Negative Outcomes
Positive observations evoke a similar sentiment. Sightings, successes, substantiations are all viewed as vindication with missing, contrary and disappointing results cast as aberrations or glitches. Humans prefer to be right (confirmation bias) but they can also misjudge the information value of tests (a high prior probability of positive) and actively suppress unpalatable truths. But sometimes a great deal can be learned from negative information, refuting scientific hypotheses when an expected event did not occur, and learning lessons from failures (or near misses).
Missing Data
Statistical methods to allow for unobserved information, subject to dependence assumptions on what is known, are advanced enough to focus on the suitability of the assumptions. In trials where such patterns are the reason for structured data collection, deliberately conservative approaches can be adopted, using allocations rather than treatments. But in general terms it is understood that setting aside information not available is poor practice leading to misconceptions of universality and sometimes substantial biases.
Logically, a view that observations of A predict an outcome B is equivalent to not observing B predicting not observing A (known as the contrapositive). But this is counterintuitive enough that people reliably do not identify the need to look for negative information to validate positive predictions. And scientific ideas have developed a philosophy (named positivism) based on refining theory by making falsifiable claims and testing them but never being able to prove their truth, only set higher bars to falsification.
And these are exemplified by the choice of framing to understand outcomes, without an appropriate contrast, seen in the Challenger shuttle crash in 1986. Data showing failure of o-ring seals only showed the temperatures of launches with positive observations, not those without (zero observations). Choosing to look at when an event has happened is different to whether it has happened, similar to the contrast of isolating positive cases of an infectious disease with monitoring test positivity in the population.
Learning from Error
In contrast to science where we are unsure and trying to find out, in many operations failures are the features that systems are designed to avoid. Whether these are mistakes, deterioration or unexpected events, quality and safety protocols are designed to assure outputs and prevent risky actions. Despite all of these precautions, risk monitoring only mitigates some errors and must manage others, so data is collected to better inform assessments and adherence to practical measure of control.
Some industries are so advanced as to simulate problems, because they are too rare to expect to be seen in training, or too catastrophic to be tolerated. But most adverse events in medical care are recorded, deaths are investigated by coroners, and fatal accidents reviews make recommendations. Clinical research protocols also require demonstration of safety in healthy patients before as treatments, and in use drugs are subject to event reporting, from side effects to cross reactions.
But there is a problem collecting data about negative outcomes, because culpability does not promote reporting, and so can even lead to institutional coverups. Psychological safety in being able to report problems, to learn from them, is seen to lead not only to higher reporting, but also better outcomes. Patterns where the stakes are high but culture not conducive to sharing difficulties have seen lower error reporting but worse outcomes; by contrast where these are rare enough ‘near misses’ also offer lessons.
Efficient Learning
Science has struggled with the desire to validate laws of nature with doubts that negative results can be due to poor experimental design or other extraneous factors. Equipoise and differential diagnosis are taken up in efficient clinical approaches, so that a test gives useful information in a similar amount in deciding between two diagnoses, or the superior efficacy of different treatments. Scientists have taken up cudgels against poor practices leading to implausible and unreliable reports of positive outcomes, setting more severe tests of their hypotheses. But sometimes the negative outcome is more informative, because it is rarer, unexpected, or entirely novel, and these are all an opportunity to learn.