From Unstructured Text to Causal Inference: Addressing Proxy Measurement Error and Omitted Variable Bias

Unstructured text offers a rich source of nuanced information that is often difficult to capture with structured survey measures. However, Natural Language Processing (NLP) tools result in labels prone to measurement error, which can cause significant bias and invalid statistical inference. Although