Safety‎ > ‎

Normalization of Deviance

posted 13 Mar 2014, 10:39 by Support SarMobile   [ updated 8 Apr 2014, 04:20 ]
In the last two safety articles we talked about the deficiencies of technical search methods Wing Blanking and Ad hoc Direction Finding Antennas used by air search and rescue units in several locations, but specifically as used by the Civil Air Search and Rescue Association. In this article we will dive deeper into how deficient methods may come to be used, and illustrate the process by examining an attempt to develop and propagate a new technical method by several members of a sub-unit of CASARA (which in their lexicon is known as a Zone). We will examine a paper1 written to document the technique. If needed, a copy of this paper should be available through CASARA; if not contact us at the email address provided at the end and we will try to assist. We will conclude by drawing a similarity the use of these techniques in search and Normalization of Deviance as described by Dr Diane Vaughan.

It is not our intention to embarrass or belittle anyone. In fact this article has been delayed to give those zone members and CASARA leadership an opportunity to communicate any changes of opinion, operations or outlook they may have had. Our attempts to reach out to the members and CASARA were, for the most part, ignored. When not ignored, those responding made no effort to distance themselves or the organization from the paper, nor to enumerate any difficulties they had with the content of the paper. Quite the contrary, in fact.

There is a good amount of material in the paper, we will not be addressing it all. There is a full technical critique of the Cardinal Pass technique that was written by a former member of the same CASARA Zone2. We can provide electronic copies of that paper on request. We will be closely examining four topics discussed in Accuracy of ELT Searches1:
  1. The relationship between the technical knowledge and qualification of the people involved in the development of the Cardinal Pass.
  2. The veracity of the content of the paper and the effect it may have on public confidence in CASARA.
  3. The basic premise of the Cardinal Pass (and the “Traditional Aural Null” presented as a foil for the Cardinal Pass).
  4. What science and engineering tells us about this premise, and some specifics of how a true traditional aural null works.
For the last point we would like the reader to be familiar with our Practical Guide to Aural Searches. You may wish to read that now, if you have not already done so. As an extra benefit we will also be able to show, by using the failures contained in Accuracy of ELT Searches1, to help convince the reader they should be describing Aural Null techniques in terms of the Radio Horizon, not equal signal strength, and using Procedure C instead of Procedure A or B.


The academic qualifications of the people involved are, in general, quite good. One holds a PHD in Physical Oceanography, an other works for the Advanced Cognitive Engineering Laboratory at Carleton University. One might be forgiven for assuming that these two at least would be sensitive to the need for familiarity with the science and engineering involved. And yet, early in the paper they make this statement:

The overall topics of wave propagation and the mathematics of the various aspects of finding maximums and minimums are the subject of a vast literature in the fields of mathematics, physics, and engineering. Nevertheless, in the case of CASARA actually looking for an ELT we have little need to understand either advanced mathematics or complex physics.1

We find it puzzling that they would on one hand acknowledge the vast amount of prior art, but on the other hand dismiss the need to understand. We can accept that for the average CASARA member, following a clear set of instructions, a recipe of sorts, would not need to understand all the science involved. But if CASARA is to design the procedures, and compile the recipe, then someone in CASARA would need to have a comprehensive understanding of the science involved. They attempt to justify this position by continuing: “... we have a small number of transmitters (usually one) and reflectors to deal with.”1 Immediately their lack of knowledge and experience in the field is exposed. First: since the purpose of the method is to find a transmitter with an unknown location, the number of reflectors can not be known until the transmitter is located. Second: even a small number of reflectors can seriously impair the technique. On the other hand it is possible that by choosing the location for the transmitter location they may have improved the performance of the technique by avoiding conditions in which the technique fails. Of course staging a test like this does not prevent failure conditions from occurring during an actual emergency.


The success of scientific discourse as well as search and rescue depend to a large extent in our ability to trust the statements made by scientists and searchers alike. When that trust is betrayed in one part, then trust fails in all parts. Someone who would intentionally misstate facts on one occasion can reasonably be expected to misstate facts on another occasion. This is why it is unusual to report that data has been gathered under one set of circumstances when it was gathered under a different set of circumstances; even if the circumstances are thought to be equivalent.

