Uncertainty and Risk in Civil Engineering Practice

Life is short and the art long; the occasion instant, experiment perilous, decision difficult. (Hippocrates as quoted in Fox,)

See also Project Definition , Project Life Cycle and Phase Models , Project Delivery Methods
See also Technical Outcomes for Civil Engineering:Risk and Uncertainty

Basic Considerations

As human beings, ... being alive means seeking opportunities and taking risks. As knowledge professionals living in the 21st century, this means coping with an increasingly complex number of uncertainties for humans living in this environment. We seek to understand better how these uncertainties can be characterized and managed. The essence of this article and the experience for engineers in general and civil engineers in particular is that manageable uncertainty is, by definition, termed risk, and its kindred cousin is termed hazard. This causes us to experience the

" ... human dread of and fascination for risk and the increasingly important role of risk analysis within societies ... " (Ibid.)

and, by extension, civil engineering. McDaniels et al. argue that risk management has been fundamental to our social and governance development for the past 10,000 years. The scale and shape of the uncertainties faced in this period shaped the societies that have developed today. The central thrust of this effort over the centuries has been to reshape and re-frame our understanding and conception of uncertainty from one of complete unknowing and simple acceptance as our fate in life to one of management (cf"Against the Gods" concept in Bernstein's book [21).

All the knowledge professions and disciplines have struggled with managing uncertainty, for it is impossible to manage unbounded uncertainty. [Note 1]

As outlined below, professions such as civil engineering have been successful in acting in the face of uncertainty lies in the profession's ability to reduce a broad variety of uncertainties into a series of increasingly smaller and crucially bounded subsets that can be managed. These are called 'risks. More recently, this process has evolved, and individual disciplines such as Civil Engineering (CE) have developed their knowledge and models to perform risk analysis. Risk is also taught as a distinct discipline and specialty practice within civil engineering. Still, it has some unique features that set it apart from other more classical practices within CE. As such, it is one of the first of some very specialized CE practices that use knowledge about civil engineering knowledge, or meta-knowledge. Other examples of civil engineering metaknowledge are project controls and quality controls.

Semantic, Epistemic and Logical frameworks

The semantic, epistemic, and logical frameworks for uncertainty and risk have several dimensions and layers of logical frameworks. The semantics problems include interchangeable usages for risk and uncertainty and hazard, uncertain and imperfect, and a lack of definitional material or context. The semantical scheme for this article will be to proceed from the distinction between certainty and uncertainty through marginal refinements and reductions of uncertainty up to the point of causal uncertainties. Beyond this point of semantics will be the logic frameworks or the subject of taxonomies of professional knowledge. First, it is about simple, testable phenomena, and then, it moves on to complex taxonomy schemes for artificially constructed phenomena or engineering projects.

Semantics framework

Certainty and Uncertainty

There is no such thing as absolute certainty, but there is assurance sufficient for the purposes of human life. (John Stuart Mill)
If you tried to doubt everything, you would not get as far as doubting anything. The game of doubting itself presupposes certainty. (Wittgenstein # 115 from On Certainty)
It is important to note that the references to epistemic or epidemiological knowledge in this article are assumed to relate to three forms of knowledge, namely:

• Knowledge that ( descriptive or declarative or propositional knowledge)
• Knowledge how ( or "know-how"), and
• Knowledge by acquaintance.

Certainty has been defined as "an epistemic property" of knowledge in all its forms and the state of our beliefs about that knowledge. Certainty about any belief about knowledge in any form implies that it is not subject to doubt or skepticism. This immunity to criticism can be dogmatically based, emphasizing the importance of a propositional-based sense of truth over experiential, sensory perceptions.

The demarcation line between dogmatic and non-dogmatic beliefs lies in the presence and recognition of specific criteria and information that would make the believer change their beliefs. There is an underlying requirement for an empirical framework of testability and falsification to recognize this and limit dogmatism. For example, one could argue that they would change their minds if God asked them to do so. Similarly, one could construct a concept of epistemic as opposed to dogmatic belief certainty as the ability to know anything that one chooses to know and can be known or inherent omniscience.

A second kind of certainty is epistemic, when conviction reflects the highest possible support for a belief. In this sense, knowledge is separate from beliefs, although someone may have beliefs about a property, such as certainty of knowledge. Logically, it has been shown that for any such system of knowledge, there will be statements that are unprovable within the system. Secondly, the knowledge system cannot demonstrate its own consistency. ([incompleteness theorems])

Certainty, in real life, is useless or often damaging (the idea is that "total security from error" is impossible in practice, and a complete "lack of doubt" is undesirable) ([Physicist Carlo Rovelli])

Uncertainty, on the other hand, arises immediately in the slightest amount of doubt or criticism. Similarly, a general definition of uncertainty could be given as "...any departure from the unachievable goal of complete determinism." Reasoning under uncertainty is very different than performing the same under certainty. In reasoning under certainty, one has complete knowledge and deduces without doubt and equally important, without limitation, thereby concluding free from error. Reasoning under uncertainty, one works in a state of incomplete, inconsistent, and limited knowledge; doubts cloud any statements or assertions and thereby taint any deductions/inferences resulting in the potential for error.

Knowledge professions reason in a state of incomplete, inconsistent and limited knowledge with doubts that cloud any statements or assertions and taint any deductions/inferences resulting in the potential for error.

It is also important to note the semantical and logical correlation between 'reasoning under uncertainty' or 'acting in the face of uncertainty' with the 'potential for' or 'presence of' error. The presence of uncertainty is invariably linked to the potential presence of error. The contingent nature of uncertainty logically implies the contingent nature of error.

Knowledge Professions use error as a proxy for uncertainty such that within discipline knowledge frameworks, uncertainty can be managed. The rationale for this belief is that error can be reliably described, quantified, explained, and ultimately reduced in ways that uncertainty cannot.

Uncertainty implies a range of variation that is impossible in certainty. Concepts and definitions of uncertainty are unbounded and vast but start from the point of complete and total ignorance. In describing uncertainty, statements can range from complete ignorance to relatively high degrees of belief that there is less potential for error in our reasoning or action. The latter is based upon confidence in justified knowledge, favorable past experience, and reliable parameters/models. One can't make that statement about ranges in certainty. If certainty, by definition, precludes doubt of any type or nature, then how could that be graduated to any degree?
Uncertainty, on the other hand, can be reduced through disciplined efforts to acquire knowledge. In short, we can reduce or even manage uncertainty (to a degree); how can certainty be improved? Are there higher degrees of perfection? The answer is no. We reason from an uncertain starting point. Yet this effort presumes, as Mill and Wittgenstein postulated, that we can develop or acquire an increasing degree of belief or conviction that there is less potential for error in our reasoning or action. This position assumes that some way exists to identify what a lesser degree of uncertainty would look like.

Knowledge professions acquire an increasing belief or conviction that there is less potential for error in their reasoning or action based on justified knowledge, favorable past experience, and reliable parameters/models.

Uncertainty is, in some form, amenable to description and explanation through observation and analysis. It is explainable and predictable to such a degree that it interests professionals such as scientists and engineers. Professional knowledge of this type has explanatory or predictive power, albeit limited in scope and application to relevant subjects such as physical objects or phenomena and functionality.

