This dissertation rethinks the concept of democratic backsliding, proposing a unified framework termed "regime contraction," which emphasizes the incremental erosion of democratic institutions. The research explores how domestic institutions, particularly NGOs, face legal and extralegal attacks as states transition toward autocracy. Chapter 3 justifies the use of the Freedom House dataset and introduces innovative methods such as leveraging large language models (LLMs) like GPT for understanding institutional dynamics. It highlights the use of LLMs to analyze changes in legislatures and governance, offering a novel approach to event-based analysis. Chapter 4 focuses on the validation of the Event-Regime Institutional Structures (ERIS) dataset, comparing it with established sources like Freedom House and V-Dem, and discussing internal and external validation methods to ensure robust results. It also includes a discussion of threat immediacy, illustrated through case studies of Russia and Nicaragua. Chapter 5 brings NGOs into the central analysis, explaining their vulnerability during democratic regressions. The chapter presents logistic regression models showing how NGO attacks increase as institutions weaken and governments resort to legal and coercive tactics to control civil society. The dissertation contributes both theoretical and empirical advancements, offering insights into the strategic nature of democratic regression and providing tools for identifying and countering regime contraction.