During the remainder of this article we will be concentrating on the report of an experimental flight documented in Accuracy of ELT Searches1. There may have been several purposes for this flight, but within the context of the paper it appears to have been to collect data on the performance of the Cardinal Pass and compare it to the performance of a different technique. In short the procedure was to place a radio transmitter in a known location, then use each technique to estimate that location. Each estimate was then compared to the known location giving an error distance used as the performance metric:

A practice ELT was set out in a clear field just to the West of Wakefield, Quebec approximately 16 nm NW of Rockcliffe Airport (CYOW). The weather was clear and bright. The position of the ELT was determined by GPS while setting it out to within a couple of metres and the location was chosen so that the radiation pattern of the ELT would not be distorted from circular by any reflectors or obstructions.1 [emphasis added]

Quite straightforward. However, the ELT location was not determined by GPS. An attempt was made to gather GPS coordinates using an iPhone application after the flight when the ELT was retrieved. This attempt failed to produce a usable location. The navigator of the flight resorted to using Google Earth to produce the known coordinates of ELT. These facts were reported by the navigator (Muir L.R.) to the Zone Training Officer shortly after the flight. They were later confirmed by the spotter on the flight (Casey M.).

Perhaps Dr Muir believed that Google Earth was capable of providing a position accurate to within a couple of meters. This is not the case3. Even if it was, we believe reporting a position derived from Google Earth as being derived from GPS is ethically very questionable. For us, the decision to misstate the source of the known location coordinates cast a pall over other decisions the authors made. At times we were torn between believing the authors made a choice because they lacked understanding of the mathematics and physics, and believing they made the decision to intentionally improve their results. It is very troubling to have these thoughts about an organization that one might have to rely on for rescue. The following pictures show image to image and image to GIS data registration errors in the search area from the Google database as it existed in December 2013.

Figure 1: Road GIS data (yellow lines) relative to imagery.

Figure 2: ELT location. Note imagery registration error.


Accuracy of ELT Searches1 describes two airborne electronic search techniques, the Cardinal Pass and what the authors call a Traditional Aural Null. These techniques are in fact almost identical. They both attempt to use received audio volume to measure relative distance to the transmitting ELT. In both cases the receiver frequency is tuned away from the transmitter frequency. This is where the Accuracy of ELT Searches1 so-called Traditional Aural Null differs from what is the canonical Aural Null. The true Aural Null is conducted with the receiver tuned to the actual transmitter frequency. To differentiate these techniques we will call the Aural Null from Accuracy of ELT Searches1 an Off-Tuned Aural Null, and the true traditional Aural Null, an Aural Null.

The Cardinal Pass is based on listening to the received audio volume while the aircraft transits the the search area on one of the four cardinal compass directions North, South, East or West. According to the proposed theory the loudest volume will be heard when the airplane is at its closest point of approach to, and therefore abeam, the transmitter. The navigator then records the Latitude or Longitude as appropriate for the direction of flight. If the aircraft is heading North or South the Latitude is recorded, if the aircraft is heading East or West the longitude is recorded. The Latitude from a North-South pass is combined with the Longitude from an East-West pass to determine the location of the ELT.

The Off-Tuned Aural Null is performed by listening to the received audio volume while the aircraft transits the search area. The receiver frequency is off-tuned from the transmit frequency. When the volume is at some fixed level, often called the minimum received signal strength, the location of the aircraft is recorded. In theory each location where the same volume level is observed, at the same frequency offset is at the same distance from the transmitter. This would put the points on the circumcircle of a simple, convex, cyclic polygon and the circumcentre would be the location of the ELT. Perhaps this is why an effort was made to choose a location “so that the radiation pattern of the ELT would not be distorted from circular by any reflectors or obstructions.1 Unfortunately an airplane crash will not always be able to replicate these desired conditions.

These theories may sound plausible, but are they? Certainly radio signal strength at a receiver is dependent on the distance from the transmitter. Logically the signal strength should have an effect on the volume of the radio audio output. The problem is that distance is not the only factor that affects received signal strength, and received signal strength is not the only factor that affects the audio volume.

Knowing all the Factors

Since we are concerned with measuring distance by using radio waves let’s look at some examples. The simplest way of measuring distance is with a ruler or tape measure. By laying something of known length between the transmitter and receiver we can determine the distance. With a tape measure the distance is a function of how much tape is required to reach between the two antennas. We can simply read this value off the tape. This is impractical for long distances so we might use a surveyor’s wheel. Distance is measured by counting the rotations of the wheel and multiplying by the circumference. This is not as direct as measuring with a tape measure. If the wheel slips during measurement, or is rolled along a curved path, the measure will be in accurate. It will also be inaccurate if the wheel has worn and no longer has the circumference we expect.

