The Failure of Dr. Weiss and Mr. Aschkenasy Testing Procedure
What is the Biggest Failure of the Dr. Weiss and Mr. Aschkenasy and the Best Part about the Initial BBN Testing?
First of all, what was the mission of Dr. Barger and BBN, along with Weiss and Aschkenasy?
Was it to test the hypothesis that the Dictabelt recording was made in Dealey Plaza and recorded the gunshots of 1963?
No. Their proper mission was to test out two hypotheses:
Hypothesis 1: The Dictabelt recording was made in Dealey Plaza and recorded the gunshots of 1963?
Hypothesis 2: The Dictabelt record was not made in Dealey Plaza and so did not record the gunshots of 1963?
Both hypotheses need a fair chance of being found. I don’t think that Dr. Barger gave Hypothesis 2 a fair shake but Weiss and Aschkenasy were particularly biased in their approach.
The BBN Hypothesis which they pursued with single mindedness, is that from the recorded sound, one can deduce:
• The location of the shooter,
• the location of what the shooter was aiming at, that is, the small area the bullet struck within,
• the location of the microphone.
and that the Dictabelt recording likely recorded the sounds of these shots.
They are searching for a correlation between an impulse in the 1963 impulses (7) and the 1978 test impulses ( 432 ). But finding a correlation might not indicate success. The correlation might be a matter of luck. Afterall, checking through hundreds of correlations, one might get lucky (or unlucky, depending on how you look at it) and find a spurious correlation. The lower one sets the correlation threshold, the more likely a false positive will occur. So, one needs some way of telling of the correlations you found are true correlations or false positives.
Again, the key is, if correlations are found:
• have some way of telling if your correlations are likely true, or are likely false positives.
Your test procedure must have this property. You must have some way of detecting false positives. The more ways the better.
How should your hypothesis by tested? In general, in science, you want’ to be strongest critic of your hypothesis. When designing test procedures to see if it has support or not, you want to make it possible for it to fail, in case it is a false hypothesis. Indeed, you want to make as easy to fail as possible. This is highly counter-intuitive but true. And goes against human nature. You don’t want your hypothesis to be proven false.
For the Acoustic Tests you want to give your hypothesis the maximum number of ways it can fail. There are four ways the BBN Acoustic Test can fail:
Hurdle 1: No correlations, that is considered strong enough, is found between any of the 1963 impulses ( 7 ) with any of the 1978 recordings ( 432 ).
Hurdle 2: Correlations are found but they contradict each other.
Hurdle 3: Correlations are found but they contradict a highly probable fact of the case, that the shots were aimed at the limousine.
Hurdle 4: Correlations are found but they contradict another highly probable fact of the case, that they are not consistent with a motorcycle following behind the limousine with reasonable speeds and not reversing directions.
Success is most strongly indicated if some of the 1963 impulses ( 7 ) pass all four hurdles. And none of the impulses failed the second hurdle and none or at most one failed the third hurdle and none failed the fourth.
The BBN tests were designed that the hypothesis could be tripped up in any one of four ways.
The BBN tests did not pass with flying colors, because, while they did find correlations between the 1963 impulses and the 1978 test impulses, too many of them failed for other reasons.
• Hurdle 1: Four somewhat strong correlations were found for four of the shots ( coefficient 0.8 ) and a much weaker one for a fifth shot ( coefficient 0.6 ), favored by Dr. Thomas.
Basically, the tests got by Hurdle 1 for at least 4 of the 1963 impulses, although a stronger correlation coefficient than 0.8 would be desirable.
• Hurdle 2: In three out of four cases where there were multiple correlations found, for the same shot, for both a shot from the TSBD and from the Grassy Knoll.
The BBN tests failed to get by Hurdle 2. Some clear false positives were found. Casting doubt on all correlations found.
• Hurdle 3: In most cases the correlation did not indicate a shot aimed at the limousine. In only 4 out of 15 correlations, was the data consistent with the shot aimed at the limousine.
While one can speculate about hypothetical “diversionary shots”, too many diversionary shots are just too unbelievable. Why wouldn’t even a diversionary shot be aimed at the limousine to increase the chances of success?
The BBN tests failed to get by Hurdle 3.