The very act of reducing the scope of uncertainty to a limited set of phenomena implies choice and, ultimately, the decision to act in the face of such uncertainty, but acting or judging in this manner adds more value than doing the same under relative ignorance.

Uncertainty, as described above, implies a hybrid nature. Examples of this are economic or decision-theoretic applications. In economics, perfect information allows one within the limiting framework of perfect competition in economic games to make decisions with 'perfect knowledge.' The crucial difference here is the distinction between 'perfection,' broadly, a state of completeness and lawlessness, and 'perfect, which can be thought of as an analytical limit that is approachable but can never be attained. Arbitrarily delimiting the realm of knowledge into defined, finite boundaries allows the simulated production of perfect choice, assumed crucially, to be free from error. The hybrid nature of this form of 'simulated' certainty produced within boundaries defined by assumptions and, therefore, clouded by doubt is itself a form of uncertainty. The importance of this is the semantical and logical association that exists between certainty and perfection versus the hybrid concept of a perfect anything, whether knowledge, infraction, competition, etc., or in the case of engineering, elastic, permeable, conductive, etc., being contained within an admittedly imperfect matrix of uncertainty.

Simply put, a finite, heavily bounded piece of uncertain knowledge can be improved through the simulated use of assumedly perfect information, but the perfection of knowledge cannot be attained or simulated.

Engineering and Economics, for example, simulate finite elements of perfect knowledge or information, heavily bounded with assumptions and a range of applications. By doing so, these professions reduce the bounds and variability of uncertainty and thereby manage it. Using these hybrid models to acquire knowledge takes place in an economy (regulated or institutional) and environment (transparent and structured), which puts pressure on the various professions to recognize and use qualitative and quantitative knowledge better to reduce uncertainties in their activities. The challenges in this reflect the nature of the profession's or discipline's knowledge (e.g., scientific, engineering, or medical), the policies and structure of the relevant professional knowledge bodies, process objectives and constraints, and "the elusive demands of politics." [6] One common theme runs through all of these efforts to address the presence of uncertainty in professional activities.

"The available scientific information upon which a decision must be made is almost always a mixture of engineering knowledge and uncertainty. Regardless of the information or methodology, there is the potential for error. This is true regardless of the relevant science, whether archaeology or aeronautics, economics or engineering, pharmacology or toxicology or epidemiology. Achieving the best social decisions requires not only understanding and using what we know but also appreciating and weighing the extent of our uncertainty. In making the best use of ... information in ... decision-making, it is still true that the beginning of wisdom is knowing what it is we do not know." (Vern, Op. cit., Reformatted and engineering substituted for scientific)

Driven by practical objectives and benefiting from past experience, knowledge professions such as engineering start by reasoning from uncertainty to develop explanations, calculate, and make predictions. This professional knowledge converges on but never reaches certainty, producing a "legitimacy" from the fruitfulness of its use.
Top of current page

Bias and Uncertainty

Up to this point, the concept of error, as outlined in uncertainty, could easily be simulated by a truly random variable. There should be no detectable differences around the unseen or specified statistical mean. However, several instances in experience point to a tendency towards one extreme or a pattern of error, a collective that is in itself an error of errors, namely bias.

Arguably, this could be part of the choice uncertainty discussion below, but it makes more sense to be looked at on its own. Acting in the face of uncertainty is an exercise of personal knowledge or experience. Whether using professional knowledge or individual experience, the potential is there to make a series of choices that, in some instances, make errors more likely than they would be if considered in the context of a statistical analysis.

This tendency for an individual to make the same error in a predictable or repeatable manner is termed bias. Personal experience, beliefs, knowledge, data, and models all have inherent built-in factors, errors, or defects that create predictable error patterns when applied in decision-making. Walker (1998) argues implicitly that such bias can also be considered systemic error. This type of uncertainty is difficult to identify except on theoretical grounds when the magnitude and direction of the bias are known. Most of these biases, whether systemic or knowledge-based, individual or practice-based, require disciplined efforts to "validate" such methods, models, and judgments at all layers while bounding the uncertainty.

This is the challenge to professional knowledge that must be overcome, namely, developing and validating models and methods for managing bounded uncertainty that minimize this tendency towards repeatable groups of error or bias. Therefore, from this point forward, the discussion will refer to 'uncertainty and its proxies, error, and bias.'

Propagating Error and Uncertainty

Partitioning uncertainty and its proxies, error, and bias into logically distinct subsets involves or is associated with multiple-step execution or choice frameworks. Combining those measurements or choices into single parameters or choices involves, in some cases, assembling the data/information in a unique or limited range of sequences. This implies that associated errors in observing, calculating, or choosing will have unequal impacts or consequences depending on where in the chain the error occurs. Also implied in this is the possibility that subsequent errors will be causally linked to some degree. (In statistical practice, this depends upon whether the errors are independent of each other or correlated, specifically, co-variant.)

Professions develop knowledge of how uncertainty and its proxies, error, and bias propagate through an ordered and normative series of observations or choices. This allows the profession to prioritize error and bias reduction to achieve an optimized reduction of uncertainty.

Top of current page

Cognition, Metacognition and Uncertainty

The concept of choice and the resultant error in uncertainty can be further delimited depending on the role of cognition, meta-cognition, and independent external elements. The rationale for this argument is that such choice activity in the face of uncertainty is a conscious, cognitive act.

An argument could be made that such cognition is a prerequisite for the exercise of choice and the introduction of error, as presented and discussed above. The counterfactual to this argument is the example of uninformed choice. Choices introduce as much as error or more as informed choices could be made. (???) Choice, in this sense, is indifferent to knowledge. Purposeful choice in an environment of objectives requires knowledge, method, and experience, but uninformed, speculative choice does not.

" Cognition is the set of all mental abilities and processes related to knowledge, attention, memory and working memory, judgment and evaluation, reasoning and "computation," problem-solving and decision making, comprehension and production of language, etc. Human cognition is conscious and unconscious, concrete or abstract, as well as intuitive (like knowledge of a language) and conceptual (like a model of a language). Cognitive processes use existing knowledge and generate new knowledge." (Accessed at Wikipedia)

"Metacognition is "cognition about cognition", or "knowing about knowing" and can take many forms. It includes knowledge about when and how to use particular strategies for learning or problem-solving. There are generally two aspects of metacognition: knowledge about cognition and regulation of cognition."

Some types of metacognition knowledge are:

Person knowledge (declarative knowledge) which is understanding one's own capabilities.

Task knowledge (procedural knowledge), which is how one perceives the difficulty of a task, which is the content, length, and type of assignment.

Strategic knowledge (conditional knowledge) is one's capability to use strategies to learn information. (Accessed at Wikipedia)

Like metacognitive knowledge, metacognitive regulation or "cognitive control" contains three essential skills.

Planning: refers to the appropriate selection of strategies and the correct allocation of resources that affect task performance.

Monitoring: refers to one's awareness of comprehension and task performance.

