GEOSTATS: The controlled experiment
- Geostat people:
Everybody wants to prove something, yet
mighty few will give the reader the error terms he needs
to judge the evidence. A few journals even say just slip
in "p = 0.95"; do NOT give the ANOVA.
Well in the recent dolphin case, it is not the
application of a test of randomness as the observations
are ordered, etc. The matter is acquiring an experi-
mental attitude that involves randomization and repli-
cation. The control is something known sometimes, and
the experimental something to be defined. This much can
be a long hard path for a beginner or an expert. Sampling
size demands may be way beyond our capacities, yet we
assume consecutive samples are reasonably smaller than
genuinely independent ones. Words like independent and
normal are historic and convenient, not rigorous. The
thing being measured is the response variable, and the
controls are the independent variables by convention.
The shorter the interval, the better... more
points, less variance and higher cost. However, it may
be mendacious to connect dolphin sightings as each can
be a different one. We might need a tag-and-release
system for a population as in PAP Moran (1951, Biometri-
ka). The rub is that while these things are simple,
they are costly.
"Random" often covers unknown assignable causes.
If we get a coefficient of determination of r squared =
0.85, and then we correlate some other variable to find
0.92, we are making real progress. Might we find another
variable and reach 0.97? We would take that as a lawlike
relation and would put that 0.03 down to random
Much of use is under Quality Control. Despite
the Genichi Taguchi effect, Experimental Designs by WG
Cochran & GM Cox (1957) and Planning Experiments by DR
Cox (1958) remain good books as they can form attitude
thus intuitive level.
Two other books are GEP Box & GM Jenkins (1970)
and GM Jenkins & DG Watts (1968) that deal with time
series. These 4 books are in the Fisher-von Mise-Feller
Dolphins can be forecasted. Monitoring programs
justly popular in biology are available off the net.
Still, they surely require massive inprovement.
We want a high number of observations taken at
uniform distances especially if we translate time to
distance. If a ship has uniform speed, and we have its
direction, we know where it will be after a given time.
Nonetheless, dolphin activity is cyclic. The same
dolphin is both coastal and far oceanic. Autocorrelation
and crosscorrelation with a control should provide
However, the sighting alone without the ecological
oceanic parameters associated is not playing the game.
This is the missing set of assignable variables. No
investigations on fluctuations in the sizes of dolphin
populations are worth the powder to blow them to Hell.
Part of that is from confusing random
with determined effects. Part is by
We do have an empty set, so face up to it.
So she says, "Don´t you understand that if
I don´t have positive findings, my funding
will be cut off?" Or, "How can I defend my
Next, we have about 3 species like spotted
to worry about. Even with 20/20 vision,
the observer has only a foggy notion of
which species she saw.
Getting a dolphin-beamed radio signal (tag)
certainly will make data so that one has to have online
computer interfacing. (Ask the Bell Laboratories.) The
opposition is using tuna-beamed signals and has found
the S/N ratio intolerable.
PS Do you remember who said, "You can fool some of the
people some of the time, but you can´t fool all of the
people all of the time." ?
PPS As the dolphinic student inappropriately requested a
test for if-it-were-random, this quote is added from Box
"Much of statistical methodology is concerned with models
in which the observations are assumed to vary indepen-
dently. In many applications, dependence between the
observations is regarded as a nuisance, and in planned
experiments, randomization of the experimental design is
introduced to validate analysis conducted as if the
observations were independent. However, a great deal of
data in business, economics, engineering and the natural
sciences occurs in the form of time series where ob-
servations are dependent and where the nature of this
dependence is of interest itself."
*To post a message to the list, send it to ai-geostats@....
*As a general service to list users, please remember to post a summary
of any useful responses to your questions.
*To unsubscribe, send email to majordomo@... with no subject and
"unsubscribe ai-geostats" in the message body.
DO NOT SEND Subscribe/Unsubscribe requests to the list!