• Hurdle 4: In this case the hypothesis held up well. But I believe this likely was a result of limited time to do the work in 10 days, make 3.024 comparisons and calculations by hand. Likely they thought they found one shot, then used that location to guess where to find the next shot, assuming an 11-mph speed of the limousine. If finding a “lucky” correlation was not too unlikely, the correlations found would match the speed of a motorcycle moving at a steady 11 mph.
The BBN tests seems to get by Hurdle 4 well, but I believe that there is another explanation that has nothing to do with successfully finding true correlations.
In any case, not all four hurdles were cleared. The “Dealey Plaza Shooting Was Recorded” hypothesis failed.
The BBN tests were designed, either by accident or on purpose, to fail pretty easily. There were a lot of things that could go wrong. This is not a criticism of the BBN test procedure, but praise. If your hypothesis is false, you want your test procedure to reveal this. I think that, in some ways, in many ways, the BBN procedures are a model of how a hypothesis should be tested.
Not that this was done perfectly. A good procedure won’t work if you simply dismiss all false positives as irrelevant ‘false alarms’ and ignore them. But as long as you present the raw data, as they did, it is still possible for others to detect probable false positives.
If the BBN hypothesis was strong, they would find correlations, with no false positives, with no false alarms, and this hypothesis would have passed all four hurdles with flying colors.
But if the hypothesis was false, it was all too easy to find false positives, or “false alarms”, with the procedure the BBN followed. It seems that for each microphone they tested, they tested all 12 tests shots. Maximizing the odds that multiple correlations would be found, maximizing the odds of detecting false positives.
The great thing about the BBN procedure, is they didn’t do the following.
• Every time they find a correlation for a shot at the wrong target, they decide not to record it, on the grounds “it is just a false alarm”. It is more than a “false alarm” It is an indication that your procedure is collecting false positives. This should be recorded, and they did.
• Every time they find a correlation for a shot at the correct target, they decide not to test that microphone for any other target, on the assumption they found a true shot. Why search for other possibilities, like a shot from another location, when you know any found correlation is impossible. The same shot can’t be from both the TSBD and the Grassy Knoll, can it? But you should search for these “impossible shots” because if you find them, this indicates that you are collecting false positives.
Yes, in their final conclusions, they dismissed the false positives. But they did not fail to record them and to list them in their reports.
The only thing I suspect they did wrong is not do a through check of all 3,024 combinations. Due to a lack of time. And this could give a false sense that all the data matches a motorcycle moving forward at 11 mph throughout the entire 10 seconds, when a through check of all 3,024 possibilities may reveal a lot of contradictions in the position of the motorcycle, just as there was were in the position of the shooter and the position of the target.
In contrast, what method did Dr. Weiss and Mr. Aschkenasy use? Dr. Weiss and Mr. Aschkenasy took the recording of the 1978 Grassy Knoll shot at Target 3 recorded at Microphone 3 ( 4 ) and calculated what it should have sounded like:
• If the microphone had been positioned a few feet differently from 3 ( 4 )
and:
• If the position of the 1963 Grassy Knoll shooter had been positioned a few feet differently from where the 1978 Grassy Knoll shooter was positioned.
We don’t know how many different positions for the microphone they might have calculated. If they made calculations for every position in a grid, one foot apart, 15 feet up the street through 15 feet down the street, and 10 feet wide, that give (15*2 + 1) * (10 + 1) or 341 different positions for the microphone. This is a little more complicated than I described because their assumption was that the microphone would be moving so the calculations would be for what the waveform would be if the microphone started in one position and steadily moved about 5 feet over 300 milliseconds. But, either way, they may be calculating over 300 different positions where each position is not a stationary point but a line segment that is 5 feet long.
For each hypothetical position of the microphone, they might have run calculations for many different hypothetical positions of the Grassy Knoll shooter, positioned several feet to the left or right of the 1978 shooter, and maybe at various depths from sticking the rifle over the fence to being a various number of feet back. For all we know, they may have made calculations for 300 different positions.
If it was that many, it might by 341 * 300 combinations or over 100,000 different combinations of a Grassy Knoll positions with a microphone position. So, perhaps, Dr. Thomas’s 1 in 100,000 calculations might be correct. Even with 1978 computers, it would be possible to run this many comparisons if one let a computer crunch numbers for a few hours, or days, if need be.
In any case, Weiss and Aschkenasy may have calculated a thousand, or maybe many more than a thousand, theorical waveforms, and compared them to the 1978 waveform formed by a test shot from the Grassy Knoll, at Target 3, from microphone 3 ( 4 ). So, there were plenty of opportunities for generating false positives.
