
    anyone who reads the entire (or most of the) thread.

    But more importantly, had you NOT seen it fit to disagree with "the
    only one" (which was a FACT, based on this thread), I would not
    have had to point it out to you that you were quoting OUT of context.
    Did you know of ANYONE who said s/he would KEEP the 1(1)366
    other than ME and you who said it earlier, and then retracted it?

    If not, then you had no POINT. To make your comment relevant,
    I drew the inevitable inference that you must have thought I meant
    something else other than in THIS thread.

    Point #1: NONE of your fictitious "most posters in this thread" said
    it.
    Now you MISSED Point #2.

    The agreement was on the RANDOM NUMBER generator or METHOD
    of generation. The ISSUE of discussion is conditioned on that agree-
    ment on the generator being acceptable, but the different ACTION
    different folks would take.

    Two SEPARATE issues.

    Bob Dole admitted he would discard it. Herman, after the understood
    condition about the generator being acceptable, STILL found the
    1(1)366 unacceptable. One does NOT follow from the other, as you
    seem to imply.

    Jerry, I am sure you had another meeting waiting for you when you
    dashed your post -- but you keep on MIS-READING what had been
    said by ME, as well as what had been said by OTHERS, to have
    arrived at your "conjecture" that was not supported by any of the
    posts in THIS thread, other than your own careless reading!

    Nobody that I know of -- on Bob Dole or Herman Rubin, to mention
    only those two.

    The ONLY paragraph of yours I snipped was prefaced by my comment
    that it resulted from your FALSE PREMISE. YOu didn't explain
    anything.
    You derived something from a false premise -- I gave you the specific
    problem of draft lottery for Iraq, and you rejected the 1(1)366 whereas
    you had accepted the same before. You were simply contradicting
    YOURSELF -- all in a single post.

    You are not exactly a good one to tall others that they don't
    articulate
    themselves. I have NO PROBLEM whatsoever understanding what
    they say. I have a terrible time putting together what YOU are TRYING
    to say, because you mis-read, mis-quote, and you contradict yourself
    on such a simple question as what would you DO with a randomly
    generated sequence. NOBODY but you made something out of
    nothing. Did Fisher say something about throwing away some
    randomized outcomes that he didn't like the looks?

    SUCH AS?

    Then you had better list the other trillions of permutations that are
    undoubtedly as objectionable for whatever reason you gave -- and
    even using your OWN WORDS:

    JD> If one could demonstrate that 1(1)366 was the result of the
    JD > agreed upon procedure and not due to chicanery, then that's
    JD> the outcome because it would be chaos otherwise.

    which I readily agreed, especially about the CHAOS part, which
    I had previously called, ill-defined, and undefined restricted sets.

    RF> Kept. You sure are wordy. Then we are the only two (so far)
    who
    RF> said we would keep that outcome, given the random generation
    agreement.

    Generator do NOT generate "random permutations". Random permutations
    are obtained by applying special RULES to get random numbers from the
    generator and THEN determine the random permutation from it. You are
    making the same ERROR as did David Heiser, about random permutations.

    I made my position very clear. If I DO decide to randomize, then I
    don't fudge the outcomes of randomization. Pure and simple. If
    I DON'T randomize, then the question of what I do is moot, but my
    OPINION remains that there shouldn't be selection chaos out of
    the whim of the randomizer.

    I even used Savage's example of a random throw of blocks resulted
    in the spelling of "C H R I S T M A S". Shoudl that be in your
    resctricted set? If you think so, you should rethink again! Just
    think of the number of different ways 'christmas" is spelled in
    different
    language, not to mention other equally recognizable patterns.

    -- Bob.
     
    Reef Fish, Mar 22, 2006
    #74
  15. David W. Cantrell

    Jerry Dallal Guest

    Bob,

    For "random", substitute "pseudo-random" where appropriate.

    --Jerry
     
    Jerry Dallal, Mar 22, 2006
    #75
  16. David W. Cantrell

    Reef Fish Guest

    Jerry, you are on a ROLL today! Your committee meetings must be
    too short and boring.

    Thank you for your correction, without which I am sure the other
    readers of this thread would be hopelessly lost.

    For your contribution in this post, I hereby nominate you for the
    No-Bell
    Prize of the N-th degree Nit-Picking Award of the Year. <:-o)

    Only earlier TODAY, I had said to Jerry,

    RF> I stated to you, that ANY pseudorandom/random number generator

    Don't you think that's better than just pseudorandom or random?

    Should I have clarified it further by explicitly excluding quasirandom,

    and hemisemidemipseudoquasirandom?

    -- Bob.
     
    Reef Fish, Mar 22, 2006
    #76
  17. David W. Cantrell

    Herman Rubin Guest

    I would never use pseudo-random for randomization. There
    can be no assurance that pseudo-random has the desired
    properties.
     
    Herman Rubin, Mar 22, 2006
    #77
  18. David W. Cantrell

    Jerry Dallal Guest

    Bob,

    You are welcome. I am happy to provide your light entertainment, even
    if it's only by agreeing with you. However--and you can tell me I
    missed this, too, but since it's the only item I should see it this
    time--WHY do you randomize? I know WHAT you do when you randomize, and
    I agree completely, but WHY do you randomize?

    --Jerry
     
    Jerry Dallal, Mar 22, 2006
    #78
  19. David W. Cantrell

    Reef Fish Guest

    That's a BIG subject, but I'll give you a tiny example, of what had
    taken place dozens of times -- and the actual reason of WHY, on
    the random permutation of FOUR numbers.

    Students in my Data Analysis courses are required to make 10
    minute presentations of their projects in which they have labored
    for weeks during the course, in the same setting as presentations
    of contributed papers at Annual Stat Meetings. Since the ORDER
    of the presentations may have some effect on how the
    presentations are PEER-graded by the other students,
    randomization was always used so that each of the 4! or 24 orders
    are equally likely and students could only blame the order they
    get on LUCK of the draw. (For the same reason that the
    ORDER of performance of competitors in Intervational Violin
    and Piano competitions are always randomly drawn).

    You may note that the order (1 2 3 4) has the same chance
    of occurrence as any other of the 23 permutations. Nobody
    ever complained when (1 2 3 4) was actually drawn on some
    of those occasions!

    That should give a sobering through whenever you are
    tempted by the non-Bayesian Devil to throw away any of
    your randomized outcomes.

    Now is I had 366 students ... the analogous reason should
    be obvious WHY you don't throw away any of the actual
    permutations drawn.

    -- Bob.
     
    Reef Fish, Mar 22, 2006
    #79
  20. David W. Cantrell

    Jerry Dallal Guest

    NOW, we've got something worth discussing!

    I don't disagree with anything you say, but there is a lot more to be
    said. Not at this moment, as I really do have business to attend to,
    and this discussion deserves care and attention. As you said in an
    earlier post, there's no reason why it can't wait a few days.

    About 25 years ago, I was sitting at my desk getting ready to prepare a
    randomization scheme for a pilot study and found myself wondering,
    "Why?" It was a small study that would never see the light of day and I
    had no biases. Why not just say, haphazardly, "You, you, you: A; you,
    you, you B"? Never being too embarrassed to ask an honest question, I
    walked into Herb Robbins's office and posed my question. He kindly
    pointed out what I was doing by randomizing and I became enlightened.

    I'll pick this up again this weekend.

    --Jerry
     
    Jerry Dallal, Mar 22, 2006
    #80
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.