CUR - The Council on Undergraduate Research

Effects of Heterogeneity for Continuous, Binary and Correlational Data Meta-Analysis Using MATLAB

Mohammad I. Jamil, Department of Industrial Engineering, University of Houston, 4800 Calhoun Rd, Houston, TX 77004 Aditya K. Jaiswal, IntelliScience Training Institute, 2139 South 10th St, San Jose, CA 95112 Sridharan Pariyangat, IntelliScience Training Institute, 2139 South 10th St, San Jose, CA 95112 Rohith Sunil, IntelliScience Training Institute, 2139 South 10th St, San Jose, CA 95112 Sohail Zaidi, Department of Mechanical Engineering, San Jose State University, 1 Washington Square, San Jose, CA 95192

Meta-analysis is a powerful tool to synthesize medical data emerging from various research groups. It assists to comprehend the full scope of the study and its impact on the subsequent decisions based on final research outcomes. In this work, we identify the need of developing open source MATLAB models that can perform meta-analysis on the given data from various resources. Our previously developed MATLAB models performed meta-analysis synthesis on the data by using random effect and fixed effect models. In each case summary effects were calculated by performing meta-analysis on continuous, binary, and correlational data obtained from examples described by Borenstein et. al. [2007]. MATLAB models were found robust enough to provide the desired analysis [Uma, Jaidev, Zaidi,  NCUR 2019, SCCUR 2019]. In the current work, we address the concept of heterogeneity by identifying and quantifying it in effect sizes through our MATLAB models. As the observed variation in the estimated effect sizes include true variation and random error, there is a need to isolate true variance and then use it to create to identify various perspectives on the dispersion. To achieve this goal, a new MATLB model is developed that determines the Q statistics (a measure of weighted square deviations), the results of a statistical based on Q (i.e.P), the between-studies variance (T2), the between-studies standard deviation (T), and the ratio of the true heterogeneity to total observed variation (I2). This analysis provides evidence of heterogeneity in the true effect size. MATLAB model was tested by using data described in various examples by Borenstien et. al. [2007] and was found successful in addressing the issues of heterogeneity in the true effect sizes. The work presented in the conference will cover all the details and various aspects of the new MATLAB Model. 

Additional Abstract Information