Stu Moment
University of Illinois Human Factors Division
©2008, Stu Moment
Stu Moment web page

APA reference: Moment, S. L. (2008). Partial Relationships in Human Error Systems, University of Illinois Human Factors Division Proceedings, Retrieved [month] [day],[year], from http://www.humanfactors.illinois.edu/research/HumanElementArticles/PartialRelateInHumErrSys/

Copyright notice: This publication may be used, reproduced, printed and redistributed for personal, academic, research or non-commercial purposes as long as 1) it is not modified, 2) credit is attributed to the Human Factors Division at the University of Illinois at Urbana-Champaign and the author, and 3) the copyright notice and this notice are reproduced on any copies. If you have any questions regarding distribution of this paper contact the author.

Note: This article expands on concepts advanced in the article, Compact Introduction to Human Error, http://www.humanfactors.illinois.edu/research/HumanElementArticles/CompactIntroToHumanError/

Introduction

A human error system gives a macro look at the total environment of a task. It may contain embedded micro analysis at points in the system. Human error systems may show multiple lines of error contribution or may incorporate separate parallel defenses. A Human Environment Design or Diagnostic System (HEDS) is a computationally viable combination of structures and elements [brought to light] by the pioneers of human error philosophy (Moment 2008). This structure accommodates the many different approaches to error analysis published in the 1980’s and 90’s.

Jens Rasmussen categorized error types into knowledge-based, rule-based and skill-based (Rasmussen1982) (Rasmussen 1983). These categories are used religiously today. Donald Norman took error analysis beyond an individual’s performance, professing that analysis should include cognitive, physical and social systems (Norman 1980). David Meister stated that systematic error cause may be in hardware, software, data, procedures, assignment of performer responsibilities, operator training, ability and motivation (Meister 1989). James Reason developed the “Swiss cheese” model of accident causation which shows layers of latent error conditions in various aspects of the task and organization (Reason 1990) (Reason 1997).

HEDS divides tasks into simple, three layer structures. The layers are System Elements, Modifier Elements and Task-Type/Error-Type Elements. Within a system, each individual vertical structure is a “partial.”

Systems, designed for task accomplishment have a non-certain reliability which may be expressed numerically. Modifiers, both physiological and psychological, act on system designs to change the efficiency of a system. Tables 1 and 2 give a sample of system and modifier elements, useful in the basic illustration of concepts. Task-Type/Error-Type elements are specific to the purpose of an individual organization therefore are not specified in this document.

System Elements

Modifier Elements

Error Structures

Examination of a simple error system [brings to mind] simple probability theory. In our models, the term probability equivalent, PE, will be used to allow for analytical developments showing more complex cause to effect math. Figure 1 illustrates a system as a simple single partial.

Partial structures show the relationships between elements acting on a particular system. Most tasks are accomplished with multiple systems. The PE’s are placeholders which are used until values or functions are thoughtfully assigned to elements. Temporary assignment of probabilities helps to visualize the cumulative effects of system elements. The oval “micro” symbol pinpoints locations where cognitive engineering or other task models can be incorporated to improve design or analysis.

Note that nodal PE’s are shown. Branch PE’s which could be conditional, dependent on a specific system-modifier combinations, have a more respected analytical presence but are more difficult to construct until data is collected. The model will allow branch PE’s to be substituted for nodal PE’s.

Figure 1 - Simple Single Modifier in a Partial

We may assume that multiple modifier elements can interact. Robert Wilkinson noted that studies show sleep-deprived people can perform well in short task durations but lose concentration after approximately ½ hour on a task (Wilkinson 1969). Figure 2 diagrams such a relationship.

Figure 2 – Multiple Integrated Modifiers of a Single Partial

Parallel systems are introduced to aid task accomplishment. Some are defenses to alert potential error situations or to redirect a task. Some are an aid to a human practitioner. These two types of parallel systems have different computational effects within the system. The parallel independent monitoring system is shown in figure 3. The parallel processes do not interact, at least by design. Figure 4 shows a parallel system, parallel integrated, where the additional system does react with the primary process system. The computational effects of the interaction may be examined within a framework of micro-theory.

Figure 3 – Parallel Independent Vertical Partials

Figure 4 - Parallel Integrated Vertical Partials

Function Types

The functions which form both individual element values and element interaction computations can take many forms.

Algebraic/Table
Algebraic/Table functions are [customarily] useful where multiple factors form a PE of a single element. Figure 2 shows a relationship where PE =  f(a,b,time). Many such functions will occur in error systems design and analysis where relations between elements are known or perhaps, just hypothesized. Modern programmers can form functions with [plot] formed tables. Table lookup and interpolation of observed data makes the task for the researcher, analyst and the programmer easier. Figure 5 and its subtitle emphasize this method.

Figure 5 – Plotted relationships on with 3 dimensions. Real or assumed data, plotted in this simple form give the modern programmer all the information needed to produce usable functions for your system models.

