Engineering investigators are obliged to utilize the “Scientific Method” when conducting an investigation into a product failure. The basic elements are: Observe, Hypothesize, Test, and Conclude.

Occasionally, an investigator will obtain sufficient information from the “Observe” phase, so that only one hypothesis is plausible. In such instances, the “Test” phase is not explicitly necessary, and the “Conclude” phase is simply a confirmation that the sole plausible hypothesis is indeed the correct understanding of the problem.

With greater frequency however, during the early stages of an investigation, the investigator is faced with two or more plausible hypotheses and it would be premature to draw a conclusion without first performing tests and analyzing the results in order to rule-in or rule-out one or more of the competing hypotheses.

Some investigations are so complex that multiple hypotheses are on the table, and the available physical evidence and eyewitness information is not sufficient for an investigator to pinpoint a “single” causation scenario, even after physical and logical tests are performed. The appropriate conclusion in such cases is “undetermined”.

So what happens when two investigators come to different conclusions – one says the evidence points strongly to a single conclusion, and the other says there are multiple hypotheses that cannot be ruled out? To which expert should a jury listen? The expert whose hypothes(es) rank high enough on the quality scale!

From this investigator’s perspective, there are eight gradations of quality from low to high that can be used to illuminate hypotheses that should be “rejected”, “avoided” or “accepted”. We have created an illustration of this scale, in the form of a color spectrum, and it is pasted below, with the levels listed subsquently as text-only.

1. “Impossible”
2. “Contradictory Evidence”
3. “Speculation”
4. “Possible, but not Tested”
5. “Corroborating Evidence”
6. “Demonstrated Mechanism”
7. “Statistical Confidence”
8. “Proven or Certain”

As one can imagine, the last four levels (5 to 8) fall into the “Accept” category, which constitutes a “more likely than not” quality level. Levels 3 and 4 fall into the “Avoid” category – which means they can’t be ruled out, but insufficient supporting evidence is available. And finally, levels 1 and 2 apply to hypotheses that clearly fall into the “Reject” category. Collectively, the four “Avoid” and “Reject” categories constitute a “not likely enough” quality level.

As one example, consider an incident where a total of five hypotheses (“A” through “E”) have been proposed by two investigators:

Hypothesis “A” is at Level 2 – it is contradicted by some (not all) of the evidence.
Hypothesis “B” is at Level 3 – it sounds interesting, but has no basis beyond conjecture.
Hypotheses “C” and “D” are at Level 4 – they can’t be ruled out based on available data, but no validation testing has been performed.
Hypothesis “E” has supporting elements from Levels 5, 6, and 7 – the test data corroborates the hypothesis and none contradicts it; the scientific literature has published examples of similar prior incidents caused by a validated mechanism; and to a high level of statistical confidence, the hypothetical mechanism couldn’t have been caused by random variations alone.

In this situation, the five hypotheses (A to E) are not equal, so even though none of the five is officially “ruled out”, a conclusion of “undetermined” would be WRONG. In fact, only one of the hypotheses draws all of its support from the “Accept” category and no support at all from the “Avoid” and “Reject” categories.

Hypothesis “E” is the correct “more likely than not” conclusion, based on a logical evaluation of all the data – even though strictly speaking, it doesn’t rise to Level 8 – “Proven or Certain”.

Many investigators routinely apply a grading system informally (or even subconsciously) to such lists of suggested hypotheses in a given case. The spectrum presented here is simply a formalized representation of such systems that, at their core, comprise large doses of common sense.

The purpose of “Investigation Anecdotes” is to inform our readers about the intriguing field of engineering investigations. We hope you are instructed by this content, and we encourage you to contact us if you seek additional information. Copyright Martin Thermal Engineering, Inc. (2013)