Emergence of patient-focused measures in health policy research and sustained interest in large-scale assessments in education has led to broad use of psychometric models to assess unobserved traits such as quality of life, attitudes, and knowledge. Detection of Differential Item Functioning (DIF), including DIF attributable to complex combinations of covariates, is relevant in such a context. However, DIF detection techniques often rely on the comparison of pre-specified groups, with numeric covariates arbitrarily split (often at the median). Strobl et al. (2015) proposed Rasch Trees for DIF detection. This latent class approach, utilizing model-based recursive partitioning, serves as a global test for DIF. Rasch Trees can automatically detect groups exhibiting DIF without the need for pre-specification of groups or numeric cutpoints. We propose Rasch Tree Bagging (RT-Bag) as a new ensemble method for the detection of DIF. Combining several Rasch Trees built with random subsamples of data, RT-Bag has the potential to both identify latent groups that exhibit DIF and quantify the relative importance of covariates in group formation while retaining the benefits of Rasch Trees.