Introduction to model convergence playlist

In April 2022 I gave a one hour seminar on model convergence at CARMA. In this video I talk briefly about my involvement in CARMA and the content of the seminar. This video introduces the playlist that contains extended versions of the content of the CARMA seminar. Model convergence (or non-convergence) is an issue that all researchers eventually face. Non-convergence means that a statistical software does not produce results or produces results that are not trustworthy. Non-convergence problems can be difficult to troubleshoot because they require some understanding on how a model and an estimation technique work. Causes of non-convergence can be grouped into two categories: computational issues and model identification issues. The playlist starts by introducing model convergence and why it might fail on a general level. Thereafter, I explain the meaning of model identification followed by how numerical optimization works. After this technical intro, there is a series of explanations on how different techniques can be used to troubleshoot model non-convergence. The playlist concludes with an explanation of how one can practice troubleshooting non-convergent videos.