Evaluating: refers to appraising the final product of a task and the efficiency at which the task was performed. This can include re-evaluating strategies that were used. (Accessed at Wikipedia)

Nothing has been said up to this point that limits this structure to an individual, or a collective, or two entities making cognitive choices that are incompatible or inconsistent to varying degrees. Interpretations of this scheme include an individual professional exercising judgment in a decision that, in turn, relies upon the collective judgment and decision of the discipline as an underlying basis versus the decision of another individual to oppose/protest or otherwise contest that decision. Thus, the framework of cognitive/meta-cognitive applies to individuals, collectives, and controversies. In economic theory, this could be recast to cognitive/meta-cognitive roles in the theory of rational choice, economics of regulatory practice, and game theory.

By solving useful problems in the face of inherent uncertainty, discipline-specific frameworks of cognitive/meta-cognitive activity apply to individuals, collectives, and controversies. In economic theory, this could be recast to cognitive/meta-cognitive roles in the theory of rational choice, the economics of regulatory practice, and game theory.

Top of current page

Probability and Uncertainty

Probability is a coherent approach to uncertainty in the physical and mathematics or "classical" domains such as engineering. [8] and is the "...is certainly the best-known and most widely used formalism for quantifying uncertainty.[9] Cox’s theorem ("Any measure of belief is isomorphic (but not necessarily equal) to a probability measure") is a well-known argument for the validity of that argument. [Note 5]

Another writer, Knight (1921,1956), presented the following taxonomy of probabilities: [10]

Classical 'a priori' probability: As an idealized model, numerical probabilities are computed based on generalized principles of assigning equal likelihoods and mutually exhaustive possible outcomes (Runde, 1998) to all set members. Such probabilities are assigned to "...outcomes based on a judgment of indifference between those outcomes, that is, based on the absence of any evidence of real influences in play that may render any one outcome more or less probable than any other." (Runde, op. cit.) This builds upon Knights's concept of an 'absolutely homogeneous classification of instances that are completely identical except for indeterminate factors. This judgment of probability or logical probability is on the same logical plane as the propositions of mathematics and ultimately inductions from experience.'(Knight, 1956, pg.225)

Bayesian 'a priori' probability: In contrast to interpreting probability as the "frequency" or "propensity" of some phenomenon, Bayesian probability is a quantity that we assign for the purpose of representing a state of knowledge or a state of belief. In this view, a probability is assigned to a hypothesis, often using the basis of past or prior experience, whereas under the frequentist view, a hypothesis is typically tested without being assigned a probability based upon prior experience. The Bayesian interpretation of probability can be seen as an extension of propositional logic that enables reasoning with hypotheses, i.e., propositions whose truth or falsity is uncertain. Bayesian probability belongs to the category of evidential probabilities; to evaluate the probability of a hypothesis, the Bayesian probabilist specifies some prior probability, which is then updated in the light of new, relevant data (evidence). The Bayesian interpretation provides a standard set of procedures and formulae to perform this calculation. [11]

Statistical or 'a posterior probability: Empirical evaluation of the frequency of association between predicates, not analyzable into varying combinations of equally probable or logical, 'a priori' alternatives. Knight argued that any "high degree of confidence" that the judgment that experience will remain valid for future predictions is "...still based on an 'a priori" judgment of indeterminateness." (Knight, Op. Cit.) Knight's argument to support this is what will be discussed further below that first, "...the impossibility of eliminating all factors not really indeterminate; and, second, the impossibility of enumerating the equally probable alternatives involved and determining their mode of combination so as to evaluate the probability by a priori calculation." (Knight, Op. Cit.) Knight noted that the main difference between this form and that of logical or inductive probability was the presence of empirical information, which allowed the analyst to identify propensity and direction in the observed phenomena. In this case, 'a priori' probability values may be derived from first principles and "statistical probabilities are determined a posteriori by the empirical method of counting instances." (Runde [1998] quoting Knight, op. cit.)

Estimates: The distinction here is that there is no valid basis ('a priori' or 'a posteriori') for any classifying phenomena. For Knight, this form of probability presented the greatest logical difficulties of all ... but its distinction from the other types must be emphasized, and some of its complicated relations indicated..." (pp. 224-5, emphasis in the original)

Arguably, Knight's concept of probability can be viewed as "..a continuum of probability situations, depending on the degree of homogeneity of the 'instances' in question" [12] ; i.e. going from a logically distinct but equally likely set of elements to a unique set with one member with statistical frequency in the middle. Knight's main motivation for distinguishing between a priori probability and statistical probability underscores his opinion that "(i) the 'mathematical or a priori type of probability is practically never met with in business, while the second is extremely common'; and (ii) 'the statistical treatment never gives closely accurate quantitative results' (Runde quoting Knight pp. 215-16). Runde argues that Knight couldn't conceive of (empirically tabulated) occurrences that are used in daily commerce classes of instances that we have to make do within the course of everyday economic life could be divided into subclasses of instances that are sufficiently homogeneous to permit the determination of what he calls 'real' probability (p. 217).

In modeling degrees of uncertainty, any measure of belief is isomorphic but not necessarily equal to a probability or statistics measure.

Top of current page

Explanation, Prediction and Uncertainty

It should be clear that the general notion of uncertainty synonymous with complete ignorance must be bounded as part of a scheme to produce a useful concept of uncertainty in professional practice such as civil engineering.

The first principle that must be introduced is that such uncertainty must be bounded by reliable and valuable knowledge gained from past experience combined with analytical and computational capabilities to address an immediate and real problem of interest. This could be viewed as the economic interest argument. Only uncertainties associated with physical phenomena and economic scarcity will be addressed.

The second principle is that the knowledge model is capable of producing statements or assertions in the form of hypotheses that themselves are testable. Testability, in this sense, is defined as the property applying to an empirical hypothesis and involves two components:

The logical property that is variously described as contingency, defeasibility, or falsifiability, which means that counterexamples to the hypothesis are logically possible. Contrast this to a logical tautology, which is always true that is true in every possible interpretation. The practical feasibility of observing a reproducible series of such counterexamples if they do exist. In short, a hypothesis is testable if there is some real, non-zero expectation of deciding whether it is true or false of verifiable, reproducible experience. Upon this property of its constituent hypotheses rests the ability to decide whether a theory can be supported or falsified by actual experience data. (Source Wikipedia.)

Knowledge models must be capable of producing expressions that can explain phenomena (physical, economic, and social) that meet the testability criteria described above in 'reasoning under uncertainty' or 'in the face of uncertainty' with the 'potential for' or 'presence of' error and bias.

Variability and Aleatory Uncertainty

The discussion up to this point has focused on uncertainty associated with various forms of knowledge referred to as epistemic uncertainty. Acting in the face of epistemic uncertainty introduces choice uncertainty (discussed below) and results in the production of explanations and predictions. It is implied that there will be testable hypotheses in the form of outcomes of interest to the knowledge professions. Inherent in nature is a variation of results, which may appear to uncertain but isn't. Observations of phenomena will not be the same, and part of the experimental process is to narrow those results down to within an acceptable level of tolerance. Statistical measures seek to identify underlying and unobserved measures of central tendency. These parameters are not directly observed but are derived to a degree of confidence. In some cases, variation around a mean or within a statistical range may be perfectly acceptable to the knowledge profession. Examples of this are in the physical sciences and medicine. In Engineering and, to a lesser extent, economics, the process or phenomenon may be required to be more controlled to a narrower range than what a natural or "unmanaged" amount of variation would allow. Several courses of action present themselves.

