--- In [email protected], Peter <[EMAIL PROTECTED]> wrote:
>
> 
> 
> --- L B Shriver <[EMAIL PROTECTED]> wrote:
> 
> LB, your last post in this thread was truncated, so I
> couldn't include it, but I wanted to compliment you on
> your observations. As you note, science is, at its
> best, an international, public discourse. I can
> understand MIU's reluctance to hand over the raw data
> for reasons that have nothing to do with this
> discourse. All movement research is for one purpose
> only: to promote the teaching of TM/TM-Siddhis. It's
> for PR only. Those in charge, MMY, aren't interested
> in developing a coherent theory of the field effects
> of consciousness. They just want to sell TM. The MIU
> researchers won't hand over the raw data because the
> ME is very weak, almost noise, not pattern. It can
> easily be shown not to exist using alternative, and
> more traditional, statistical methods used in this
> type of research.


Yes. 

And in statistical methods such as multi-variate regression (and ARIMA
which they used -- which can be thought of as a specialized subset of
of regression methods), a large number of "model specifications" can
be developed and tested. 

(A model specification being the articulation of dependent variable
with various control and explanatory variables aka independent
variables. Such as: crime is a function of weather, LE funding and
unemployment. OR, crime = f(weather, lagged abortion rates, education
levels) OR, crime = f(severity of punisment levels, conviction rates
(agressive prossecution), police on the street. OR crime = f(lagged
head start programs, lagged pre-natal care, lagged school lunch
programs, and lagged classroom size).

Many, many model specifications can be tested. A good analyst and
research team will look to at least half a dozen key parameters to
evaluate how well each model explains the variations in the dependent
variable: i) overall model fit via R^2 and global F test, ii) the
significance of each independent (control) variable, aka t-tests,
i.e., was it a random effect?, iii) were the independent variables
correlated with each other (a bad thing, called multi-collinearity),
iv) are the variables correlated with past values of themselves aka
autocorrelation (a bad thing), v) are the residuals random or skewed
relative to the dependent variable aka hetroscadisity, (a bad thing),
is the model specification consistent with "theory", aka does it tell
a reasonable and plausible story -- or were a million independent
variables tested, and chosen ONLY due to good fit (aka, which chan
happen via 'spurious corrleation" but really are just randome effects,
vii) is the data "good", viii, were the number of independent
varibales less than 10-20x the number of observations, etc.  

Either by inexperience, or via intent to manipulate and arrive at a
pre-selected result, a researcher can shoose model specifications that
show a particular effect, via one paramenter, but are weak in other
parameters. But these parameters or diagnostics (like the eight above)
 can be swept under the carpet and not cited in the research results,
or worded in a best-spin sort of way (characteristic of some TMO
reseaarchers, IMO). 
 
Thus, its critical to make the full original dataset available to
other researchers to test the hypothesis via "their" approach to model
specification and selection. If a suboptimal specification were chosen
by the original researchers, because one or two paramenters shined,
but others sucked, this "illusion" can be uncovered by indepedendent
analysis and comparision of the results of different model
specifications -- and the full spectrum of the relevant parameters and
diagnostics associated with them. Or it may be found that alternative
model specifications, strong on all levels, produces a different
conclusion than the original research. This may indicates something
important is missing in one or both models, and more analysis is
necessary.










------------------------ Yahoo! Groups Sponsor --------------------~--> 
Get fast access to your favorite Yahoo! Groups. Make Yahoo! your home page
http://us.click.yahoo.com/dpRU5A/wUILAA/yQLSAA/JjtolB/TM
--------------------------------------------------------------------~-> 

To subscribe, send a message to:
[EMAIL PROTECTED]

Or go to: 
http://groups.yahoo.com/group/FairfieldLife/
and click 'Join This Group!' 
Yahoo! Groups Links

<*> To visit your group on the web, go to:
    http://groups.yahoo.com/group/FairfieldLife/

<*> To unsubscribe from this group, send an email to:
    [EMAIL PROTECTED]

<*> Your use of Yahoo! Groups is subject to:
    http://docs.yahoo.com/info/terms/
 


Reply via email to