Using radio waves there are two well known methods of measuring distance: RADAR and GPS. Pilots may be familiar with a third: Distance Measuring Equipment (DME). RADAR and DME actually function in a very similar way. Primary surveillance RADAR works as we have seen in war movies. A large radio transmitter sends out a pulse of energy that is reflected back off the target. The time the pulse takes to return is measured. By multiplying the speed of the RADAR signal in the atmosphere by one half of the time we can compute the distance. Secondary surveillance RADAR and DME are very similar, except that they don’t depend on the signal bouncing off the target. Instead an airplane will carry a transponder that listens for a signal from the secondary RADAR and replies with a signal of its own. With DME the aircraft has the equipment that sends out the initial signal to which the DME station replies. Again the round trip time can be used to compute distance. A GPS satellite sends out a repeating, but very long signal that is synchronized to its internal atomic clock. Just as we can listen to the lyrics of a song “The Star Spangled Banner” and know if we are listening to the start, middle, or end, the GPS receiver can listen to the signal and know what time the satellite clock read when that signal was sent out. The difference between the clock time taken from the signal and the correct time, multiplied by the speed of the signal is used to calculate the distance from the satellite to the GPS receiver. Several such distances may be used to compute a position.

Unlike direct measurement methods, like the tape measure, indirect methods (the surveyor’s wheel, RADAR, DME, GPS) the primary measurement (number of rotations, or time) is determined by more than just the distance. Usually these factors are small (wear of the wheel, changes in the speed of radio signals in the atmosphere, measurement paths that are curved) and may be correctable if enough information is known. Before these techniques were used to measure distances, these other factors had to be examined, understood and shown to be insignificant, or methods developed to deal with them.

The Cardinal Pass, and Off-Tuned Aural Null use received volume to measure distance or relative distance. In 1945 a Danish-American radio engineer named Harald T. Friis4 working at Bell Labs derived an equation used in telecommunications engineering that computes the power received by an antenna under idealized conditions given another antenna some distance away transmitting a known amount of power. The equation in the form appropriate when the receiving antenna is on a moving airplane specifies the following parameters:
  • the power output of the transmitting antenna;
  • the inverse of the square of the distance from the transmitter to the receiver;
  • the gain of the transmitting antenna in the direction of the receiving antenna;
  • the gain of the receiving antenna in the direction of the transmitting antenna;
  • the polarization vector of the transmit antenna;
  • the polarization vector of the receive antenna;
  • the equation does not account for multipath and other fading effects. 
Some of these values can be assumed to be constant. The power output of an ELT will be relatively constant over the period of a typical search unless the battery is in bad condition. The polarization vector of the ELT will also likely be constant, though there are stories of survivors manipulating the ELT orientation as search aircraft were visible to them. The polarization vector of the receive antenna can be held constant by only measuring volume when the search aircraft is flying straight and level. The remaining factors affecting received signal strength are:
  • the inverse of the square of the distance;
  • the gain of the transmitting antenna in the direction of the receiving antenna;
  • the gain of the receiving antenna in the direction of the transmitting antenna;
  • multipath and other fading effects.
We want the first factor to be able to detect closest approach during the Cardinal Pass, or equal distance during the Off-Tuned Aural Null. However if any of the other three factors, individually or combined, alter the signal strength significantly, and they certainly can, then the volume heard from the radio won’t indicate accurate distance.

There are some other factors introduced at the receiver and its operation. The Friis equation is only concerned up to the receiver antenna. These factors are:
  • The receiver frequency offset from the transmitted frequency. Larger offsets reduce the strength of the signal processed by the receiver.
  • The receiver automatic gain control, which tries to keep the signal being processed by the receiver at a constant level.
  • The volume control.
It is easy to see that when the antenna gain and fading factors are small these techniques will work. In the case of an actual crash or other emergency it is quite possible that the transmitter gain pattern and fading factors could be large resulting in these techniques failing. It can be extremely difficult for a pilot to arrange for a crash that results in a symmetric ELT transmission pattern, and is in an area with few reflectors.This knowledge comes along with an understanding of the appropriate mathematics and physics which the authors of Accuracy of ELT Searches1 decided they did not need to know. Again we wonder why the authors spent so much effort to ensure a circular pattern and limit the number of reflectors when searchers would not be able to do the same in a real emergency. The paper1 does not provide any quantification of these extra factors, nor does it provide any way to deal with them.