So, given the procedure of Weiss and Aschkenasy, how many hurdles does their hypothesis have to pass? Is it four, like the procedure developed by Dr. Barger of BBN? No.
• Hurdle 1: Must find at least one strong correlation between the waveform calculated from one unique microphone position (which moves 5 feet during the duration of the impulse) with one unique Grassy Knoll position.
• Hurdle 2: It is alright to find multiple correlations, provided they don’t contradict each other. If two contradictions are found, but both are quite close to each other, that is find. But finding two correlations that different by a lot, the location of the microphone differs by 15 feet and the location of the shooter differs by 20 feet, that’s a failure. One of these correlations has to be a false positive and so both must be considered suspect.
This is not near the number of hurdles the BBN procedure provided. Since only one 1978 shot is being tested, there is no chance of getting correlations that match two different firing positions separated by over 200 feet. Or getting correlations that match for two different targets separated by over 100 feet.
And depending on how Weiss and Aschkenasy did things, the odds of being tripped by either hurdle can be minimized.
They might calculate 10 different microphone positions with 1 Grassy Knoll position. If no correlations are found, with could run the calculations for the same 10 different microphone positions with 10 different Grassy Knoll positions. If no correlations are found, they keep increasing the number of microphone positions and shooting positions until a waveform was generated that matched up well with the 1963 waveform. And call it a day.
We don’t know if this is how Weiss and Aschkenasy did this. They never provided any sort of details of their test procedures. How many combinations were initially calculated in their first computer run. How many in their second. They just provided us with the end result, this position for the Grassy Knoll shooter, combined with this position for the microphone, produces a calculated waveform that matches up very well with the 1963 waveform.
The lack of more details about the history of their test runs, how many test runs were made, how many combinations did each test run make, what were their results, makes me a little suspicious.
The way the procedure should have been run is to run this procedure with different test shots. There were 69 rifle shots fired on that day in 1978. Run these calculations for all the 1978 Grassy Knoll shots that were fired at Target 3, not just for one. If the other Grassy Knoll shots generate microphone and shooter position correlations that are very close together, that is good. If they differ too much, that means false positives have been found.
Expand this to some of the test shots from the TSBD. Yes, I know. Weiss and Aschkenasy consider these shooting locations have been ruled out. Run the test anyway. Using different nearby shooting positions, not just the sniper’s nest. If this procedure is just finding random correlations in data that is “white noise”, it will also likely find correlations for some TSBD shots, just as the BBN procedure found correlations on the same shot with both the TSBD and the Grassy Knoll position.
If this is done, and it is found that no correlation is found for any of the TSBD shots, no correlation is found for any of the Grassy Knoll shots fired at Target 2, only correlations that were test shots from the Grassy Knoll at Target 3, and all these correlations agree with each other, that would be powerful support that they were not finding false positives, that the found correlations were legitimate evidence of a shot from the Grassy Knoll.
What if there is only enough time to do the calculations for one 1978 test shot? Raise the issue with BBN and the HSCA. Stress how important it is to test multiple shots, so a serious effort can be made to find false positives, which will alert them that they are finding correlations in “white noise”. They need more time. Or, failing that, they need more experts to be drafted into this project to get the tests done.
What if the answer comes back that they must made do with the time and resources they have? They can still do more. It would take hardly any addition time to run the computer program, and just compart one shooter position, the 1978 Grassy Knoll position, with one microphone position, 3 ( 3 ). Does the calculated waveform match the real 1978 waveform recorded at microphone 3 ( 3 )? Do the same with several microphone positions, 3 ( 1 ), 3 ( 2 ), 3 ( 3 ), 3 ( 5 ), 3 ( 6 ) and 3 ( 7 ). This would give demonstrate they can indeed calculate the waveform from different microphone positions. And if this could be done with at least one other 1978 test shot, from the Grassy Knoll, at Target 3, so much the better. But that may take a little more time.
And then to make certain to put in the final report that the Weiss and Aschkenasy procedure was not tested enough to confirm the hypothesis. They think the hypothesis was good, but it was not stressed as much as they would like. This is what they should have done, but didn’t.
In conclusion, it does not look like Weiss and Aschkenasy stressed their hypothesis very much at all. Not nearly as much as BBN did. Which may be why the BBN procedure found many clear false positives while the Weiss and Aschkenasy found none.
