22.5 Aptitudes, on-task performance, and response sensitive adaptation

As reviewed above, micro-adaptive systems, including ITS, demonstrate the power of on-task measures in adapting instruction to students' learning needs that are individually different and constantly changing, while ATI research has shown few consistent findings. Because of the theoretical implications, however, efforts for selectively applying aptitude variables in adaptive instruction is continuing. It has been suggested to integrate some aptitude variables in the micro-adaptive system. For example, Park and Seidel (1989) recommended to include several aptitude variables in the ITS student model and use them in the diagnostic and tutoring processes.

22.5.1 A Two-level Model of Adaptive Instruction

TO integrate the ATI approach in a micro-adaptive model, Tennyson and Christensen (1985, 1988; also see Tennyson & Park, 1987) have proposed a two-level model of adaptive instruction. This two-level model is partially based on the findings of their own research in adaptive instruction over the last 2 decades. First, this computer-based model allows the computer tutor to establish conditions of instruction based on learner aptitude variables (cognitive, affective, and memory structure) and context (information) structure. Second, the computer tutor provides moment-to-moment adjustment of instructional conditions by adapting the amount of information, example formats, display time, sequence of instruction, instructional advisement, and embedded refreshment and remediation. The microlevel of adaptation takes place based on the student's on-task performance, and the procedure is response-sensitive (Park & Tennyson, 1980). The amount of information to be presented and the time to display the information on the computer screen are determined through the continuous decision-making process of the Bayesian adaptive model based on on-task performance data. The selection and presentation of other instructional strategies (sequence of examples, advisement, and embedded refreshment and remediation) are determined based on the evaluation of the on-task performance. However, the response-sensitive procedure used in this micro-adaptation level has two major limitations, as discussed in the Bayesian adaptive instructional model: (a) problems associated with the quantification process in transforming the learning needs into the Bayesian probabilities, and (b) the capability of handling only limited types of learning tasks (e.g., concept and rule learning).

For variables to be considered in the macro-adaptive process, Tennyson and Christensen (1988) identified the types of learning objectives, instructional variables, and the enhancement strategies for different types of memory structure (i.e., declarative knowledge, conceptual knowledge, and procedural knowledge), and cognitive processes (storage and retrieval). However, the procedure for integrating components of learning and instruction are not clearly demonstrated in their Minnesota Adaptive Instructional System.

22.5.2 On-Task Performance and Response-Sensitive Strategies

Studies reviewed in the micro-adaptive models demonstrated the superior diagnostic power of on-task performance measure to pretask measures and the stronger effect of response-sensitive adaptation to ATI or nonadaptive instruction. These results indicate the relative importance of the response-sensitive strategy compared to ATI methods. The student's on-task performance or response to a given problem is the reflection of the integrated effect of all the variables, identifiable or unidentifiable, involved in the student's learning and response-generation process. As discussed earlier, a shortcoming of the ATI method is adapting instructional processes to one or two selected aptitude variables in spite of the fact that learning results from the integrated effects of many identifiable or unidentifiable aptitude variables and their interactions with the complex learning requirements of the given task. Some of the aptitude variables involved in the learning process could be stable in nature, while others could be temporal. Identifying all of the aptitude variables and their interactions with the task-learning requirements is practically impossible.

Research evidence shows that some aptitude variables (e.g., prior knowledge, interest, intellectual ability) (Tobias, 1994; Whitener, 1989) are important predictors in selecting instructional treatments for individual students. However, some studies (Park & Tennyson, 1980, 1988) suggest that the predictive value of aptitude variables decreases as the learning process continues, because the involvement of other aptitude variables and their interactions may increase as learning occurs. For example, knowledge the student has learned in the immediately preceding unit becomes the most important factor in learning the next unit, and motivational level for learning the next unit may not be the same as in learning the last unit. Thus, the general intellectual ability measured prior to instruction may not be as important in predicting the student's performance and learning requirements for the later stage or unit of the instruction as it was for the initial stage or unit.

In a summary of factor analytic studies of human abilities for learning, Fleishman and Bartlett (1969) provided evidence that the particular combinations of abilities contributing to performance change as the individual works on the task. Dunham, Guilford, and Hoepner (1968) also found that definite trends in ability factor loading can be seen as a function of stage of practice on the task According to Fredrickson (1969), changes in the factorial composition of a task might be a function of the student's employing cognitive strategies early in the learning task and changing the strategies later in the task. Because the behavior of the learner changes during the course of learning, including the learner's strategies, abilities that transfer and produce effects at one stage of learning may differ from those effective at other stages.

