Vessey and Galletta (1991) define cognitive fit as the match between task, problem representation (e.g. mode of presentation of data) and individual problem solving skills. Goodhue (1995) defines TTF as the ‘extent that technology functionality matches task requirements and individual abilities’. Three components appear consistent across these two definitions: task, technology (which in Vessey and Galletta’s work is what provides the problem representation) and individual abilities.
Operationally, these two widely used fit constructs are quite different. For example, the survey instrument for measuring TTF identifies 12 components (see, for example, Goodhue, 1998), whereas cognitive fit is not measured per se, but rather manipulated in experimental studies that employ the construct. While it is possible to make some mappings between the two constructs, such mappings are not the same as an integrated theory of fit. For example, one of the 12 dimensions of TTF is ‘presentation’, which is operationalised with items like ‘data is presented in a readable and useful format’ (Goodhue, 1998). This dimension may loosely capture the relationship between problem presentation and task that is at the crux of the graphs versus tables debate that cognitive fit has attempted to resolve (Vessey and Galletta, 1991). While the mapping is possible, it is still unclear theoretically what precisely constitutes a comprehensive definition of fit.
Examining fit relative to other behavioural constructs in information systems requires mapping again from scratch. There is no unifying theoretical framework to provide guidance. Consider the well-known constructs of perceived usefulness and ease of use (e.g. Davis, 1989). From a fit perspective, perceived usefulness could map into how well the technology supports the task requirements, and ease of use could correspond to the ‘fit’ between user abilities and the technology. While these mappings seem intuitive, there is no clear theory of fit to justify them and empirical investigation of the relationships has not been forthcoming.
At the outset, fit has three key components — task, technology, and individual characteristics. A unified theory of fit must therefore address these components. For expositional ease, the components are described separately below, although they are inextricably linked in practice.
Following Wand and Weber (1990), a two-part view of information systems is employed here: technology-as-tool and technology-as-representation. Technology-as-tool provides the physical interface for manipulating the technology as representation. ‘Representation’ implies a model of the real world task (e.g. a mathematical model embedded in a decision support system, or a graphical representation of a document) as opposed to a designer’s system metaphor or a mental model inside the head of the user. In decision support systems terms, it is the Representation part of the ROMC (Representations, Operations, Memory Aids and Controls) design approach (Sprague and Carlson, 1982).
Distinguishing between technology as tool and as representation is useful. It can help organise various literatures addressing behaviour with information technology. For example, research on the psychology of decision models (e.g. Cooksey, 1996; Hoch and Schkade, 1996; Melone et. al., 1995; Blattberg and Hoch, 1990; Kleinmuntz, 1985; Einhorn, 1972) bears on technology as representation, whereas work in human-computer interaction (e.g. Davern, 1997; Gerlach and Kuo, 1991; Norman, 1986; Card et. al., 1983) bears on understanding technology as tool.
When considering fit with technology, the question thus arises as to whether it is fit with the tool, the representation, or indeed the fit between the tool and the representation. The tool versus representation dichotomy thus provides at least one basis for conceptualising different types of fit to populate a taxonomy. Any definition of fit must be able to capture both roles of technology: tool and representation.
More broadly, the value of this tool-representation dichotomy is evident in considering other behavioural constructs in the information systems literature. Consider the well-known constructs of ease-of-use and perceived usefulness from the technology acceptance model (Davis, 1989). Does the ease-of-use pertain to the technology as tool or as representation? Likewise for the usefulness construct. By capturing the dichotomy explicitly in a theory and taxonomy for the fit construct there is no such confusion.
Following Wood (1986), task is defined here as comprising three components: products, required acts, and information cues. Products are ends, required acts and information cues are means for achieving the ends or goals. Specification of products or goals should detail the level of performance (in other words, the quality of the product, such as the accuracy of a sales forecast). Behavioural requirements (acts to be carried out and information cues to be used) will vary with the level of performance required in the task product. In practice there may often be substantial choice amongst sets of required acts (more than one way to achieve the goals of the task), even given a specific target level of performance.
The decomposition of task into goals and acts can be carried even further. User interactions with computer-based systems have often been described in terms of a hierarchy of tasks (Gerlach and Kuo, 1991; Card et. al., 1983). Rasmussen (1988) identified three levels of abstraction for computer supported work tasks: functional purpose, generalised, and physical (for an application of this framework in information systems, see Davern, et. al., 2000). In a similar vein, Silver (1990) distinguished between mechanical and conceptual tasks in decision support system usage.
