Performance Assessment in Science

The call for alternative assessments of science achievement grows out of the current constructivist reform in science curriculum and cognitive research. This article presents and applies guidelines for developing performance assessments aligned with this research and reform. We sample classroom activities or tasks from a domain of activities and construct performance assessments with them. Using this approach, three hands-on science investigations were constructed so that each could be scored by observers in real time. These investigations were considered benchmarks for performance assessments. Because these investigations are costly to develop and administer, surrogates were developed: student notebooks in lieu of observers, computer simulations of the investigations, free response questions paralleling parts of the investigation, and multiple-choice items with alternatives keyed to student hands-on performance. Data have been collected from over 300 fifth- and sixth-grade students using these assessments. We found that hands-on assessments can be developed through an extensive, iterative, development process; hands-on assessments are very delicate instruments. Moreover, they can be scored reliably, even in real time. However, with both benchmarks and surrogates, task heterogeneity -variations in an individual student's performance among tasks-limits the generalizability of performance to the larger domain of interest. Similarly, method heterogeneity-variations in an individual student's performance depending on whether the hands-on investigation, computer simulation or pencil-and-paper exercises was used-limits the exchangeability of the surrogates for the benchmarks.