Consider the true Aural Null for a moment. If you have read our Practical Guide to Aural Searches, you know that an Aural Null is not performed by looking for locations with a particular signal strength, but by looking for locations on the radio horizon of the search aircraft from the transmitter. An ELT, even on 121.5 MHz has plenty of transmit power to reach LEOSAR (Low Earth Orbit Search and Rescue satellite) platforms in orbit, even if they are no longer listening. Reaching the radio horizon a few dozen nautical miles away is not a problem. Because we are using the radio horizon we won’t use the Friis equation but the line of sight propagation equation. This is much simpler having the following parameters:
  • the radius of the earth;
  • the refraction index of the atmosphere for the signal5;
  • the altitude of the search aircraft above the transmitter (we assume the transmitter is on or very near the surface). 
The radius of the earth is, of course, constant. The search aircraft can be flown at a constant altitude, and this is one of the specifications of the Aural Null. That leaves the refraction parameter. This parameter varies with atmospheric weather conditions. During most search flights the weather will not vary enough to make a significant change. So the radio horizon will be a constant distance from the transmitter while the aircraft is conducting the Aural Null. This is exactly what we want to be the case. The receiver frequency is not offset from the transmitter frequency because we want to receive all available signal strength to make detection of the radio horizon easier. The automatic gain control helps us by ensuring as much receiver gain as is necessary, but no more, is available. The volume control allows us to adjust the volume for comfortable listening to ensure our hearing does not become fatigued, again ensuring accurate detection of the radio horizon. All of this allows us to use the receiver to accurately map points on the radio horizon and use either Procedure A, B or C to determine the transmitter location.

Now that we know the factors, we can examine the results reported in Accuracy of ELT Searches1 and see what they tell us about the Cardinal Pass and the Off-Tuned Aural Null.

Cardinal Pass

Below is a map depicting the flight and computation of the ELT position during the Cardinal Pass.

Map 1: Cardinal Pass1

You can see the path of the airplane and the locations identified as the maximum signal points during first the West to East pass, and then the North to South Pass. It is interesting that the aircraft began to circle to the left immediately after the Max Long position. One would think, given the description of the Cardinal Pass that some amount of travel beyond the point to confirm passage of the max signal point. No matter. They determined the position1 to be 45° 37.79’ N 75° 56.45’ W. The actual position, taken from Google Earth, was reported1 as 45° 37.697’ N 75° 56.333’ W. So the technique was able to fix a location to within 0.125 nautical miles (or 760 feet, or 232 meters) from the known location. This is really quite good. Of course they knew where the ELT was located before the aircraft left on the search, and they had placed the ELT in a location that they thought would allow the technique to produce accurate results.

Off-Tuned Aural Null

Below is a map depicting the flight and computation of the ELT position during the Off-Tuned Aural Null.

Map 2: Off-Tuned Aural Null1

If you believe it looks similar to the Cardinal Pass map, you are correct. In fact it depicts the same flight path. This means that while the paper1 describes a number of benefits that accrue to this technique due to the way it was performed, this technique was, in fact, performed incidentally to the primary Cardinal Pass. And it suffered from that and other problems.

First the benefits that the authors suggest:

Note that using the three waypoints (ELT 1,ELT 2, ELT 3) in the Aural Null B procedure in this example works to the advantage of the Aural Null graphical method since, in general, we would not have these waypoints accurately without the use of a GPS and if we had not been using the Off-Tuning method in the first place.1

So yes, the coordinates of the off-tuned aural null points are more accurate because they used a GPS receiver instead of plotting the positions on the map. But could that not be done anyway, assuming a GPS is available? They continue:

In addition, the graphical construction shown here was done on a computer with a large screen and could not have been done to the same accuracy on the navigator’s lap in a small aircraft.1

If you have read our Practical Guide to Aural Searches - Polygon Methods you would know that an Aural Null computed from three points (a triangle) can result in only one circumcenter location, not a cocked hat as depicted in the image above and described in the paper: “The graphical solution is also shown with the most probable location (graph-ELT) being within the ‘cocked hat’.”1 This branch of mathematics, geometry of triangles, is one that all CASARA members should be familiar with. If the authors had been familiar they would not have confused the Aural Null procedures with triangulation and recognized their error. It seems that even with a computer and a large screen they could not achieve an accurate geometric construction. Their time might be better spent becoming familiar with the mathematics rather than assuming others share their lack of geometric skills. We have members, and have observed many navigators, who can plot an accurate Aural Null pattern while flying in small aircraft.

