EXECUTIVE SUMMARY:
- We extend a s/w that correlates day by day planetary parameters with
(lagged) daily UFO sightings.
- We use the s/w to matches patterns found in the observations against
a 2nd model of a "UFO fleet" operating across the solar system
according to simple rules. All the simplest variations on the theme
were pre-computed to match against the actual correlations we observed.
- Looking at UFO types "all", "Lights" and "Non Lights" we find the
simplest apparent set of assumptions that matches the observations
is: the probability of a flight from A to B depends on the present
distance between A and B; most UFO's originate nr Saturn; "Light" UFO's
seem to originate nr Neptune (and maybe even the Sun, the s/w finds);
Non Light UFO's originate mostly from Saturn with some from Neptune.
- Other origins can't be eliminated because only the simplest set of
scenarios were used in this study. Even more slightly complex
assumptions are likely to produce different results given the
chaotic and inter-correlated nature of planetary movements.



We've seen in a previous post how the positions of the planets seem to correlate highly with day by day UFO sightings across N America. In
particular we found the distance between Saturn and the Earth seemed
to explain a big chunk of sightings and other data showed it was not
likely a significant number of UFO sightings were just a matter of
confusing Saturn with a "real" UFO. (E.g. the same patterns were seen
for day/night sightings and sightings of objects that don't seem to be confusable with a planet).

To extend that work I've added a more planetary parameters to the
mix. The apparently position of each planet -- declination and right
ascension -- boost the list to 13 parameters for each "normal"
planet, and 15 in the case of Saturn with its rings. We can now
correlate this set of time series against UFO data for various types
of object to see how well each planetary param can predict future
sightings. It's assumed a lag between the change of a planetary
position "now" and UFO activity in N days might indicate "something"
is travelling between that planet and Earth and taking maybe around N
days to do it. In this way we found last time a lower bound on
putative UFO movements between planets was substantially sub-FTL.

Working with an AI s/w has been overall a boon to this work. While a
complex program spitting out results can be tedious to check and
almost impossible to debug, it can also help in its own
development. In this case the s/w, which uses some standard stats
tools to do some of the heavy lifting, discovered one package gave
unusual results in some cases. This is often the case when packages
are pushed (by statisticians) further than their designers envisioned.
But a "problem" with AI's is they tend to push whatever tools they are
allowed to use *well* past design limits. They have generally no
concepts how a package is typically used or what kind of data people
normally give it; they just use it if, when, and how they see fit.
And sometimes the package obviously breaks or spits out an answer that
is wrong but without printing a warning message with it the AI might
pick up on.

After much nail biting the problem in question was tracked down and
fixed. The up side is the stats s/w now outputs numbers with a known
and hand-checked (:) error bound -- generally +- 5%. If 2 R2
correlations (normally I use the R2 statistic from various time series regressions) differ by say 10% from their common average then the
larger one is "most likely" really the larger. If they are closer than
10% then the ordering can't be unambiguously determined.

So the AI can now "confidently" print out a list of correlations of
each UFO type (shape/color) against day by day (and hour by hour if
necessary) planetary movements at the time and order them from largest
to smallest. The larger correlations are then the "most likely" ones
to be "real" and not due to some luck-of-the-draw in the data. And
the order of the correlations show e.g. which planets are "more
involved" in the particular kind of UFO activity, and which are less
involved.

For example. We can correlate the daily UFO sighting counts against
planetary movements from an ephemeris s/w and find the table:

Planet Parm R2
(in TS regr of planet param predicting
daily UFO sightings)
mercury Dec 0.56812030
neptune rg 0.50386267
venus Dec 0.47351440
uranus rg 0.45647785
saturn rg 0.43902437
venus RA 0.41095631
jupiter Dec 0.27301541
pluto rg 0.23986765
mars RA 0.21302930
Dec == declination (celestial latitude) of planet in degrees
RA == right ascension (celestial longitude)
rg == distance between planet and Earth in AU

The table shows the "most involved" param with daily UFO sightings in
total is the declination of Mercury. The next most correlated is the "geocentric distance" to Neptune. (Note the diff in R2 between 1st
and 2nd is slightly more than 10% of either). Etc.

And now the complication. The AI has warns me already that we cant
just assume most UFO's are coming from Mercury because of the first R2
in the list. They may be coming VIA Mercury, for example. And in the
latest twist the AI s/w has proved to itself and me the correlation
between Mercury's declination and UFO sightings may be "induced" by
something else because the movement of Mercury is synchronized with
other planets in the solar system. The distance between planets and
their periods are not random numbers. They are correlated by Bode and Kepler!

Up to recently I had assumed planetary parameters were "more or less" statistically independent, close enough. But it turns out some are
way way NOT independent of others.

It is therefore necessary to get the AI to look at the pattern of
correlations and decide which set of simple assumptions would results
in numbers than most look like the correlations we found. We need the
s/w to essentially "get inside the head of" UFO captains and simulate
flying between planets under different kinds of assumptions, and pick
out which set produces results most like the correlations we have
above. Only in SOME cases will the largest correlation straight out
point at the most interesting planet. Sometimes the real interest
will be lower down on the list or maybe be "hidden" and not appear on
the list at all.

To that end I quickly made another model that takes the planetary
parameters output from the ephemeris s/w, takes a set of assumptions
and strategies a UFO captain might use to decide when and where to fly
from their current location, and collect the correlations as we have
measured above for "real UFO sightings", and determine which strategy
looks to be closest to those observations.

While there is no guarantee we will find the *actual* strategy used
to fly between different planets of our solar system, we will likely
end up with *a* simple set of assumptions that produce a similar
result to the one we actually see. It will give us a "mental model"
of what may be going on, rather than "fact".

It would obviously be interesting to see what those set of strategies
might look like.

Drum roll Mr Music! ....

The output from testing a big set of simple strategies is as follows:

Strategy R2 (measure of likelihood the set of
assumptions named
explains the observed set of Dec/RA/rg correlations) only6-t 0.81282165
only72-t 0.47853941
only60-t 0.37906993
only76-t 0.35098475
only64-t 0.33747267
dis1-t 0.29975186
everywhere-t 0.25808418

--- SoupGate-Win32 v1.05
* Origin: www.darkrealms.ca (1:229/2)