Many of us may assume that the most recent studies of feedback are the result of several current trends and accept ed paradigms-for example, the information-processing model and newer theories of motivation. However, three definitions of feedback dating back to the early 1900s are surprisingly similar to the ones we use today. Kulhavy and Wager (1993) refer to these as the "feedback triad" (p. 5) and point out that these definitions still prevail in our views of feedback we currently hold. The first view was that feedback served as a motivator or incentive for increasing response rate and/or accuracy. Secondly, feedback acted to provide a reinforcing message that would automatically connect responses to prior stimuli-the focus being on correct responses. Lastly, feedback provided information that learners could use to validate or change a previous response-the focus falling on error responses..

The earliest studies of feedback date back to E. L. Thorridike's Law of Effect that postulated that feedback would act as a "connector" between responses and preceding stimuli (see 2.2.1.3; Kulhavy & Wager, 1993). Researchers such as Thomdike were examining the use of postresponse information as early as 1911 (cited in Kulhavy & Wager, 1993). Thorndike's work showed that a response followed by a "satisfying state of affairs" is likely to be repeated and increases the likelihood of learning. The view of feedback as information emphasized the role that the learner had in learning, with the ability to adapt his or her response according to information in the feedback and thus correct his or her errors. The first researcher to emphasize error correction was Sidney Pressey (1926). However, a later study using his "teaching machine" (see 2.3) emphasized both the error-correcting function of feedback as well as its acting as a punishment for errors-a Thorndike viewpoint that supports the notion of feedback as a reinforcer (Pressey, 1927). Thus we see that the confusion in the feedback research began quite early and that, given the early "feedback triad," the research has not evolved as much as one might expect.

Thorndike's pioneering work paved the way for the next avenue of research on feedback, B. F. Skinner's study of programmed instruction (Skinner, 1958; see 2.3). Using principles from the Law of Effect and the application of reinforcement on learners, Skinner proposed that a solution to instructional problems lay in the use of strategically designed classroom materials that would take learners through information in a step-by-step fashion, shaping behavior and strengthening desired responses. By the year 1960, the programmed instruction movement was well under way, purporting that feedback in programmed instruction served as both a reinforcer and a motivator and perpetuating a confusion between learning and incentive.

During this period, instructional errors were either ignored or considered as "aversive consequences" to be avoided (Skinner, 1968). The fact that errors were deemed as aversive implies an emotional element from which the early motivational view of feedback was derived. The viewpoint that incorrect responses cause distress and influence self-concept is used even today (Fischer & Mandl, 1988). Kulhavy and Wager (1993) suggest that such motivational variables should be separated from the feedback message, keeping them extrinsic to the lesson content itself. Certainly this would help remove the confusion between the instructional content of feedback and other factors that might affect performance.

Programmed instruction (see 2.3.4, 22.4.1) emphasized an operant approach to learning (see 2.2.1.3.2), one that had the concept of reinforcement at its heart. Programs were designed to shape a student's responses, using a small lockstep approach with a high level of redundancy, Operant psychologists of the time argued that learning tasks should be analyzed and broken down into small enough steps such that the probability of a successful response was ensured (Cohen, 1985). By telling a student that an answer is correct, the student is "reinforced" to answer correctly again on a later test (Kulhavy, 1977).

Around 1970, most researchers began to doubt the feedback-as-reinforcement view. In fact, 10 years of research under this paradigm showed no systematic effects for feedback (see Kulhavy & Wager, 1993). Studies provided little evidence that feedback following positive responses acts in a reinforcing manner (Anderson, Kulhavy & Andre, 1972; Bardwell, 1981; Barringer & Gholson, 1979; Kulhavy, 1977; Roper, 1977). Researchers then had to look at the basic functions of feedback to discover what was actually occurring. A series of studies by R. C. Anderson and his colleagues found that students will not use feedback as the researcher intends unless this use is controlled (Anderson et al., 1971, 1972). For instance, students will simply copy answers from feedback if allowed to do so, with little or no processing or learning of information. Kulhavy (1977) coined the term presearch availability to describe the ease in which learners can find a correct answer without reading the lesson material. If presearch availability is high, then students will usually copy the answer itself, bypassing the instruction and yielding little learning (Anderson & Faust, 1967). In programmed material, feedback significantly facilitates learning only if students must respond before seeing the feedback.

The data collected by Anderson and his colleagues (Anderson et al., 1971, 1972) not only provided insight into the importance of the learner's processing of the lesson material before his or her response to a question but also, perhaps more importantly, provided indication that feedback functions primarily to correct errors, not merely to "reinforce" correct answers. Numerous studies during this time supported feedback's ability to correct inaccurate information (Anderson et al., 1971, 1972; Bardwell, 1981; Barringer & Gholson, 1979; Kulhavy, 1977; Kulhavy & Anderson, 1972; Roper, 1977; Tait, Hartley & Anderson, 1973). Concurrently, cognitive psychology was coming into vogue (see 5.2), and many educational psychologists were shifting from a behavioral to a cognitive view of feedback. Such researchers became more interested in how feedback influenced primary cognitive and metacognitive processes within a learner (Briggs & Hamilton, 1964; Kulhavy, 1977). Consequently, feedback was said to serve primarily as information and not as reinforcement.

Examining feedback from an information-processing perspective (see 5.4), the learner participates in the system to correct his or her errors. Kulhavy and Stock (1989) use the concept of servocontrol theory, contrasting the two feedback systems (feedback-as-reinforcement vs. feedback-as-information) as either open loop or closed loop. Feedback acting as reinforcement would be an example of an open-loop system, in which errors are ignored because the system is not affected by input information. The operant approach does not provide effor-correcting mechanisms. In contrast, the feedback-as-information position acts as a closed-loop system. Since this type of system has ways of correcting errors, errors are of primary importance. In light of this view, studies indeed emerged which made the correction and analysis of errors a major goal (Anderson et al., 1971; Birenbaum & Tatsuoka, 1987; Elley, 1966; Gilman, 1969; Kulhavy & Parsons, 1972).

It is from the information-processing perspective that most research of the past 20 years has been conducted. In a later portion of this chapter, the prevailing concerns of researchers during this period to the present will be discussed in detail. But first, it is helpful to present two current models of feedback as a framework for what follows.