There is a mathematical formula relating the coordinates of the three vertices of a triangle to the coordinates of the circumcentre of the triangle. We will use that formula later. For now we note that the graphical construction resulted in a position of 45° 37.122’ N 75° 57.020’ W for a position error of 0.750 nautical miles1 (4557 feet, 1389 meters). At fully six times the error of the Cardinal Pass, In spite of the claimed advantages, the Off-Tuned Aural Null seems hopelessly outmatched. But is it?

First, there is a fourth candidate vertex that does not appear in the graphical construction:

An arbitrary decision was made to make a turn to the right to begin the start of chord 2. Since the audible signal did not reappear on 123.25 MHz while the aircraft was proceeding East, the right turn was continued until, upon passing the track on a Westerly heading, it was faintly reacquired for a very short time while passing across chord 1.1

No reason is given for not including this position in the Aural Null computation so we will consider it. Thus we have four points (ELT 1, ELT 2, ELT3 and ELT 4). We also note that, from the GPS track, the aircraft did not roll wings level while proceeding East. Recall from earlier that if the airplane is not wings level the polarity vectors of the transmitter and receiver antennas come into play adding more factors that can obscure the distance measurement. The signal was detected when the airplane rolled wings level to fly chord 2. An even better candidate may have been missed simply because they chose to continue the turn rather than roll wings level. These four points don’t meet the criteria for Procedure A and Procedure B takes only three points. We could use Procedure C, but for now let’s see what can be done with Procedure B.

We need to eliminate one of the points. There are a number of strategies that should work. We could apply the logical requirement that Procedure B is based on a triangle, and that ELT 1, ELT 2 and ELT 4 are all on chord 1. This suggests that ELT 2 should be discarded since it is in the middle of chord 1. We could compute all possible positions from all possible permutations of the four points taken three at a time; then eliminate the point that gives the highest error value. Our software to compute a Procedure C result does this. Again ELT 2 is identified as the point that should be discarded.

Finally we can use the properties of polygons that have circumcircles: simple, convex, cyclic polygons. What we need is a procedure that given a set of vertices, will order the set to form a simple, convex polygon and discard the vertices that do not belong. Luckily such a procedure was developed by Ronald Graham6. Our Procedure C software also uses the Graham Scan. Again ELT 2 is discarded.

The ultimate test however is the distance of each point from the transmitter location. Since we are hoping that all our points are on a single circle with the transmitter at the centre, the distance from each point to the transmitter location will tell us how good the data collection method is. Let’s look at some actual numbers.

Points Used

Position Result

Position Error

ELT 1, ELT 2, ELT 3

45° 37.171’ N 75° 57.028’ W

1327.47 meters

ELT 1, ELT 3, ELT 4

45° 37.643’ N 75° 56.704’ W

491.124 meters

Table 1: Positions computed from two sets of Off-Tuned Aural Null points.

Table 1 shows the results of direct computation of the Off-Tuned Aural Null position using the circumcentre formula for a triangle on the surface of the earth. The top row is using the same three points (ELT 1, ELT 2, ELT 3) as the graphical construction from the paper1. The error is reduced from 1389 meters to 1327 meters. Not a lot. The second row is using the three points selected by the Graham Scan and other methods (ELT 1, ELT 3, ELT 4). The error is now reduced by almost 1 kilometer to 491 meters. Quite substantial. We are puzzled as to why they would not use ELT 4, unless it is to make the Cardinal Pass look even better by comparison.


Distance from Transmitter

RMS delta Range


2398 meters

861 meters


2150 meters

1186 meters


3146 meters

1309 meters


2745 meters

797 meters

Table 2: Distance of each point from the transmitter location.

