2012 Volume 8 Issue 1 Pages 9-18
When the equating coefficients of two IRT based scales are estimated by Mean and Sigma method in common examinees designs, we sometimes obtain inappropriate estimates due to error variances of the examinees latent trait estimates. Noguchi and Kumagai (2011) proposed a correction method for such cases and examined the validity of the corrected method using a simulation study. However, we found that this method could generate inappropriate estimates in some cases. In this paper, we have proposed a new method of estimation method using estimated population distributions, which can reduced the effect of the error variance.
In simulation studies, when the effect of the error variance was very large, this method was able to estimate appropriate coefficients, more accurately than the simple Mean and Sigma and, or Noguchi and Kumagai methods. Using PISA 2006 data, we examined the effectiveness of this method and found that it was not affected by the error variances of the examinees latent trait estimates.
