Combining expectation maximization(EM)algorithm and Lasso method,we propose a methodology to estimate a mixed-frequency dynamic factor model on large macroeconomic panels with rag-ged edges.We apply this approach to nowcast China's gross domestic product(GDP)growth and decom-pose the resulting forecast revision into contributions from the news.We find that,(1)Our method im-proves the out-of-sample forecast accuracy in nowcasting Chinese GDP growth,compared to the existing methods;(2)From 2005Q1 to 2022Q1,growth of fiscal expenditure,growth in retail sales,growth of industry's value-added and growth of total imports affected the real-time forecast of China's GDP growth the most.