643 – Small Sample Property and Statistical Inference
Conditional Maximum Likelihood Rasch Model in Data Harmonization
Yan Wang
University of California at Los Angeles
Honghu Liu
University of California, Los Angeles
Heidi M. Crane
University of Washington,
Data from different studies often have large variability and data collected with various instruments usually have low comparability, even if they are in attempt to measure the same concept or construct. Pooling individual data is scientifically and technically very challenging. It requires the generation of harmonized datasets across studies. Data harmonization aims to promote common measure for the key indicators that can permit certain degrees of comparability over time and across studies. This common measure will be used to combine the datasets and therefore to increase the sample size and to allow for adjustment of confounding factors. We will review the statistical methods that will accommodate these differences to create the common latent trait to harmonize the measures. Finally, the method will used on a real data in practice to create the harmonized measures across different studies.