Some thouhts that have been swimming around.

I checked the OED... as i suspected improbable refers to unlikely in an
extreme form. so, for example - the crumbling of the WTC buildings was
improbable on 9/10/01 but neither unpredictable nor impossible.

unpredictability, on the other hand describes a situation in which it
is impossible to predict, which i would argue strongly against on the
basis that prediction is always possible even if it later, by whatever
arbitrary standard has been agreed to in advance, turns out to have
been incorrect...

Than there is also a temporal issue that i suspect may be important to
consider. If i am doing polling data on election day, and I choose my
sampling frame well, and i am not doing the polling in the state of
florida in 2000, then when i draw my inference from my sample to the
whole population of voters - anywhere but florida, i am not predicting
in terms of a future state of the world, but merely describing
something that has already happened. i am still likely to be wrong
about the estimate though the 95% confidence interval will likely be
very accurate, tho potentially useless information if it is not
sufficiently narrow for whatever decision-making has to occur in
reporting, interpreting, or acting on my results...

On the other hand, when i am trying to forecast a future state of the
world (i am mindful that some may argue that there is no such thing as
objective, forward moving time, but that is another discussion
entirely, and i am already on record as ridiculing victor stenger for
his 'backward traveling in time particles' thesis, anyway), and this is
the area that I, were I in martha's purple dress, would be more
concerned about. Suppose, for example, I were talking with a student in
the 1950s or 1960s and they, having just run a statistical regression
on teachers salaries and liquor sales. I would be very concerned if
their proposed solution to excessive alcohol drinking was to recommend
that teachers salaries be lowered on the presumption that there was a
causal link between salaries and drinking that might be rectified in
the future by adjusting teachers' salaries. That is a wholly different
matter than reporting about the current state of affairs to making a
prediction that, based on a certain intervention - a future state of
affairs would come to exist.

I think there are also some issues in terms of how we approach
knowledge - rather than asking about our ability to produce unitary
knowledge by building up, one dimension at a time, perhaps we ought to
think of it the same way that galileo did when considering the ages old
proposition that heavier objects fall faster than lighter objects. In
his 'thought experiment', galileo reportedly argued thusly - if a
heavier object falls faster than a light object, then if I put both
objects together they will be heavier than the heavier object and ought
to fall faster. On the other hand, it would be unreasonable to assume
that the lighter object, which on its own would fall more slowly, would
not hold the combined object back because of its slower speed. The only
way to view the situation more accurately would be to assume that they
all fall at the same rate of speed...

Now, shifting to the question of excluding any form of 'knowing', if we
assumed we had 'unitary knowledge' that just happened to incorporate
inferential statistics or any quantified knowledge, would it be
justifiable to exclude that knowledge from our unitary knowledge? If
so, on what basis would that exclusion be made? Could we then transfer
the same justification for excluding any other aspect of knowledge -
aesthetic representation, phenomenological inquiry, one 'qualitative'
method after another on the basis that its contribution was
reductionistic compared to the remaining wholeness?

Then there is yet another issue - are all things inherently
immeasurable? I used to have polite discussions with a peer about this
issue all the time a few years ago. The question of something being
immeasurable is, at best, temporally determined. Could I have measured
X-rays in 200 BC? probably not. Can I do that in 2004 AD? Yes. Can I
measure the time til death of a healthy human being, barring knowing
that they are on death row in texas? No, not unless I have prior
knowledge of the exact moment of their death. I would argue that the
assertion that things are 'not measurable' must be no less justified
than that they are measurable - perhaps more so. Person X may not have
the skills, aptitude, tools, or technical efficiency to measure a
phenomena - but that does not automatically translate to the phenomena
not being measurable by someone who possesses the appropriate skills,
aptitude, tools, or technical efficiency to measure the phenomena. So
someone says they want to study sadness but feels that no extant
approach to measurement is appropriate. On further discussion they say
they want to compare sadness in two populations, high school graduates
five years after graduation, and their HS peers who did not graduate at
the same time. Either qualitative or quantitative methods might be used
to do this. We might seek to interview 8 people from each group about
the experience of post-hs sadness. the problem, of course, is that the
only people who will show up will be people who identify with sadness -
so the comparison may be flawed if our interest was to compare the
percent of people who are sad in each group. If, on the other hand, we
are more interested in the unique ways that sadness might manifest and
be reported by HS grads v HS non-completers, it would likely be foolish
to ask a randomly selected portion of each group to take the Beck
Depression Inventory...