The first would be to ignore the aleatory phenomenon and choice, which is rejected outright.

The second would be to assume that, in effect, such variation is itself a form of minimally managed uncertainty and treat it the same as epistemic uncertainty. This is not a desirable outcome as reducing Aleatory error and bias requires

A third approach would be to acknowledge that process variation is, in some forms, manageable. This means making choices and acting in a way that presumes that the natural occurrence of error can be reduced, or more importantly, bias can be reduced. There is ample precedent for the third approach in civil engineering design and project management. In this sense, the argument is made that the aleatory response to managed efforts results in an elastic variation range. Presumably, an effective management effort either "shifts" the central tendency of the process results or reduces the variation range of occurrence or some other population parameter. Choosing to act and accept that narrower range rather than the broader natural variation may result in an outcome that, while well within the bounds of the expected variation, is outside the arbitrary control limits established by the choice and action. While this phenomenon mimics the broader uncertainty and its proxy error, it is not an error. The information on the variation was known, and the ability to produce narrower results was an error, albeit a pseudo-error, when compared to the broader classes of uncertainty discussed above, such as total ignorance.

Looking at uncertainty using this third approach means that the error in choosing a new target parameter for this process may contain epistemic errors (our knowledge of the phenomenon may be erroneous, or the model flawed), aleatory uncertainties as discussed, or, more importantly, choice error or bias in thinking that we can move the process results to meet the requirements.

In civil engineering practice, this may mean applying professional judgment in developing partitioning schemes or allocating uncertainty between the three dimensions (Epistemic, Aleatory, and Behavioral) of uncertainty.

Lastly, raising the topic like this brings the questions of effectiveness and efficiency to the fore. Advancing knowledge models that reduce error and bias at first are largely choice models resulting in justifications and favorable experiences. At some point in the process, the knowledge profession advances explanations and predictions that require reliable mathematical or logical concepts that are reduced to models and parameters. The question of efficiency becomes important; the broader uncertainty may have been reduced, but the process is economically inefficient. The profession is urged to become more frugal or, for example, reduce the variability of the outcomes. Knowledge professions often operate under conditions that require them to acquire knowledge and reduce uncertainty effectively and efficiently. Doing so requires understanding both epistemic uncertainty, as discussed above, and statistical process variation or aleatory variation.

Knowledge professions manage and reduce uncertainty in the form of error and bias in an economically effective and efficient manner. This effort requires understanding both epistemic and aleatory uncertainty.

Top of current page

Choice and Behavioral Uncertainty

The discussion up to this point has focused on uncertainty associated with various forms of knowledge, referred to as epistemic uncertainty, as well as the inherent variations in results, or what was termed aleatory uncertainty. Acting in the face of that epistemic and aleatory uncertainty introduces yet another uncertainty into the process, namely choice uncertainty. This uncertainty reflects uncertainties associated with first-person and third-person choice. This is particularly relevant for engineering with project stakeholders.

Summary of Semantics Framework for Uncertainty

Professional disciplines have gradually re-framed their understanding of uncertainty and, thru experience and analysis, have arrived at the following meta-knowledge concepts of uncertainty:

• Uncertainty and its proxy, error, have three components: epistemic uncertainty, aleatory uncertainty, and behavioral uncertainty.
• Driven by practical objectives and benefiting from past experience, knowledge professions such as engineering start by reasoning from uncertainty using non-dogmatic beliefs, which recognize the existence of specific criteria and information that would make the believer change their beliefs using coherent approaches such as probability.
• Knowledge professions develop knowledge of how uncertainty and its proxy, error, propagate through an ordered and normative series of observations or choices that meet testability criteria in 'reasoning under uncertainty' or 'in the face of uncertainty' with the 'potential for' or 'presence of' error. An error can then be described, explained, and reduced in ways that uncertainty cannot. This allows the professions to prioritize error reduction to achieve an optimized reduction of uncertainty and produce a "legitimacy" from the fruitfulness of its use.

Top of current page

Uncertainty versus Risk/Hazard

Up to this point, nothing has been said about risk or hazard. Now, risk can be defined. Simply put risk or hazard are finite, logically distinct subsets of the broader uncertainty. Risk and hazard are currently defined interchangeably. First and foremost, risk is a specific subset of general uncertainty. Any arbitrary subset of general uncertainty can be defined as a risk if it meets the following criteria:

• Uncertainty is associated with the potential for material error when acting or choosing in the face of such uncertainty in an environment of physical, economic, and social phenomena.
• Uncertainty and its proxy, error, can be partitioned into logically distinct subsets associated with multiple-step execution or choice frameworks in physical environments linked to a unique or limited range of sequences.
  • Subcriteria 1:Resulting errors will have varying impacts or consequences depending on where the error occurs and the degree of causal linkage or influence in the chain.
  • Subcriteria 2:Potential errors can be sequenced and prioritized to achieve an optimized reduction of error or bias and, by proxy, uncertainty over time.
• Uncertainty and its proxies, error bias, and environments are isomorphic to coherent approaches such as probability.
• Uncertainty and its proxies, error, and bias are isomorphic to coherent approaches such as knowledge models capable of producing expressions that can explain phenomena (physical, economic, and social) that meet the testability criteria in 'reasoning under uncertainty.'

This is a definition of general risk. There is no differentiation for engineering or economics, medicine or insurance. There is no difference between a 'good' risk consequence and a 'negative' one. Redefining general risk into simpler terms gives the following:

General Risk is a subset of general uncertainty that is associated with the potential for material errors and biases when sequentially acting or choosing in the face of such uncertainty in an environment of physical, economic, and social phenomena; isomorphic to probability measures and methods and capable of testability.

Top of current page

Logical Framework for Specific Risk Taxonomies

As noted above, at the most fundamental level, uncertainty is ".....the subjective perception of ignorance." [13] [Note 6] At this point in the discussion, the focus has moved from discussing uncertainty to its framed subset, risk.

Further, the discussion moves from general risk to a professional or discipline-specific risk taxonomy. Beyond this point, risk will be used in place of uncertainty.

