Discussion: View Thread

  • 1.  [MG-ED-DV] Bacal's Howl's

    Posted 05-06-2000 16:09
    > Robert and others,
    > I have been away from the bench a few days but hope it is not too late to
    > address some of your grievances. Pls see embedded comments below.
    >
    > ----- Original Message -----
    >
    > > There is NO WAY (repeat sotto voce -no way) that you can justify
    > > or support applying a normal distribution to anything but randomly
    > > selected populations. The only way this makes any sense is if you
    > > hire randomly, completely randomly.
    >
    > [JR]First, it seems to me that you confuse a) the method of sampling a
    > population with b) the characteristic distribution of the population.
    There
    > are many ways a population can evolve to a normal distribution regardless
    of
    > how the
    > population is observed.
    >
    > > If you actually do have a normal distribution underlying employee
    > > performance, your hiring practices would be absolutely broken.
    >
    > [JR]Do you have a rationale for this seemingly nonsensical statement?
    > And notice that the distribution does not "underlie" employee performance,
    > it is simply a characterization of the spectrum of employees performances.
    > It is not a
    > foundational or driving factor but only a view of the spectrum of employee
    > performance appraisals.
    >
    > But more importantly, notice that the discussion was not about
    performance.
    > It was about appraisals of performance. We hold open the possibility
    that
    > the appraisal is far from a true estimate of actual performance.
    >
    > > Since the assumption about normal distributions is wrong, so is
    > > everything that follows.
    >
    > [JR]If you get beyond the statistics of management and HR of management
    and
    > get on to the cybernetics of management then you can comprehend that the
    > issue is not about the initial assumption but about a process for
    converging
    > to true appraisals of performance by appraisal writers. So the initial
    > assumption can be dead wrong and everything that follows can be quite
    useful
    > . If you understand self-correcting systems an assumption can be wrong
    and
    > the outcome resulting from that assumption can be quite illuminating. It
    is
    > called testing the hypothesis. Didn't they teach that in your school?
    >
    > > Lest you don't understand the concept here: If you take all the
    > > heights of players in the NBA, do you think you will get a normal
    > > distribution? Or weights of people who go to weight watchers?
    >
    > [JR]I have no reason to believe that the spectrum of the height of NBA
    > players
    > should approximate a normal distribution. In fact, I would expect a
    Poisson
    > distribution. More importantly we should note that this is the old Red
    > Herring trick. However, I do not mind a hypothesis that a normal
    > distribution describes the heights of NBA players as long as you test the
    > hypothesis.
    >
    > Enough of 7th grade statistics. Let's focus on performance appraisals.
    >
    > Although I am not in favor of individual performance appraisals I will say
    > that if such are to be done then the appraisal should address the fit
    > between a) the challenges of the job and b) the contribution of the
    > individual. A
    > performance appraisal is not a recitation of activity. It is a measure of
    > variance from expectation. The variance can be either positive or
    > negative. And we keep in mind that this is the actual performance
    variance
    > compounded with additional errors on the part of the appraiser (which may
    be
    > either positive or negative, as well).
    >
    > Also, I will say that the Quality of Performance Appraisals should be a
    > major factor in the appraisal of any person authorized to write appraisals
    > of others.
    >
    > So for each individual we have a job demand vector and an perceived
    > achievement vector. If each employee in the group achieved exactly 100%
    of
    > the job demands there would
    > be no normal distribution of group appraisal results. Likewise, if each
    > employee in the group achieved exactly 50% of their job demands there
    would
    > be no normal distribution. But it is likely that some employees
    > overachieved and some underachieved. So we get a distribution of
    variances.
    > Whether this distribution is normal or not is irrelevant. Likewise,
    whether
    > I assume (make a hypothesis) it is normal is irrelevant -- as long as the
    > hypothesis is tested. After all, we are concerned here not with improved
    > employee performance but with improved accuracy of appraisals.
    >
    > We should note that in the case where 100% achievement is demonstrated
    very
    > little learning will likely transpire in the future because there is
    > apparently little error
    > being made so no basis for learning.
    >
    > So now we can ask what distribution of variances a good manager would like
    > to see.
    > + All employees performing at 100%? No future in that.
    > + All performing at 130%? Nice job by the employees but the manager has
    to
    > be
    > appraised negatively for spending money on competencies that obviously are
    > not needed.
    > + All performing at 50%? Not quite prudent from any viewpoint.
    >
    > So let's assume that in the ideal case we would want performance to
    average
    > about 90%. Can we decide what spectrum of variance would be ideal? Sure
    we
    > can. No one comes to work with identical learning style, enthusiasm,
    > proficiency, etc. every day. Some days you eat the bear and some days the
    > bear eats you. Aha! The makings of a normal distribution.
    >
    > If each person exhibits a normal distribution over time regarding
    individual
    > performance what would the group distribution look like?
    >
    > Now do you want to howl that the distribution will not be exactly normal?
    > Goodness, then howl if you must --- while the rest of us manage.
    >
    > The central fact remains that very few people are qualified to write
    > appraisals on others but if the HR system makes it happen we should focus
    on
    > the performance of the writers, first, and the performance of the worker
    > bees, second.
    >
    > And to those who are tired of this topic I apologize for the long posting.
    >
    >
    >
    >


