From: "Steven J. Powell" <email@example.com> Date: Sun, 19 Jan 1997 11:49:22 -0500 Fwd Date: Sun, 19 Jan 1997 17:24:17 -0500 Subject: Re: Philosophy of Science and UFOs >Date: Fri, 10 Jan 1997 11:33:38 -0800 >To: UFO UpDates - Toronto <firstname.lastname@example.org>, email@example.com >From: firstname.lastname@example.org (John Bindernagel) >Subject: Re: UFO UpDate: Re: Philosophy of Science and UFOs > I would like to mention two points: >1. Regarding the issue of whether 60% or 80% or 95% of reports are valid - >Sasquatch author John Green reminds us that if only one report out of the >thousands in our database is valid we still have an authentic problem >(phenomenon?) to explain. Absolutely. >2. We scientists like to discard data that does not fit our preconceptions >or established trends, assumming (hoping?) they are errors in measurement. >These "outliers" sometimes turn out to be very important and we ignore them >at our peril. There is the issue of "intellectual integrity" wherin we >really should not refer only to the data which support our case and quietly >omit those embarrassing exceptions and examples which may support an >opposing view. (We have lots of these in the sasquatch database.) Terminology is also always an issue. We don't want to "discard" _any_ data. Instead, we want find the best subset of data upon which to do some analysis. Obviously, we first separate out the data that has a solid or very good explanation. We don't discard it or throw it away or use it to start this evening's fire - we just place it aside in a named subset and continue on with the remainder. Here begins the discussion on whether the remainder is 5% ot the total or 30% of the total. Some favor each extreme (and _both_ estimates are extremes) and it is really irrelevant which number turns out to be right (unless it isn't my number <grin>). With our pre-screened unexplained data we have to make some more decisions. In this unexplained dataset will still be all sorts of things that probably need to be separated. For example, IC1 (Imaginary Case #1) has one eyewitness to an odd aerial craft. The eyewitness owns his a very large farm in Kansas, he's a lawyer and a doctor, a private pilot (retired USAF B-52 pilot) and pillar of the community. Sighting duration is under 3 minutes, and the observation is relatively featureless - just a slightly reflective bright nocturnal object moving around irradically. IC2 has one on-ground eyewitness, a teenager, three other independent non-colocated on-ground eyewitnesses (a diner waitress, a garage mechanic and taxicab driver) one with videocamera and one with SLR, the local airport tracks something on radar and local TV crew film several helicopters flying over the area 30 minutes later. IC1 has almost no details, IC2 has every possible detail, over 90% concurrance among the various eyewitnesses to those details. Do you put these two sightings in the same dataset - Yes or No - and why? Rule Number One: Your reason has to be fully generalized and fully objective such that the exact same rule can be applied to every other similar data integrity problem <grin>. So far I have not mentioned anything about explanatory hypotheses because it would be premature to do so. If we can stick to that guideline we can generally safely avoid the problem of screening our data _based on_ preconceived hypotheses. If we're to do any screening then it needs to be objective. I have a non-answer answer to my question above. I'd prefer to do a first-pass pre-screen of the unexplained data and separate out _all_ the single-witness cases. Then, with the multi-witness case dataset look for trends/patterns. If any trends/patterns are found (whatever they might be) then look for those in the single-witness case dataset. If we get matches _then_ we can start arguing the details. I imagine that a Bigfoot sighting databse has the same type of problems...
[ Next Message | Previous Message | This Day's Messages ]
This Month's Index |
UFO UpDates - Toronto - Operated by Errol Bruce-Knapp