Han et al. note that, in general, "(t) taxonomies are valuable not only in their comprehensiveness but in their coherent reduction of uncertainty to conceptually discrete elements." Uncertainty (and, by implication, risk or hazard) taxonomies offer an approach or tool to more precisely identify and define ontological uncertainty so that it can be quantified, analyzed, and communicated. Viewed this way, uncertainty is not a single, monolithic phenomenon but "...multi-dimensional with theoretically distinct domains and constructs that are potentially measurable and related to different outcomes, mechanisms of action, and management strategies." [14] Taxonomic schemes for classifying the different kinds of scientific and, by extension, engineering uncertainty require identifying the various kinds of potential error associated with descriptive scientific or engineering information and knowledge. An example of this is the following quote from a legal authority:

"I would especially stress the need for an agency to disclose the uncertainty that surrounds its determinations. And by uncertainty, I mean the agency's ignorance as well as its quantitative estimates of error." (Emphasis added) [15]

Given the necessity of acting in the face of enormous uncertainties [16], discipline taxonomies must offer a clearly coherent and probabilistic approach to uncertainty that is sufficiently general in nature, logically distinct and exhaustive in scope and "...provide decision-makers with a foundation for understanding the nature of discipline information..." [17] Top of current page

Scientific Risk Taxonomies

The scientific method is a body of techniques for investigating phenomena in its environment, acquiring new knowledge, or correcting and integrating previous knowledge. Inherent in that process are uncertainties of many forms. Any one taxonomy of scientific uncertainty has largely been focused on statistical models used to assess and quantify sampling, errors, and parameters, although several different taxonomies are conceivable: note-24 [18]

Environment-specific, Phenomena oriented, model-centric taxonomy of uncertainties

• • Parameter uncertainty the source of which is model parameters that are inputs to the computer model (mathematical model) but whose exact values are unknown and cannot be controlled, or whose values cannot be exactly inferred by statistical methods. Subsets of this type of uncertainty include but are not limited to:
    • Experimental uncertainty is also known as observation error, which comes from the variability of experimental measurements. An example is repeating an experiment measurement several times using exactly the same settings for all inputs/variables and recording the variability. It is a subset of the larger uncertainty component, parameter uncertainty.
    • Parametric variability comes from the variability of the input variables of the phenomena model and is also a subset of the larger uncertainty component, parameter uncertainty.
  • Structural uncertainty, or model inadequacy, model bias or "systemic bias", or model discrepancy, which comes from the lack of knowledge of the underlying true state of the phenomenon or its environment. It depends on how accurately a mathematical model describes the true state of the phenomenon or what is termed "model fitness". Due to the inherent nature of uncertainty in any knowledge body (scientific or engineering), models are only an approximation to reality. Therefore, any conclusions drawn from the model can be very misleading when the underlying basis is not plausible or lacks validity. note-25 [19]]
    • What is missing in this context is any concept of output variability due to the model itself. Part of this is due to the lack of recognition of "process" in scientific investigation. The concept of the scientific method started with a single investigator, such as Galileo or Michael Faraday, working in their laboratories. It has grown into planet and solar system scale experiments such as data gathered on planetary flybys such as Mars, Pluto, and Ceres to the hunt for the Higgs Boson. Ultimately, science has become more like engineering in the scale and complexity of its investigations and theory aggregates, such as the "Unified theory" of particle physics. Subsets of this type of uncertainty include but are not limited to:
    • Algorithmic uncertainty, or numerical uncertainty, comes from numerical errors and numerical approximations per implementation of the computer model. Most models are too complicated to solve exactly. For example, the finite element method] or finite difference method [may be used to approximate the solution of a partial differential equation], which introduces numerical errors. Other examples are numerical integration and infinite sum truncation, which are necessary approximations in numerical implementation.
    • Interpolation uncertainty comes from a lack of available data collected from computer model simulations or experimental measurements. For other input settings that don't have simulation data or experimental measurements, one must interpolate or extrapolate to predict the corresponding responses.

Top of current page

Commentary

A more comprehensive taxonomy of uncertainties that acknowledges knowledge, data, and linguistic uncertainties is:

• Epistemic uncertainty -Epistemic uncertainty or systematic uncertainty, in contrast, reflects limitations in the current “state of knowledge” underlying models themselves, originates from competing theories or models, is not readily quantifiable, and is manifest by subjective confusion or indecision. It includes uncertainty due to measurement limitations, insufficient data, extrapolations and interpolations, and variability over time and space. Epistemic uncertainty is uncertainty about "...some determinate fact ... because of a lack of complete information. cite_note-26 [20]] Epistemic uncertainty can be classified into six main types:cite_note-27 [21]]
• Measurement error - Measurement error is uncertainty that manifests itself as (apparently) random variation in the measurement of a quantity. Repeated measurements will vary and demonstrate classic statistical behavior.
• Systematic error - Systematic error occurs due to bias in the measuring equipment, model, analysis, observation, or sampling procedure. It is formally defined as the difference between the true value of the quantity of interest and the value to which the mean of the results converges as sample or data sizes increase. Unlike measurement error, it is not (apparently) random, and therefore, results subject to systematic error alone do not vary about a true value. Systematic error can result from deliberate judgment to exclude (or include) data, parameters, or models that ought not to be excluded (or included).
  • "The only way to deal with systematic error is to recognize a bias in the process, model, or procedure and remove it. Systematic error, however, is notoriously difficult to recognize except on theoretical grounds. Corrections may only be applied when the magnitude and direction of the bias are known. Such corrections underlie the application of double-sampling methods in environmental science." (Regan, et. al., op. cit.)
• Natural variation - Natural variation or underlying process variation occurs in systems that "...change (concerning time, space, or other variables) in ways that are difficult to predict...across the full range of temporal and spatial values (or other related variables)." note-28 [22]
  • This form of uncertainty is reduceable utilizing tools such as statistical process control and replication of experiments or observations.
• Inherent randomness - randomness or aleatory uncertainty exists because the system is"...in principle, irreducible to a deterministic one (the most well-known case is described by Heisenberg’s uncertainty principle in quantum mechanics)." note-29 [23], note-30 [Note 7]
  • Even if the argument is accepted that it is unlikely that any given physical system at some level is inherently random, it is important for any taxonomy to distinguish between systems that appear random due to incomplete or inconsistent information and those that are intrinsically random. (Again, see Regan (2002)).
• Model uncertainty - Model uncertainty occurs in the process of generating a model as a conceptual representation of a particular phenomenon to create a simplified reflection of reality. This selectivity of factors and variables creates uncertainty in at least three ways. (Regan (2002))
  • Variables and processes that are regarded as relevant often represent a trade-off between system knowledge and assumed states and model objectives;
  • Models depict observed processes using logical or mathematical constructs based on underlying theories about system states or dynamics using continuous equations to describe discrete processes.
  • Curve fitting (including interpolation and extrapolation) with mathematical expressions using model variables where inputs are discrete data points.
  • Regan (2002) argues that model uncertainty is "...notoriously difficult to quantify and impossible to eliminate ...(and)...(t)he only reliable way of determining how appropriate a model is for prediction is to perform validation studies."
  • Subjective judgment occurs due to the interpretation of scarce or error-prone data.
    • The only way to address this type of uncertainty is to "...assign a degree of belief about an event in the form of a subjective probability." (Regan (2002))
• Linguistic uncertainty - Linguistic uncertainty, or "vagueness", is a source of uncertainty and includes uncertainties due to context dependence, ambiguity, and under-specificity. note-31 [24]], note-32 [25]]
  • Context dependence is uncertainty concerning the context in which a statement is to be understood;
  • Ambiguity occurs when words have multiple meanings, and further reduction to a single meaning is not logically possible in a given context.
  • Underspecificity occurs in the presence of "unwanted generality" or multiple interpretations, and reduction to a smaller set or even a single interpretation is not logically possible.
• Stochastic or statistical uncertainty - Stochastic or statistical uncertainty, which is sometimes known as Aleatoric uncertainty as well as the subject of stochastic control pertains to the parameters of a risk model, originates from sampling or measurement error, and can be quantified and mathematically expressed (e.g., using confidence intervals).

