Abstract
<title>Abstract</title> <p>This work employs a mixed-methods human-auditing and machine-learning approach, combined with elastic-net logistic regression, to empirically measure mediated economic policy uncertainty in the United States manufacturing sector from January 1985 to December 2025. The resulting monthly time-series Manufacturing Economic Policy Uncertainty index is calculated by creating a new Manufacturing Term Set to be added alongside Baker et al.’s (2016) preexisting Economic, Policy, and Uncertainty lexicons. This new set is created by first conducting a human seeding study to collect training data. Then, a Machine Learning model is trained, and an elastic-net logistic regression is fit to the Term Frequency-Inverse Document Frequency matrix to create a candidate pool of words with high term weights. A secondary human auditing team then selects words from the pool to be included in the final Manufacturing Term Set. Ultimately, the results show that industrial uncertainty and macroeconomic uncertainty generally follow similar trends, though the two clearly capture distinct underlying media sentiments. Furthermore, there is evidence that manufacturing uncertainty has a statistically significant inverse correlation with aggregate consumer confidence. In addition, while aggregate uncertainty shows a striking positive relationship, industrial uncertainty shows an inverse relationship with manufacturing output. Overall, the findings are useful for policymakers to empirically gauge the uncertainty-relieving or inducing effects of economic policies on the production sector, and for future research on the mediated uncertainty faced by manufacturers.</p>