The potential for continuing and extending research on the effects of generative activities is considerable. What is currently evident from past research is the validity of Wittrock's basic premise of active learner engagement. Further research is necessary to help in selecting the type and mode of activity. In other words, we need to ask when various generative activities are more appropriate than others, and whether they should be used in an instructorprovided format or leamer-generated one. Given past research results and these capabilities, these two broad agendas can be specified.

31.4.1.1. Selection of the Type of Generative Activity. In the previous section, a table of activities was proposed which matches generative activities to desired levels of cognitive processing. This matching must be empirically tested. Questions such as the following take this into account:

1. What are the effects of each generative activity on higher-level learning? Much of the previous research has emphasized fact and concept-level learning and has not dealt with higher-level learning such as application, synthesis, or problem solving.
2. Are there clusters of generative activities that are best used for specific leaming tasks or levels of learning? Are, for example, analogies appropriate for factlevel learning?

31A.1.2. Use of Generative Activities. Previous research has also indicated mixed results from activities requiring overt/covert responses. Because of a "veto power over learning," described by Rothkopf (1976, p. 94), further research should explore the conditions that may require Overt Manifestations to ensure that processing occurs.

3. Is there a differential effect from requiring or not requiring overt manifestations of generative activity?' What are the best strategies (instructional and mechanical) for controlling that information is manipulated in the mind?

31A.1.3. Motivation, Learner, and Knowledge Creation Processes. Another very significant area of research is identifying strategies that will enhance the perception of learner responsibility. This indicates a need to merge the learner control research with that of generative leaming. From Wittrock's writing, it seems apparent that learner control with advisement would be recommended, but it needs to be empirically tested with questions such as:

4. What are the best methods for providing advisory feedback on learner-generated conceptions of the instruction content, and what are their effects?
5. What is the effect on leaming of directive, embedded, or inductive control when motivation level varies? Several strategies have been proposed by various researchers. Directive control, as defined by Rothkopf (1976), takes the form of directions that are given to a learner to perform a particular task. Embedded strategies are similar to Rothkopf's inductive control in that they may not be obvious to the learner. Inductive control does not force a response, however, while an embedded strategy expects the learner to perform the behavior before going on (Rigney, 1980).

31.4.1.4. Instructor Provided or Learner Generated? Some of the research results reported earlier indicate that both developmental and cognitive strengths may play a part in selecting appropriate and successful activities. Besides learner-generated activities in which the learner actively makes connections, Bovy (1981) suggests that instructorprovided activities supplant cognitive connections that are provided for the learner by the instruction itself (instructor generated, not learner generated, but personally relevant). There is also another category of instruction in which no control is provided: offering no suggestions, no forced responses, and no supplanted cognitive strategies. The following table proposes a matching of cognitive strengths with levels of thinking and recommended generative activities. If the activity is one that matches the cognitive strengths of individuals, then perhaps it should be presented in a learner-generated format. If it is an activity that would frustrate the learner-i.e., it is not a cognitive strengththen it should be presented in an instructor-provided format, so that the mental effort can be concentrated on the meaning of the message, rather than on a frustrated attempt at using a technique that does not match one's cognitive style. Providmg no guidance may well be saved for learners with welldeveloped metacognitive abilities (see Table 31-4).

Research designs should then test the effect of these three presentational strategies (learner generated, supplanted, or no control) for each generative-learning strategy matched by cognitive style or other individual difference factors against desired levels of learning or the cognitive processing requirements of the specific task. Cognitive developmental issues should also be considered. The following research questions should yield very important prescriptions:

6. 6. Is there an appropriate use for supplanted vs. generated learning? Does this vary by task or learner?
7. 7. Which activities match with developmental levels of learners?