To illustrate the task hierarchy consider the example of producing a sales forecast. There is the substantive task for which the product is the sales forecast. The required acts and information cues for making the forecast could involve collating and modelling a sizable time series of past sales. The size of the time series (the information cues) and the collation and modelling efforts (required acts) are contingent on the level of performance at which the product (sales forecast) is defined (i.e. the desired accuracy). Typically, the task leads the human forecaster to use some computer-based aid. Using technology support to produce a forecast involves running the time series data against some forecasting model. Thus there is the task that has as its product a computer model of sales (a sub-task of the substantive task), which involves the technology as representation. Subordinate to this task are the series of sub-tasks the products of which are specifications of some part of the model inside the computer or refinements to it (e.g. adding variables, etc). There is also a series of subtasks, the product of which is the time series data in a form suitable for running against the model. These more operational tasks involve the use of technology as a tool rather than as a representation, and are a primary concern of human-computer interaction research.
The type of fit and relevant measures of performance vary with the level at which a task is defined. At a minimum, it is necessary to distinguish among three key levels in the task hierarchy. There is the level of the super-ordinate or substantive task that motivates the whole exercise (e.g. producing a sales forecast). There is also the level of the computer modelling task [1] which involves the technology as representation (e.g. conceptual development and manipulation of a sales forecasting model inside the computer). Finally, there is the level of the more operational tasks in the hierarchy that involve the use of technology as a tool (e.g. data entry and other more physical operations). It is also evident that different users will work on different and multiple tasks and sub-tasks at different stages of interaction with an information system. By way of example, Figure 1 depicts a sample task hierarchy for a production mix optimisation problem using the spreadsheet package Excel.
Newell (1982) provides a framework by which it is possible to understand individual behaviour in terms of knowledge and goals — the principle of rationality — which states:
If an agent has knowledge that one of its actions will lead to one of its goals, then the agent will select that action (Newell 1982, p. 102).
In simple terms, if an individual’s goals are known then from observing their behaviour his knowledge can be inferred. Alternatively, if the goals and knowledge of an individual are known his behaviour can be predicted.[2] Newell’s framework complements the definition of task in terms of products, required acts and information cues presented above. Products are goals. Information cues and required acts define the structure of the task to which the individual user must apply their knowledge and abilities. Thus, such questions arise as: Does the user have the knowledge to carry out the required acts? Does he or she have the requisite knowledge to effectively utilise relevant information cues? Does the user’s knowledge and abilities about how to achieve the task goal/product correspond to the set of acts and information cues supported by the technology?
To understand behaviour (and ultimately performance) with information technology thus requires understanding both the goals and the knowledge of the user. Importantly, this does not require absolute definition of how an individual will achieve a goal. The Principle of Equifinality (McClamrock, 1995) suggests that there are multiple ways to achieve a given goal or product. Colloquially, this is often rendered as ‘there is more than one way to skin a cat’. Equifinality and the task hierarchy suggest that multiple actions may achieve a goal and there may be multiple goals implying different actions. Thus, the principle of rationality simply constrains actions, rather than dictates them. This degree of flexibility and equifinality evident in the application of the principle of rationality will prove important in understanding fit as a dynamic emergent property of user-technology interactions, as discussed in section four below.
Any definition of fit must consider task (goals/products, required acts, and information cues), technology (tool and representation, both of which support finite sets of goals, acts and relevant information cues), and user (goals, knowledge of how to carry out relevant acts and knowledge of how to make use of relevant information cues). It must also be consistent with the principle of equifinality. In its most general form fit is defined here as:
An agent’s potential to achieve a given performance level at a given task.
Notably, the definition is silent on the mechanics of how potential becomes performance, which is consistent with the principle of equifinality. The distinction between fit and performance is somewhat analogous to the distinction in physics between the potential energy of a ball about to be dropped and the kinetic energy of the ball in motion, having just been dropped. More formally, it draws on Chomsky’s (1965) classic distinction between competence and performance theories of behaviour that is the foundation of much of modern linguistics. Competence, as used by Chomsky, defines the capability for idealised performance. Performance is the realisation of this capability, which in practice may not quite achieve the full potential.
The general definition above is purposely couched in terms of an ‘agent’. Changing the agent in question provides a basis for generating different types of fit. For example, the agent may be the individual user, the technology-as-tool or the technology-as-representation. The roots of this definition are in the broader cognitive science literature, where Johnson et al. (1992) provide a powerful definition of the fit a cognitive agent has with their task environment:
Fit, then, characterises the degree to which an agent’s expertise (a) reflects the requirements for success in … performing tasks and (b) is in accordance with the structure of available task information. (p. 307)
Adapting Johnson et al’s definition to the present context, fit is formally defined here as:
Fit: The degree of functional correspondence between an agent’s knowledge and the structure and features of the environment that specify supported actions and available information, relative to a specific task.
With its roots in psychology (e.g. Gibson, 1979; Kochevar, 1994) this definition lends itself to a theory of dynamics as discussed below. Importantly, it can also be readily mapped on to the different components of fit described earlier.