Toggles
Toggles are [functions] where the existence of some parameter makes another function exist. If a monitoring system requires electricity, and electricity is available, than PE may equal 1. If the electricity is off, the PE definitely equals 0. Modern, technology assisted human performance systems may contain many toggles in vertical or horizontal relationships.

Step-functions
Step-functions are toggles which are activated at a particular threshold value of a related element or combination of elements. Since they are toggles, they produce values of 0 or 1 but their activation is dependent on other relationships. When the quantity of manpower reaches a threshold, a facility or piece of equipment may suddenly fail. The ‘step’ in the step function was reached.

Probability Combinations
Probabilities, if truly applicable to a partial, provide quick computation of system/modifier effects. Probabilities may also apply to some horizontal combinations of elements. If the probability equivalents in figure 1 are accurate, there is a 1 – (.9997 * .995) = .0053 chance of error. It should be noted that each element’s PE is shown in the figures as reliability, which would have been computed as 1 minus the chance of error for that particular element.

The probability of success in parallel independent systems, as illustrated in figure 2, should be the best of each system if the parallel system, when it functions, truly keeps error from occurring.

In repetitive tasks, probabilities show a disheartening phenomena. The probability of success, Ps, repeated n times = (Ps **n). Parallel independent systems greatly improve reliability in repetitive tasks.

Note: Despite the ease of probability calculation, resist the assumption that vertical partials should be calculated with probability math. Functions like the horizontal function f(a,b,time) in figure 2 may be applicable in vertical relationships. PE’s are placeholders and serve to help visualize a vertical partial during development. There is no need to simplify a model to pure probability once the computational order and the [make-up] of the transform is determined.

Accommodation of Micro-analysis

We use the term micro-analysis to refer to models of cognitive control, attention or other models which help to analyze performance of a task at what would be considered to be a local level in the macro system. Inclusion of micro models should lead to the discovery of methods for improved performance in many of the partials.

Cognitive control architectures
A collection of cognitive control models are presented in the book, Adaptive Perspectives on Human-Technology Interaction (Kirlik 2006). The models are illustrated with studies which produce simple, yet useful findings, hence they will contribute to local considerations when tasks are similar to those previously studied.

Cognitive control architectures relate the human’s assessment of the environment to judgments, decisions or actions. Ecological effects are only [crudely] involved in current frameworks of error analysis, yet may be the foremost contributor to error assessment when knowledge-based errors occur.

Attention processes
A multitude of attention frameworks are presented in the book Applied Attention Theory  (Wickens, McCarley 2008). Models of attention control, visual aspecs of attention, resource choice and time sharing may be integrated into many nodes or branches of partials.

Computation Order

The order of computation in complex relationships like those shown in figures 2 & 4 will normally be from the inside out. The combination of sleep deprivation and lack of currency would be calculated before the vertical relationship. Likewise, the interaction between process A and the supportive system of knowledge resources, illustrated in figure 4, should be computed first.

Given an understanding of computational order, micro-analysis involving the achievement of specific subtasks in a system may be easily integrated into the HEDS.

Limitations and Summary.

The assignment of PE’s at the level of system and modifier elements do not account for interactions between elements and, therefore, may appear to be too simple for some theorists. But the structure of partials gives a foundation on which to build interactive relationships. The HEDS structure can accommodate many different views of human performance analysis. Perhaps its main advantage is that it is of dual use. The same structure is used for both analysis and system design.

References:

Kirlik, A. (2006), Adaptive perspectives on human-technology interaction: Methods and models for cognitive engineering and human-computer interaction. New York: Oxford University Press.

Meister, D. (1989), The nature of human error, IEEE Global Telecommunications Conference, Communications Technology for the 1990s and Beyond, volume 2, 783-786

Moment, S. L. (2008), A Compact Introduction to Human Error, University of Illinois Human Factors Division, Retrieved 11/25/2008, from http://www.humanfactors.illinois.edu/research/HumanElementArticles/CompactIntroToHumanError/

Norman, D.A. (1980), Errors in Human Performance, University of California, San Diego, Center for Human Information Processing Report No. 8004

RASMUSSEN, J. (1982), Human Errors. A Taxonomy for Describing Human Malfunction in Industrial Installations, Journal of Occupational Accidents, 4: 311-333

RASMUSSEN, J. (1983), Skill, Rules, and Knowledge; Signals, Signs, and Symbols, and Other Distinctions in Human Performance Models, IEEE Transactions on Systems Man and Cybernetics, Vol.13(3)257-266

Reason, J. (1990). Human error. New York: Cambridge University Press.

Reason, J. (1997). Managing the Risks of Organizational Accidents. Aldershot: Ashgate Publishing Limited

Wickens, C. D., McCarley, J. S. (2008), Applied Attention Theory, Boca Raton: CRC Press.

Wilkinson, R. T. 1969. Some factors influencing the effect of environmental stressors upon performance. Psychological Bulletin. 72: 260-272.