Top of current page

Uncertainty Taxonomies in a Legal Environment

From a legal perspective, a definition of taxonomy is dictionary-based, such as "the systematic distinguishing, ordering, and naming of type groups within a subject field." given by Walker (op. cit.). Such legal concepts of scientific or engineering taxonomies are not required to be epistemically complete. Still, they must cover "the most significant aspects of ...(current, good discipline practices) ... and of the descriptive information commonly encountered by decision-makers" in a logically distinct manner. (Walker, 1991, p. 571) These taxonomies must be sufficient to catalog the types of uncertainty associated with descriptive assertion, including the critical subset of cause and effect assertions. This legal/decision-maker framework has presented descriptive uncertainty as having six components (Walker, 1991). Although not part of the Walker taxonomy, underlying the scheme is linguistic uncertainty, or "vagueness," as a source of uncertainty, including uncertainties due to context dependence, ambiguity, and under-specificity. note-33 [26]] Arguably, these uncertainties underlie the most basic element in the taxonomy.

• Linguistic - or "vagueness", is a source of uncertainty and includes uncertainties due to context dependence, ambiguity, and under-specificity. (Regan, 2002)
• Conceptual- the definition by choice and design of descriptive concepts or variables to be used as predicates where the subject of an assertion identifies what is being discussed and the predicate provides information about the topic or characterizations such as what the subject is, what the subject is doing, or what the subject is like (Wikipedia).
• "Conceptual uncertainty, or the potential for conceptual error, arises whenever predication occurs. Whenever a concept is used to describe something, using certain concepts instead of others begins to structure how we understand the object, event, or instance under discussion. Predication or conceptualization generates useful information about things, but it also can inhibit our ability to think about those things with concepts other than those selected. The concepts used may not be the most fruitful or the best designed- either for scientific purposes or for making wise, fair, effective, and efficient decisions." (Edited for reading, Walker, 1991)
• Similarly, choice and design of descriptive variables enhance and restrict data set membership, creating potentially incomplete, incompatible, or inconsistent data sets, "...the classification categories employed, and the relationships among those categories (nominal, ordinal or scalar)." note-34 [27]]
• Concept uncertainty note-35 [28] like systematic error discussed above is notoriously difficult to recognize, except on theoretical grounds, and the magnitude and direction of the bias are known.

Both terms could be analogized to that of accuracy in ISO 5725 which looks at the ability of the concept or system's ability to produce results that are proximate to the "true" value or state of the phenomenon. Conceptual uncertainty and systemic bias are related to another measure of uncertainty: validity. Walker has defined validity as the ability of a concept to quantitatively describe and explain what the phenomenon objective "truly" is. Walker also ties the concepts together when he states, "Validity concerns the "accuracy" of the measurement data, not its precision." (Walker, 1998)
Additionally, conceptual uncertainty and associated validity issues could be propagated throughout this taxonomy. Every component of action and decision in the face of uncertainty raises validity issues. Walker recognized this when the author defined epistemic choice for concepts used throughout the other uncertainty components or layers.

• Measurement uncertainty or Misclassification error- the application of the underlying concepts or variables to specific, individual cases and the uncertainty of the reliability (Walker, 1998) of the resultant data. (It is important to remember that assessment observation and measurement are used interchangeably in this analysis.) Thus, an assessment method or procedure is considered "reliable" or "precise" in the scientific terms of ISO 5715 if it repeatedly produces consistent results.
• Measurement uncertainty is logically distinct from conceptual uncertainty, which can only occur after any conceptual uncertainty has been established. Still, misclassification can occur even when the conceptual uncertainty has been minimized. Conversely, underlying errors or errors made in measurement or classification propagate thru the reasoning chain and reduce the quality or value of the final result.
• While uncertainties associated with quantitative measurements or assessments can be statistically evaluated and described, uncertainties related to qualitative assessments are more problematic.
  • Working through the taxonomic chain from base concept to causal explanation requires amassing and integrating qualitative information and quantitative data into input parameters for mathematical models. Developing procedures for producing consistent, repeatable sets of information from assessments that minimize conceptual error or systemic bias requires professional efforts to "validate" such methods, models, and judgments at all layers in the process. Validation as a process is something that civil engineering as a practice needs to embrace more frequently.
• Sampling- in the classical statistics sense, is the selection of a limited subset of a larger population to use as a basis for making estimates about the population in the form of attributes, proprieties, or parameters. The resulting information would be determinative in making forecasts about future population sampling or possessing a desired amount of predictive power. The uncertainty about the ability of this information to correctly predict future samples is termed sampling error.
• Modeling- Modeling uncertainty arises whenever a claim is made that out of a class of candidate relationships and variable pairings, one variable pairing inclusive of constants is chosen that possesses a particular or persistent mathematical relationship to another variable X. Modeling errors arise in selecting the wrong variable pair and constants or by incorrectly specifying its constants. (Walker, 1991, pg. 599)
  • Causal- In causal modeling and analysis, the relationship between causal variables is not strictly a mathematical function. Still, it makes testable predictions and explains "...how a system of variables works, why a system works the way it does, or why it makes sense to think of certain variables as a system" at all." (Walker, 1991, pg. 609)
  • Epistemic- the choice of interpretations for fundamental, logical concepts used throughout the other components or layers.

The components of this uncertainty taxonomy can combine in different ways to "...produce different aggregate uncertainties." note-36 [29]

Only the scientific and engineering professions possess the capacity to develop and validate procedures and models that combine the different kinds of qualitative note-37 [30] and quantitative information and their associated potential for error into an aggregate measure or "scalar variables" of uncertainty.note-38 [31]]

Scalar variables are quantitative variables whose categories are related by some measure of the relevant property's incremental frequency, degree, or amount. (Walker, 1991) Examples are Modulus of Elasticity, Compression stress in elastic materials, etc.

Top of current page

Economics Uncertainty Taxonomies

Like scientific concepts of uncertainty, economics is environment-specific (in this case, read sector of economy or markets), phenomenon-oriented (in this case, read human activity and manufactured phenomena such as firms and markets guided by human judgment and decision), and model-centric.

Economics is not interested in all forms of uncertainty, only specific, meaningful subsets possessing casual connections with material impacts, often labeled as 'risks'.

One of its earlier writers (Knight, 1921, 1948, 1957) attempted to define "uncertainty" as the presence of "defects of managerial knowledge" (or knightian uncertainty and thus risk centric) as risk that is immeasurable, not possible to calculate.(Wikipedia) Risk in this sense was defined as "the ordinary risks of business activity which can ... be reduced... by applying the insurance principle." Knight does acknowledge that risk is a specific and unambiguous subset of uncertainty. (Risk, Uncertainty and Profit, 1957, pg. 19) For Knight, measuring uncertainty meant measuring the probability of its occurrence and the severity of its consequences or economic impact, e.g., measurable risk of loss versus unmeasurable uncertainty consequences, which Knight referred to as "the imperfection of knowledge". (Risk, Uncertainty and Profit, 1957, pg. 197) One of Knight's many contributions to this analysis was recognizing information and knowledge's role in economic activity.