Table 2 shows the distance of each point from the transmitter location. Remember, if these are perfect Aural Null points they will all be the same distance. For each point we computed range from the transmitter. We then computed the root mean square of the differences between the range of that point and all others. This helps determine which points are better suited to use in an Off-Tuned Aural Null computation.

The point that is farthest from the mean is ELT 3. This would tend to indicate we should drop it, but that would leave us three points in a line which wouldn’t work. The next farthest point is ELT 2, which is also the point discarded by the Graham Scan. It is unclear why they selected ELT 2 and discarded ELT 4. What is clear is that their point selection criteria did not produce a set of points that are anywhere close to equidistant from the transmitter, which is required for the Aural Null procedures to produce an accurate result. There is in fact a difference of 1 kilometer between the nearest and farthest points. This is a substantial difference when the mean is 2600 meters. It appears that the graphic procedure did not benefit at all from off-tuning. In fact it is very difficult to measure, compare or even estimate distances based on received signal strength, even when one has the opportunity to design the equipment from scratch to perform that function. It is even more difficult to do with a receiver designed for a completely different purpose, like an aviation radio telephone transceiver. While in this case, where the transmitter location was carefully chosen for the technique, they were able to get an accurate position from the Cardinal Pass, eventually (if not frequently) an actual accident will place the transmitter in a location that will cause the Cardinal Pass to fail as the Off-Tuned Aural Null did.

This point is worth emphasizing. Aural Null Procedures assume that the positions used to plot the solutions are on a circle with the transmitter at the centre. Procedure A (shown at the left) uses a standard geometric solution to find the centre of the circle, and so the assumed position of the ELT. But what happens if some of the positions are not equidistant from the actual transmitter location. The procedure will still happily give you a position for the ELT, it will just be wrong.

An Aural Null Procedure A with one wrong position will result in an incorrect computed ELT position. Similarly a Procedure B with one wrong position will also give an incorrect computed ELT position. Procedure A has a benefit in that if the navigator computes all the bisectors, they will not intersect over either the true position or the erroneous position. So there is a chance, if the navigator runs this check, to find out that there is a problem. Procedure B provides no such checks.

In Accuracy of ELT Searches1 Barr, Casey and Muir seem to imply that the Aural Null is not as good as the Cardinal Pass even with the advantage of off-tuning. Unfortunately it is actually their method of collecting the positions that results in the poor performance of the Off-Tuned Aural Null.

Other Factors

Discussions between Mr Buckley and Dr Muir that pre-date either paper1 2 are quite illuminating as to how the thinking went astray. Dr Muir was (and may still be) involved in ski patrol activities and attended training on the use of Avalanche Transceivers “The more complex problems are well-known in avalanche searches where you have multiple (up to 4-6) casualties all with beacon antennae pointing in different directions and all radiating on the same frequency leading to multiple maxima, minima, and highly asymmetrical radiation patterns.7 The problem with comparing Avalanche Transceivers with ELT beacons is that Avalanche Transceivers operate on 455 kHz and usually have a range specification of 100 meters. This means that electronic avalanche search operations take place entirely within the near field of the transmitter. In fact within about one sixth of a wavelength. Translating this experience to the frequency of an ELT 121.5 MHz transmission would mean the experience is only applicable when the searcher is within 40 cm of the ELT. It is not practical to operate a search aircraft this close to an emergency beacon. Dr Muir goes on to say “...both Mike and I have tested the methods out at least a couple of times each, so there may be some merit to all this.” Our questions to Barr, Casey and Muir is this: were these tests conducted in similarly contrived circumstances as the one described in Accuracy of ELT Searches1? And: is a handful of tests really enough to discard the sum total of our knowledge of how radio works?

In fact it is quite well known that Avalanche Transceivers suffer very similar limitations to those that plague the Cardinal Pass and Off-Tuned Aural Nulls. Spikes in the signal pattern can cause significant location errors. Some materials surrounding the transceivers, including snow, can cause significant errors in ranging.

That is not to say there is not a method of accomplishing what Barr, Casey and Muir intended. For years multipath interference and fading were problems to be overcome. With the advent of techniques like those used in third generation cellphone technology multipath can actually be used to increase the range and reliability of those links. There is of course a method by which one may use a radiotelephone receiver to reliably determine the distance to a transmitter. Experienced practitioners of the technology should be able to deduce this method, so we leave it as an exercise for the reader.