22.5.3 Diagnostic Power of Aptitudes and On-Task Performance

As discussed above, the change of aptitudes during the learning process suggests that the diagnostic power of premeasured aptitude variables for assessing his or her learning needs, including instructional treatments, decreases as the learning continues. In contrast, the diagnostic power of on- task performance increases because it reflects the most updated and integrated reflection of aptitude and other variables involved in the learning. In contrast, the student's on-task performance in the initial stage of learning may not be as powerful as in the later stage of learning because of the student's lack of understanding about the nature of the task, specific learning requirements in the task, and his or her ability related to the learning of the task. Therefore, during the initial stage of instruction, specific aptitude variables like prior knowledge and general intellectual ability may be more valuable than on-task performance or response in prescribing the best instructional treatment for the student.

The decrease in predictive power of the premeasured aptitude variables and. the increase in that of on-task performance can be represented as Figure 22- 1.

22.5.4 Response-Sensitive Adaptation

Figure 22-1 suggests that an adaptive instructional system should be a two-stage approach: (a) adaptation to the selected aptitude variable, and (b) response-sensitive adaptation. In the two-stage approach, the student will initially be assigned to the best instructional alternative for the aptitude measured prior to instruction, and then response-sensitive procedures will be applied as the student's response patterns emerge to reflect his or her knowledge or skills on the given task. A representative example of this two-stage approach is the Bayesian adaptive instructional model. In this model, the student's initial learning needs are estimated from the student's performance on a pretest, and the estimate is continuously adjusted by reflecting the student's on-task performance (i.e., correct or incorrect response to the given question). As the process for estimating student learning needs continues in this Bayesian model, the value of the pretest performance data becomes less important, and the most recent performance data become more important.

The response-sensitive procedure is particularly important because it can determine and ad use learning prescriptions with timeliness and accuracy during instruction. The focus of a response-sensitive approach is that the instruction should attempt to identify the psychological cause of the student's response and thereby lower the probability that similar mistakes will occur again rather than merely correcting each mistake. The effect of a response-sensitive approach (e.g., Atkinson, 1968; Park & Tennyson, 1980, 1986) has been empirically supported. Also, some of the successful ITSs (e.g., SHERLOCK) diagnose the student learning needs and generate instructional treatments entirely based on a student's response to the given specific problem without an extensive student-modeling function.

Development of a response-sensitive system requires procedures for obtaining instant assessment of student knowledge or abilities and alternative methods for using those assessments to make instructional decisions. Also, the learning requirements of the given task, including the structural characteristics and difficulty level, should be assessed continuously with an on-task analysis. Without considering the content structure, the student's response reflecting his or her knowledge state about the task cannot be appropriately analyzed, and a reasonable instructional treatment cannot be prescribed. The importance of the content structure of the learning task was well illustrated by Scandura's (1973, 1977a, 1977b) Structural Analysis and Landa's (1970, 1976) Algo-Heuristics approaches.

To implement a response-sensitive strategy in determining the presentation sequence of examples in concept learning, Tennyson and Park (1980) recommended analyzing on-task error patterns from the student's response history and content and structural characteristics of the task. Many ITSs have incorporated functions to make inferences about the cause of a student misconception from the analysis of the student's response errors and the content structure and to instantly generate instructional treatment (i.e., knowledge) appropriate for the misconception (see Chapter 19).

22.5.5 On-task Performance and Adaptive Learner-Control

A similar curve to the instructional diagnostic power of aptitudes (Mg. 22-1) can be applied in predicting the effect of the learner-control approach. In the beginning stage of learning, the student's familiarity with the subject knowledge and its learning requirements would be relatively low, and the student would not be able to choose the best strategies for learning. However, as the process of instruction and learning continues and external or self-assessment about the student's own ability is repeated, her or his familiarity with the subject and ability to learn it would increase. Thus, as the instruction progresses, the student would be able to make better decisions in selecting strategies for learning the subject. This argument is supported by research evidence that the strong effect of learner-control strategies are mostly found in relatively long-term studies (Seidel, Wagner, Rosenblatt, Hillelsohn & Stelzer, 1978; Snow, 1980), while scattered effects are found usually in short-term experiments (Carrier, 1984; Ross & Rakow, 198 1).

The speed, degree, and quality of obtaining the self-regulatory ability in the learning process, however, would be different between students (Gallangher, 1994), because learning is an idiosyncratic process influenced by many identifiable and unidentifiable individual difference variables. Thus, an on-task adaptive learner control, which gradually gives the learner the options for controlling the instructional process based on the progress of the learner's on-task performance, should be better than non- or predetermined adaptive learner control, which gives the options without considering individual differences or is based on aptitudes measured prior to instruction. An on-task adaptive learner control will decide not only when is the best time for giving the learner-control option but also what kind of control options (e.g., selection of contents, learning activities, etc.) should be given based on the student's on-task performance. When the learner-control options are given adaptively, the concern that learner control may guide the student to put in less effort (Clark, 1984) would not be a serious matter.