"(W)e are concerned only to emphasize the fact that knowledge is in a sense variable in degree and that the practical problem may relate to the degree of knowledge rather than to its presence or absence in toto. We live only by knowing something about the future, while the problems of life, or conduct at least, arise from knowing so little. This is as true of business as of other spheres of activity. The situation's essence is action according to opinion, of greater or less foundation and value, neither entire ignorance nor complete and perfect information, but partial knowledge. If we are to understand the workings of the economic system, we must examine the meaning and significance of uncertainty; and to this end, some inquiry into the nature and function of knowledge itself is necessary." (Knight, pg. 199) Emphasis added

Knight also reinforced the difficulties in classifying uncertainties when he analyzed life insurance. In this case, accidental death was an uncertain phenomenon compared to sickness and accident, where an "...objective description and classification of cases was impossible..." (Knight, pg. 248) Another factor that made uncertainties ineligible as risks and, therefore, uninsurable because they were unclassifiable was the exercise of judgment in making decisions by the businessman. (Knight, pg. 251) This is a simple classification of economic uncertainty and, as an idealization, is philosophically controversial. note-39 [32]

• Uncertainty and information about the economic environment are distinct from uncertainty about others’ behavior or choices.
• Risk as a specific subset of uncertainty implies that "...all possible acts are known, all possible outcomes arising from each act are known, and it is possible to assign probabilities to each act." note-40 [33]

Like the economic theory of the firm, an economic theory of uncertainty seeks to explain and predict the nature of a specific subset of uncertainty (in this case, risk relating to physical phenomena and human activity), including its behavior, structure, and relationship to the universal uncertainty of physical phenomena. In this sense, uncertainty, like firms and markets, can establish price equilibriums under the right circumstances different from those that might have occurred in either of the other two. Such a theory of uncertainty offers a partitioning scheme that decomposes uncertainty into distinct and meaningful subsets, of which risk is a significant one. It also explains how the presence and variability of knowledge cause or " drive" economic consequences. Unlike physical phenomena, economic activity creates artificial objects such as products, firms, mediums of exchange, markets, economies, and knowledge about those activities. The economics of uncertainty exists as an alternative to "certainty" models that assume away imperfect knowledge and unknown choice/judgment preferences when it is more efficient for economic decision-making. This also allows the introduction of the logical concept of "hazard" versus "risk". Frank used hazard extensively in Risk, Uncertainty, and Profit, but not to the extent that one could say it was interchangeable. An example of this is in the term "Moral hazard," where a firm could engage in "riskier" and potentially "uninsurable" behavior if the information were transparent, and, therefore, insurance coverage of losses is limited. note-41 [34]], note-42 [35]]

Top of current page

Uncertainty and Risk in Policy and Regulatory Environment

"Hume's 'just reasoner', when faced with a difficult public policy decision in current times, would probably commission a risk analysis. But how could they incorporate 'a degree of doubt, caution, and modesty?" note-43 [36]

The problem of "decision-making in the face of uncertainty" is how to regulate based on incomplete information that has the potential to be materially inaccurate. A contextual assumption is that we need to evaluate decision rules for dealing with uncertainty. As a social enterprise, risk regulation, whatever its substantive objectives, should be as effective, efficient, and equitable as practically possible. These "three E's" form a set of "process objectives" or "meta-goals." From the uncertainty standpoint, causal information can be usefully divided into two major categories: information about groups and information about individuals.

Each category, group, and individual has its distinctive types of inherent uncertainty. These are logically distinct, generally independent, and cumulative, contributing to (?????)

Top of current page

Engineering Uncertainty Taxonomies

Like scientific and economic concepts of uncertainty, engineering is environment-specific, phenomena-oriented, model-centric, external choice, and driven by decision-making in the face of uncertainty. Like economics and unlike science, engineering is not interested in all forms of uncertainty, only those with determinable and material consequences. Like economics, engineering conceives risk as a specific and unambiguous subset of uncertainty determined by measuring the probability of its occurrence and the severity of its consequences or economic impact, e.g., measurable risk of loss versus unmeasurable uncertainty consequences. Also, engineering recognizes the role information and, by extension, knowledge plays in constructing the built environment. note-44 [37]] Like economics, engineering differs from economics and scientific uncertainty in its interest in behavioral uncertainty.

• "Behavioral or interaction uncertainty is how individuals or organizations act or interact. Behavioral uncertainty arises from four sources: design uncertainty, requirement uncertainty, volitional uncertainty, and human errors." note-45 [38]]
  • A design uncertainty is a choice among alternatives over which an individual or group of individuals exercises direct control but has not yet decided upon.
  • Requirement uncertainty includes parameters of interest to and determined by the stakeholder, independent of the engineer or designer.
  • Volitional uncertainty is uncertainty about what the subject him/herself will decide. Other people’s future actions and conduct are not entirely predictable, particularly when dealing with other organizations.
  • Human errors occur during the development of a system or project due to blunders or mistakes by an individual or individuals.

Top of current page

Practice Frameworks for Uncertainty and Risk in Civil Engineering Practice

PMIBoK

PMI does not define uncertainty, but the PMI Body of Knowledge uses the term in 45 places (uncertainty) and as describing properties in 8 places (uncertainties). [See Note 8] PMI does define "risk" and uses the term over 1,300 times in the same document. In its glossary, PMI also defines terms such as "threat" and "opportunity" but not events.[391 The BoK contextually defines uncertainty when it states that project risk has its origins in the uncertainty present in all projects. (Op. Cit., pg 309)
Uncertainty is contextually defined and taken as having the property of affecting project execution and the ability to meet stakeholder expectations. Uncertainty is more than the sum of all individual risks (known and unknown). The nature of this uncertainty is not defined, only its capacity to affect something else or its properties.
Risk is defined as " ... an uncertain event or condition that, if it occurs, has a positive or negative effect on one or more project objectives." (Ibid.) Beyond this formal definition, PMI adds context when, in its risk management section (Sec. 11 ), the BoK states that risk is a caused event or condition with multiple causes and impacts. The BoK even offers specific techniques for mapping cause-and-effect relationships and influence diagrams. (Sec. 11.2.2.5) Not as evident but equally important is the recognition that not only is risk caused but the impact is "triggered". The risk trigger is an event or situation that signals that it is about to occur (Glossary, pg. 566). Risk conditions are factors that affect or otherwise contribute to risk, such as project stakeholders. Risk implicitly retains some residual element of probability in it, or a value of less than 1.0 with risk that approaches a level of near certainty is termed "issues" or "realized risk" .(Op. Cit., pg 309). PMI also links risk to underlying variations in project outcomes. (Ibid.) Risk is based upon data and information which can be assessed as to its quality (Sec. 11.3.2.3). This analysis looks to determine the value of the information and examines the degree to which the risk information is understood as well as its " ... accuracy, quality, reliability and integrity ... " (Ibid.) Part of that is understanding the relationship between occurrence and impact as outlined in PMl's Figure 11-10 Risk Impact Matrix and developing a model for prioritizing risk products and setting the threshold for action.

