The United States invented a matrix sampling technique to impute five plausible scores of student performance in its National Assessment of Educational Progress. That approach was adopted by a Trend in Mathematics and Science Study (TIMSS) for international comparison. In this paper, baseline data from TIMSS 1995 are analyzed at the seventh grade to examine correlation of plausible scores between mathematics and science. Canonical correlation is introduced to address a non-additive nature of correlation coefficient and reduce Type I error in the result aggregation. Besides revealing the impact of interdisciplinary correlation, this investigation reconfirms importance of student performance in each core subject.