Individual Submission Summary
Share...

Direct link:

Poster #220 - Early Childhood Policies in the Past Decade: Application of Machine Learning Research in Developmental Science

Fri, March 22, 7:45 to 9:15am, Baltimore Convention Center, Floor: Level 1, Exhibit Hall B

Integrative Statement

Within the past decade, state-funded preschool has roughly doubled in its coverage of 4 year-olds. While state legislature has played a critical role in rapidly expanding ECE programs across the nation, content and focus of legislations have rarely been examined. Additionally, the proliferation of new digital data sources have changed the nature of how state policymakers and researchers understand when and how government works. Thus, we investigated:
1. What are the most prominent key topics across state ECE legislations?
2. For each topic, does the successful passage of ECE legislation vary by political affiliation or track record of the primary sponsor?

The sample consists of 4,386 legislative data in the Early Care and Education Bill Tracking Database maintained by the National Conference of State Legislatures (NCSL). This database tracks ECE legislations from the 2008-2018 legislative sessions for 50 states and the territories. The NCSL database provides the full text of the bill, state, topic, status (adopted/failed/pending), legislative session time period in which the bill was deliberated, primary sponsors and their party affiliation.

We used Latent Dirichlet Allocation (LDA) (Blei, Ng, & Jordan, 2003) to parse/analyze the text of 4,386 ECE legislations and identify common occurrence of the words in various pieces of the legislation. LDA as a three-level hierarchical Bayesian model allows us to analyze patterns in legislative data and to estimate the topic probabilities (policy priorities) in a given ECE bill. We conducted an exploratory analysis of initially detected cluster structures of text and then imposed topic classification of text informed by developmental theories and early childhood research.

As expected, we conceptualized classification of two key topics: ‘ECE finance’ and ‘ECE services’ (Figure 1). Top five words most likely to appear under ‘Finance’ included fund, provided, appropriation, fiscal, and federal. Top five words most likely to appear under ‘Services’ included school, education, commissioner, health, family & children. ‘ECE services’ appears to be a comprehensive, multidimensional construct which we plan to examine further. Preliminary subconstructs of ‘services’ are what, for whom, where, and how: service components, targeted population, delivery settings, administrative leadership, and providers.

More interestingly, we found that stronger the track record of a primary sponsor, less probability the bill text focused on ‘finance’ (Figure 2). We speculate that this could be related to ‘unfunded mandates’ or concerns about viability/sustainability of ECE programs (funding) vs. (services) of existing ECE infrastructure. We have subsequently correlated the proportional presence of these topics in the legislation with the legislation’s success using the generalized (logistic) hierarchical linear model. Preliminary results indicate a positive correlation between ‘ECE services’ and successful passage. We plan to experiment with three- or four- topic solutions to see which topics the literature or the public considers important are conspicuously missing from the legislation and discuss this gap between public demand and actual legislation.

Authors