I think the appropriateness of either qualitative or quantitative (or
both) methods is ALWAYS dependent on 'what is the research question.'

One additional point for reflection - when we conduct an inquiry
seeking to capture wholeness - are we not always basing our assessment
of wholeness on a snapshot, a sampling of all possible snapshots that
we, or others, might have ever taken? Here I will acknowledge one of my
more embarrassing moments. In my youth, as a young cub just out of
school, I worked for the department of social services. I had a client
who had been convicted of second-degree manslaughter. He had 7 children
with one woman and was reputed to have fathered many other children in
the area. I did the good liberal thing of rationalizing and justifying
behavior, which I personally found somewhat troubling and formed an
image of him that I thought to be accurate, positive, and balanced. I
held him in positive regard, did everything I could to help him and his
family deal with the rigors of life… One day, as I walked through the
local 'community action agency' building, I heard a voice that sounded
familiar. I retraced my steps and looked into a meeting of the
community action board. There, wielding the gavel, was my client, the
president of the board. Now a simple measurement incorporated into my
prior thinking would have produced a radically different portrait - How
many local community action boards do you chair? __ (0=none; 1= one or
more). This answer would not have made any qualitative distinctions
between how well he performed in that role - but it would certainly
have fleshed out the unitary portrait of his life that I had woven
based on obviously extremely limited information. That my client served
in so responsible a position and held such positive regard in the
community was improbable earlier that day - but clearly not impossible.
That he held such a position and such regard was predictable - tho my
personal prediction might have been on the order of 1/1000 or less with
a 99% confidence interval… The prediction would clearly have been
proven false based on a little inquiry however - as premature
predictions are from time to time…

Like it or not, all of our data collection is constrained. I am color
blind, there are color vision based distinctions I will never be able
to make that another person can. I could, perhaps substitute
measurement of wavelengths but I may miss the affective response to
gradations of color that are appreciated by a non-colorblind person.
But certainly understanding that there is a difference in wavelength
will help me to appreciate the distinction between my own perceptions
and those of my less-impaired colleagues. On a related note - I once
pondered the implications of the crackpot former prestidigitator, 'the
amusing randi' possessing the ability to detect and manipulate a human
energy field. What I argued, would the errant assumptions of skeptics
turn out to be if randi was so skilled at observing and modifying the
HEF that he could actually modify choices made by TT practitioners in
performing an experiment such as the rosa tt test, just by being in the
room - or anywhere else for that matter? What if he was able to do this
but for purely selfish and gratuitous reasons, chose to deny something
that he could actually prove to be true based on his profoundly
extra-normal skills?

In the end, it is not, I suspect, whether we predict but what we do
with and assume about our predictions. Teachers salaries ought not be
lowered when workers salaries overall have risen and fed the increase
in alcohol consumption, quite apart from the salary rise among
teachers…

In any event - before we could defend a position that there is no place
for inferential statistics in the suhb, I think we would have to
distinguish between the different types of inferential statistics,
their actual meaning and interpretation, and how the knowledge derived
was to be employed. I would also add as a footnote to all of this, that
it may be a mistake to prematurely write off quantitative research and
quantitative measurement. I fear that this is often done simply on the
basis that some people are outsiders to the millenia old hermeneutic
circles (yes I know heidegger did not mean 'circle') of mathematics,
philosophy, and more recently statistics and probability theory. Our
decision, I think, ought not be based on the lack of understanding of
outsiders to those dialogues - that the dialogical circles are ongoing
and that the insiders to the discussions have their own understanding
that the grasp of the statements of number, measurement, inference, are
incomplete and open to endless redefining. We should never, I think,
suggest that the current shorthand linguistic conventions of
statistical inference (p-values, confidence intervals, bias, estimates,
certainty, uncertainty…) are anything more or less than the current
'state of the art' in statistical and epistemological theory…

Just a couple of errant thoughts to add to the dialogue…

bear