Since the introduction of the Card, Moran, & Newell book in 1983, The Psychology of Human-Computer Interaction, the GOMS model has been a leading theory in cognitive modeling. It is one of the few widely known theoretical concepts in HCI. The GOMS concept inspired much research in the area of cognitive modeling. This fostered efforts to create more plausible and useful extensions. GOMS models were actually devised to improve the processes of cognitive modeling and usability engineering. Their purpose was to extend psychologically-oriented cognitive models by adding the following: the ability to make a priori predictions, the ability to be learned and used by computer system designers as well as researchers, coverage of relevant tasks, and approximation.
What are the differences between GOMS task analysis techniques?
For what types of design is GOMS beneficial?
In 1990 Olson and Olson extended the basic GOMS framework. They found several significant gaps in cognitive theory that prevent cognitive modeling in general from addressing some important aspects of HCI(e.g., fatigue) and argued that cognitive models are essentially the wrong granularity or form to address certain other aspects of computer systems design, such as user acceptance and fit to organizational life.
GOMS is a procedural cookbook method of task analysis that can be used for well defined routine cognitive tasks. It is restrictive and time consuming to write a sequence-form model by hand for a large set of tasks.
Restrictions
GOMS strictly models human performance on specified tasks. These methods lend themselves well to usability analysis, training programs and help systems. For designing leading edge technology systems, GOMS has deficiencies. In these areas, there may be no expert user, and we may not know why or how the system will be used. The learning and performance on these systems could be new, unpredictable motor operations.
Our final day of presentation was spent in a critical discussion of some of the major points of contention that we had with GOMS. Some of the discourse is summarized in the following passage.
Q1. Does CMN-GOMS and KLM GOMS handle mental operations in an appropriate fashion?
A1. KLM has no explicit goals or choices between goals, whereas the CMN-GOMS model represents these explicitly.
Q2. Is there any empirical difference between the placement of the M operators in KLM vs. CMN-GOMS?
A2. KLM model puts M operator at the beginning of subprocedures, while CMN-GOMS model puts the mental time in verify operations at the end of subprocedures. This difference is irrelevant in most design situations.
Q3. How does learning time apply in the NGOMSL model?
A3. This is the only model to provide information about learning time and these predictions cover only the time to learn the methods in the GOMS model and any long term memory information they require.
Q4. In relation to selection rules, do people truly take time to think about how to perform a task in an IF-THEN-ELSE fashion? Should this be selection rules or selection heuristics?
A4. Dr. Hartson suggested performing a study by training a group of people how to do the same task by applying several different methods. Once they are all similarly trained then proceed with some task to require them to use the task they were just taught and observe(without their knowledge which method each uses. Afterward question them as to how they made their decision on which method to use
Q5. What are the implications and short-comings of GOMS by the fact that it assumes error-free performance?
A5. The GOMS Model by Mark Treglown -- Studies have found that even expert users who are skilled typists spend up to 25% of their time using a computer system either making errors or recovering from errors. If the model is to represent error-handling behavior, then models need to be complicated by adding extra methods.
Q6. Since FITTS models have been found to accurately measure repetitive movements, does anyone see an application of FITTS in GOMS?
A6. GOMS utilizes FITTS timings.
Q7. What are the benefits of using the GOMS task analysis technique over simple usability studies?
A7. The assumption made by GOMS that actions are independent and additive render the model incapable of describing the coarticulation of concurrent activities. John extended GOMS for task involving parallel activities by incorporating the concept of critical path.
Q8. What needs to be done for GOMS models to become a standard design methodology?
A8. As a rule of thumb when choosing a design technique, tools that are fast, cost effective and easy to use are a must. Methods such as Quick and Dirty GOMS(QGOMS) have been adopted for those reasons. GOMS needs to handle standard human factors issues such as readability of letters and words on the screen, visual quality of a display layout, recognizability of menu terms or icons, and memoriability of commands. It also needs to address the quality of the work environment, user acceptance, and affect(i.e. system is fun to use or does it cause fatigue, boredom). Also the social or organizational impact of the system and the resulting influence on productivity.
The critical discussion gave the class a chance to clear up some misconceptions about GOMS from previous information. It also raised some questions about the timings used in GOMS. Problems with "errorless performance" and is GOMS just practical for academic applications.