Macro-level framework

From LearnLab
Revision as of 12:25, 7 April 2007 by Vanlehn (Talk | contribs) (Draft version of a framework for the Macro Level)

Jump to: navigation, search

Draft version of a multi-dimensional framework for the Macro Level

The overarching research goal of the PSLC is to understand robust learning, and in particular, to delineate both the conditions under which robust learning occurs and the mechanisms that underlie it. Roughly aligned with these questions are our two major level of explanation. The macro level, which focuses mostly on identifying the conditions where robust learning occurs. It is based on observable conditions, activities and results. The micro level, which focuses mostly on identifying the mechanisms that underlie robust learning. It is based on inference of unobservable conditions, activities and results.

Just as medicine finds indispensable both the clinical and biological levels of explanation, the PSLC has found both its levels to be indispensable as well. The macro level is like the clinical level—it is a relatively atheoretical classification of treatments and effects. The micro level is like the biological level—it explains why certain treatments have certain effects.

Although many classification schemes may make sense for the macro level (PSLC’s cluster-subcluster hierarchy is one), this page presents a multidimensional one. That is, this classification scheme defines a set of design dimension, where each dimension has a set of alternative, non-numeric values. A point in this multidimensional space is simply a specification of a value for each dimension. Each point in the space corresponds to a generic type of instruction. Click here for a list of the dimensions currently being investigated.

Whenever possible, the values of a dimension are ordered or partially ordered by the amount of assistance they offer students, where higher assistance values raise performance during training. For instance, along the feedback timing dimension, offering immediate feedback increases performance during training (measured by uncorrected errors and time-to-completion) compared to delayed feedback, so immediate feedback is a higher assistance value for the feedback-timing dimension than delayed feedback. This ordering allows testing the PSLC’s Assistance Hypothesis:

  • Robust learning will be enhanced by providing assistance is inverse proportion to how well a student knows a component of knowledge.

Experiments are used to map out the multidimensional design space. A typical experiment compares two points (instructional treatments) that have different values on one dimesion and the same values along all others. If the assistance ordering of the compared values is not yet known, it is measured by instrumenting the training appropriately. Prior knowledge is measured by pretesting or its LearnLab equivalent. Immediate and robust learning is assessed with the usual LearnLab measures. Such an experiment determines how two indepentent variables (prior knowledge and instructional assistance) affect two dependent variables (immediate and robust learning). Such an experiment provides a test of a specific application of the Assistance Hypothesis. With hundreds of such experiments, we should be able to understand a design space with thousands or millions of points.

Note that such an understanding would be only a macro level one (like the clinical level in medicine). A micro level understanding (like the biological level) is related but different. An experiment may need to collect finer grained data (e.g., learning curves for individual knowledge components; verbal protocols) in order to test different micro-level explanations of its macro-level results.

Click here for a list of the dimensions currently being investigated.