INNOVATIVE THINKING
RIGOROUS APPROACHES
REAL WORLD IMPACT

 

Project

Microsimulation Model for Predicting Paid Leave

Official Title
DOL Microsimulation Model on Worker Leave
Dates
09/2017–03/2020
Market
Workforce Development
Practice Areas
Paid & Worker Leave
Services
Applied Research Studies
Client
U.S. Department of Labor - Chief Evaluation Office (CEO)
Methodologies
Conventional (logit regression and k-NN) and Machine Learning (Random Forest and xGBoost) microsimulation methods
Python Computer Code Example

IMPAQ is working with the U.S. Department of Labor's Chief Evaluation Office (DOL/CEO) to develop a microsimulation model to simulate leave taking and other labor market behavior under various policy scenarios. In addition to the model itself, IMPAQ is developing code to create, operate, and administer an open-sourced modeling platform with full capacity to model complex labor market behavior.

This tool will support different states' quantitative evaluation of proposed paid leave policies, as more jurisdictions across the country examine the feasibility of instituting paid leave polices. In this regard, the goals of this microsimulation model are to:

  1. Produce a transparent, validated, and publicly available tool to predict the leave-taking behavior of US workers.
  2. Simulate state-level program costs of implementing leave programs based on estimated leave-taking behavior.
  3. Conduct population analyses and analyze the distribution of program benefits by demographic characteristics.
  4. Conduct policy simulations of implementing similarly structured paid leave benefit programs in a number of states, counties, or municipalities whose cost and program parameters can be observed.
  5. Provide access to a state-of-the-art microsimulation model for researchers, policy analysts, and interested members of the public to conduct analysis on a standardized platform.
Methodologies

Previous iterations of the model used a logistic regression to calculate leave taking probability for individuals. IMPAQ is testing to see if various machine learning predictive methods (such as Naïve Bayes, Random Forest, and Nearest Neighbor methods) would perform better than regression-based methods (logistic, multinomial, and ridge regression) to simulate leave taking behavior for individuals.

IMPAQ is also applying the latest clustering methods to the FMLA survey data to determine which combination of variables best predict leave-taking behavior. Our model currently features a user-selectable full suite of microsimulation methods for simulating 16 different types of behaviors under different policy rules.

Skills & Experience

On this project, the IMPAQ team consists of Senior Researchers, Economists, and Data Analysts with expertise in statistical and machine learning methods, data programming, and paid leave. Together, they apply complex microsimulation models and translate this work to the public policy arena. Researchers from the Institute for Women's Policy Research (IWPR) also contribute paid leave expertise to the project.