Normalization of Deviance

Dr Vaughan developed her theory of the normalization of deviance by examining the Challenger Launch Decision. During the development, testing, redesign and retesting of the solid rocket boosters both Morton-Thiokol and NASA observed joint performance that deviated from expected performance. The problem was seen to be recurrent, but had no consequences, so they were able to convince themselves that the flaws were normal and acceptable.

The situation is very similar in the use of the electronic search techniques we have been discussing. Even though the results often deviate from expected or acceptable performance, and there are good scientific reasons for this deviation, there have been no consequences. The deviation has come to be seen as normal, even desirable even when it is actually deleterious to desired results.

In the case of NASA, Normalization of Deviance continued to be a problem playing a role in the Columbia disaster, and more recently the near drowning of an astronaut during a space walk. So unless and until CASARA implements a large and significant change in safety and management culture we can expect more instances where questionable techniques are used routinely. 


We believe Barr, Casey and Muir postulated the Cardinal Pass, and possibly the Off-Tuned Aural Null without considering the applicable mathematics, physics or engineering principles. The flight was carefully staged to showcase the Cardinal Pass technique. No attempt was made to implement either single or double blind protocols. These conclusions result from statements in their paper. In an attempt to show the Cardinal Pass in the best light, a set of points gathered incidental to the Cardinal Pass were cobbled together into an Off-Tuned Aural Null and presented as a Traditional Aural Null, to be a foil for the Cardinal Pass.

All of this makes it seem that the paper was not an honest attempt to present an unbiased investigation of the technique. Rather it seems to be an attempt to justify conclusions already arrived at, possibly from anecdotal observations. It seems reasonable that a similar set of circumstances lead to the adoption of Wing Blanking and Ad-hoc Direction Finding Antenna Placement.

Should volunteer search and rescue organizations use their funding to investigate techniques like this without competent technical oversight? How far have the ideas presented in Accuracy of ELT Searches1 penetrated into the accepted canon of CASARA training? The Advanced Cognitive Engineering Lab, mentioned earlier, was awarded a New Initiative Fund Grant of $2.6 million over three years to develop a training system for CASARA. What are the implications of Accuracy of ELT Searches1 on the quality of that system?

The most important point is not whether the Cardinal Pass or Off-Tuned Aural Null are effective techniques or not. Neither is it whether any of the techniques we have criticized should be used or not. It is quite clear, and professionals in the subject have clearly answered that they are not effective, and should not be used. The most important point is that time after time air search and rescue organizations have come up with techniques that have be used on actual searches without ever being scrutinized for efficacy, or used even after such scrutiny has uncovered problems. The tendency for this to happen puts lives at risk.

We present this as a cautionary tail. Search and Rescue volunteers should not adopt procedures based on conjecture without submitting them to a full and unbiased engineering or scientific study by persons qualified in the appropriate fields. As adults, these volunteers should recognize and accept the boundaries of their own qualifications, no matter how seductive a new idea may seem. Victims of an aircraft accident shouldn't have to depend on the unqualified conjecture or anecdotal experience of volunteers for timely rescue. In a recent article Randall Monroe posed this question “Would you rather bet a million dollars on a spacecraft engineer’s ability to successfully perform eye surgery, or an eye surgeon’s ability to land a probe on a comet?”


1. Barr A. Casey M. Muir L. R. “Improvement in Position Accuracy for ELT and Visual Searches” Civil Air Search and Rescue Association, Ontario Zone 12 (Ottawa), December 2010
2. Buckley H.R. “An Analysis of the Cardinal Pass Electronic Search Technique and Its Effect on Operational Effectiveness” September 2010
3. Potere D. “Horizontal Positional Accuracy of Google Earth’s High-Resolution Imagery Archive” Sensors 2008, 8, 7973-7981; DOI: 10.3390/s8127973
4. Brittain J.E. “Electrical Engineering Hall of Fame: Harald T. Friis [Scanning our Past]” Proceedings of the IEEE (Volume: 97, Issue: 9) pages 1651-1654
5. Haslett C. “Essentials of Radio Wave Propagation” Cambridge University Press, 2008 052187565X pages 119-120
6. Graham R.L. “An Efficient Algorithm for Determining the Convex Hull of a Finite Planar Set” Information Processing Letters 1 Pages 132-133 (1972)
7. Email from Dr Muir to Mr Buckley et al, March 2, 2010 4:41 PM EST