Commentary:

For defining and identifying risk, PMI notes in Sec. 11.2.2.4 that " ... Every project and its plan is conceived and developed based on a set of hypotheses, scenarios, or assumptions." Part of the risk analysis process is identifying risks to the project, such as inaccuracy, instability, inconsistency, or incompleteness of assumptions. Similarly, the quality of management plans, as well as their consistency with others and the project objectives and assumptions are 11 •• .indicators of risk in the project." (Sec. 11.2.2.1)
Similarly, the role of engineering economics in risk is embedded or implied in the PMI definition of risk when the BoK discusses risk identification as a process of assessing which risks out of the total risk exposure for the project "may affect" the project and distilling their characteristics. Embedded in this statement is the concept that not all risks can cause economic impacts (positive and negative) to the project. Also embedded is the requirement to understand risk characteristics to identify project risk (Cost and schedule). Lastly, risk identification is an iterative process of incrementing project risk documentation, much like project scope, cost, and schedule documentation.
Lastly, PMI acknowledges the complex constitution of risk aggregates or the "sum" of all risks in its material, but is it a sum? Risk information has descriptive and explanatory content. These are knowledge and meta-knowledge components. Understanding project assumptions is project-specific knowledge, but identifying risks from assumption inaccuracy, instability, inconsistency, or incompleteness requires professional or program meta-knowledge.
Uncertainty exists in all projects, but risk, a subset of uncertainty with material economic impacts, is a key meta-knowledge element for civil engineering practice.

Software Development Process- Risk Focused

The Unified Process requires the project team to focus on addressing the most critical risks early in the project life cycle. The deliverables of each iteration, especially in the Elaboration phase, must be selected to ensure that the greatest risks are addressed first.

American Society of Civil Engineers (ASCE)

ASCE does not define uncertainty, but the ASCE CE Body of Knowledge (CEBoK) uses the term in 22 places (uncertainty) and describes properties in 13 places (uncertainties).
The CEBoK does not use the phrase "uncertainty and risk" as PMI and AACE do in their publication. ASCE reverses the two (Risk and uncertainty) in twelve places regarding technical outcomes (Outcome 12). A fundamental difference between PMI and ISO is that risk is considered a subset of uncertainty. ASCE does not define "risk" but uses the term 25 times in the same document. Almost all are in the context of variation of design parameters and not in the classical context of cost, schedule uncertainty, and risk.

Commentary:

...

American Association of Cost Engineers (AACE)

AACE defined uncertainty, risk, and other related terms in its Risk Management Dictionary. AACE's overall strategy was to define risk and other terms that stem from base uncertainty. AACE first defined uncertainty as .... "(t)he total range of events that may happen and produce risks (including both threats and opportunities) affecting a project (see opportunities, events, conditions, risk, and threats" where the following sub-definitions apply:

• • Biases--A lack of objectivity based on the individual's position or perspective. Systematic and predictable relationships between a person's opinion or statement and his/her underlying knowledge or circumstances. Note: there may be "system biases" as well as "individual biases."
  • Condition (Uncertain Condition)-Any specific identifiable circumstance (such as the rate of inflation or the quality of labor available) that might affect the outcome of the project
  • Event (Uncertain Condition)-is a specific identifiable action (such as a large government project being started in the same labor area as your project) or an act of nature that might happen and that (if it does happen) could affect the outcome of the project.
  • Opportunities are Uncertain events that could improve the results or improve the probability that the desired outcome will happen
  • Threats are Uncertain events that are potentially negative or reduce the probability that the desired outcome will happen.
    • AACE defines Risk as an "ambiguous term" (sic) that is synonymous with uncertainty or a negative subset such as "threats", or an overall negative impact of all possible uncertainties.

Commentary:

...

Limitations of the definition

...

Working definition

...

Beneficial outcomes

...

See also

Wiki articles on the project.

Top of current page

Notes

1. See McDaniels et al. 2004 for an extensive discussion and bibliography on the historical development of Risk Analysis and some key milestones in the 20th century.
2. Open item
3. The Merriam-Webster dictionary defines uncertainty:

"the quality or state of being uncertain," which is a circular definition. Likewise, one could define uncertainty as the state of not being certain. Synonyms are distrust, doubt, misgiving, mistrust, reservation, skepticism, and suspicion. Another writer added indefinite, indeterminate, not certain to occur, problematical, unreliable, untrustworthy, unknown beyond doubt, dubious, doubtful, not clearly identified or defined, not constant, variable, and fitful to the list.

Han, Paul KJ, William MP Klein, and Neeraj K. Arora. "Varieties of Uncertainty in Health Care: A Conceptual Taxonomy." Medical Decision Making 31.6, (2011 ): 828-838.

Han, ct. al. noted that any definition of uncertainty encompasses " ... numerous types, sources, and manifestations of uncertainty, and ... (any) ... useful working definition of uncertainty needs to specify the concept underlying these varied meanings of the term." (Ibid.)

4. Implicit in this definition of uncertainty as a "state of' is ...

" ... a conceptualization of uncertainty as a subjective, cognitive experience of people--a state of mind rather than a feature of the objective world. Furthermore, the defining feature of this state appears to be a lack of knowledge about some aspect of reality. Importantly, however, the concept of uncertainty also implies a subjective consciousness or awareness of one's lack of knowledge, without which one could not feel uncertain; uncertainty is a form of "meta-cognition" ... ( or alternatively, its main component, meta-knowledge) ... -a knowing about knowing." (Han, 2011, op. cit., Emphasis added)

5. Cox wanted his system to satisfy the following conditions:

Divisibility and comparability- The plausibility of a statement is a real number and depends on the information we have related to the statement.

Common sense - Plausibilities should vary sensibly with the assessment of plausibilities in the model.

Consistency - If the plausibility of a statement can be derived in many ways, all the results must be equal.

Cox's theorem has come to be used as one of the justifications for the use of Bayesian probability theory. (For example, in Jaynes Jayne, Probability Theory: The Logic of Science, Cambridge University Press (2003). -preprint version (1996) at http://omega.albany.edu:8008/JaynesBook.html; Chapters 1 to 3 of published version at http://bayes.wustl.edu/etj/prob/book.pdf

6. The utility of existing taxonomies of uncertainty in civil engineering and, by extension, "risk" has been unsatisfactory for this very reason. This is particularly true in cases where the source of the uncertainty is conflicting versus incomplete information. See Regan, et. al.,"A taxonomy and treatment of uncertainty for ecology and conservation biology" (2002) for a survey of scientific taxonomies for uncertainty.
7. Regan et al. argue that " ... genuine examples of this kind of uncertainty are hard to find. Even classic cases of random experiments like coin tosses and the throwing of dice are deterministic; it is just that we do not have enough information about the dynamic processes and initial conditions to make any sensible estimates about the outcomes. Such processes are, for all intents and purposes, inherently random, but they are not genuinely inherently random. For similar reasons, complex systems such as ecosystems and weather patterns are unlikely to be inherently random. Similarly, chaotic systems are entirely deterministic. They are unpredictable because the deterministic processes generating them and the relevant initial conditions are hard to fully specify (see Stewart 1989, Sugihara et al. 1990)." (Regan (2002)
8. The CEBoK uses the phrase "uncertainty and risk" twice in reference to scheduling (Section 6.5,2) and cost estimating (Section 7.2,2) as well as reversing the two (Risk and Uncertainty) also in two places in discussing project life cycle (both in Section 2.4.1)