  • 2.  [MG-ED-DV] Bacal's Howl's

    Posted 05-06-2000 22:05
    Jack, it seems to me you are obscuring the issue here. It's simple.
    you suggested assuming a normal distribution as part of a process
    you cited (and suggested). I indicated this was nonsense, and you
    also mentioned that there was no need to test our assumptions.

    I appreciate you writing this but most of it is really obscurative.

    On 6 May 00, at 13:08, Jack Ring wrote:


    > > ----- Original Message -----
    > >
    > > > There is NO WAY (repeat sotto voce -no way) that you can justify
    > > > or support applying a normal distribution to anything but randomly
    > > > selected populations. The only way this makes any sense is if you hire
    > > > randomly, completely randomly.
    > >
    > > [JR]First, it seems to me that you confuse a) the method of sampling a
    > > population with b) the characteristic distribution of the population.

    Nope. It's clear. A random selection of people on many
    characteristics will give you a sample that follows a normal
    distribution. A highly selected sample will not. It's really not that
    complicated, and perhaps others have explained this on the list
    more effectively than I have.


    > There
    > > are many ways a population can evolve to a normal distribution
    > > regardless
    > of
    > > how the
    > > population is observed.

    I have no idea what that means.

    > >
    > > > If you actually do have a normal distribution underlying employee
    > > > performance, your hiring practices would be absolutely broken.
    > >
    > > [JR]Do you have a rationale for this seemingly nonsensical statement?

    You betcha. The purpose of selection is to choose people in the
    upper tail. The better the selection the less variance, and the more
    the distribution will no longer be normal. As in selecting all people
    who are 6'11 for a job. That's super selection, and you won't have a
    distribution at all.


    > > And notice that the distribution does not "underlie" employee
    > > performance, it is simply a characterization of the spectrum of
    > > employees performances. It is not a foundational or driving factor but
    > > only a view of the spectrum of employee performance appraisals.

    This is a statistical issue, not a "whatever babble issue". So, can
    you speak plain Engish or plan statistics, jack?


    > >
    > > But more importantly, notice that the discussion was not about
    > performance.
    > > It was about appraisals of performance. We hold open the possibility
    > that
    > > the appraisal is far from a true estimate of actual performance.

    Sure that's fair enough but who is "we"?
    > >
    > > > Since the assumption about normal distributions is wrong, so is
    > > > everything that follows.
    > >
    > > [JR]If you get beyond the statistics of management and HR of management
    > and
    > > get on to the cybernetics of management then you can comprehend that the
    > > issue is not about the initial assumption but about a process for
    > converging
    > > to true appraisals of performance by appraisal writers. So the initial
    > > assumption can be dead wrong and everything that follows can be quite
    > useful

    Uhuh. so that justifies using something that is incorrect? If I have a
    choice between believing the Earth is flat or round, but both work
    equally well for navigating to work, does it make sense to believe
    the first? Or better yet teach it to management students?

    I just have this funny thing. I get reall annoyed when I see
    educators saying..."well it might work, even if there is no
    justification for using this tool, and it shouldn't work, so I WILL
    teach it to my students.



    > > . If you understand self-correcting systems an assumption can be wrong
    > and
    > > the outcome resulting from that assumption can be quite illuminating.

    I'd rather prove an assumption, then build on it. You obviously
    seem to be comfortable with starting with a poor indefensible
    assumption and then...what?


    > > It
    > is
    > > called testing the hypothesis. Didn't they teach that in your school?

    You told me that we don't need to test assumptions or something
    to that effect in one of your private remarks. Don't they teach
    statistics and research methods in your school, and don't they
    teach you that you don't ASSUME distribution characteristics
    without testing them?

    I'm sorry, but you suggested a method that has no justification. It's
    that simple.

    I guess I just expect more from people who teach and educate
    managers.


    Visit the work911.com supersite at http://www.work911.com
    for work related articles, or to find almost anything including
    book reviews and suggestions, discussion lists and more.