Predicting the demand for preschool teachers in China under the three-child policy: a comparative analysis between Guangdong Province and Jilin Province

Dongyuan Xu, Zhichao Wang, Jiwei ZhangFrontiers in Psychology 2025 · DOI

TL;DR

This paper utilizes Long Short-Term Memory (LSTM) neural networks to forecast preschool teacher demand in China through 2035 under the new "three-child policy." By comparing the economically vibrant Guangdong Province with the aging Jilin Province, the study reveals a stark regional divergence: while Guangdong faces a growing need for educators, Jilin is bracing for a continued decline in its school-age population.

What Problem Does This Paper Address?

China is currently navigating a complex demographic transition. After decades of the one-child policy, the government has rapidly shifted to two-child (2016) and three-child (2021) policies to combat a "low fertility trap." However, these policies do not affect all regions equally.

The authors identify a critical gap in educational planning: most existing forecasts are either national in scope or focus on a single city, ignoring the massive disparities between provinces. Furthermore, traditional forecasting tools (like CPPS software or simple Logistic models) often rely on fixed parameters that struggle to adapt to the non-linear, complex nature of population shifts. This paper seeks to provide a more granular, data-driven roadmap for provincial governments to allocate teacher resources effectively.

Methods

The researchers employed Long Short-Term Memory (LSTM) neural networks, a type of Recurrent Neural Network (RNN) specifically designed to handle long-term dependencies in time-series data.

Key technical details include:

  • Architecture: A two-layer LSTM (32 and 64 neurons respectively) followed by a fully connected output layer.
  • Optimization: The Adam algorithm was used for adaptive learning rates, with Mean Squared Error (MSE) as the loss function.
  • Feature Selection: Beyond just birth rates, the models incorporated the number of children currently in kindergartens, the number of classes, and existing student-to-teacher ratios.
  • Validation: The models were tested against historical data (2012–2022), achieving high goodness-of-fit ($R^2$ values between 0.74 and 0.83).

This approach is a step up from standard demographic modeling because it can better capture the "fading effects" of previous policies and the "lagging effects" of new ones.

Datasets

The study relied on publicly available longitudinal data from:

  1. The Educational Statistics Yearbook of China: Provided data on student enrollment and teacher counts.
  2. Provincial Statistical Bulletins: Sourced birth population data from the "Statistical Bulletin of National Economic and Social Development" for both Guangdong and Jilin.
  3. Timeframes: Data for Guangdong spanned 2012–2022, while Jilin’s data was extended back to 2010 to better capture its more volatile long-term trends.

Key Insights and Ideas

  • The Great Divergence: The most striking insight is that the three-child policy is not a "rising tide that lifts all boats." In Guangdong, economic opportunity and cultural traditions ("more children bring more blessings") sustain growth. In Jilin, a "double reduction" (negative natural growth plus net outward migration) means the policy may barely move the needle.
  • Economic Coupling: Population trends are inextricably linked to economic health. Guangdong’s ability to attract young migrants expands its childbearing age group, whereas Jilin’s reliance on traditional heavy industry has led to significant "population loss."
  • Quality vs. Quantity: Even in regions where the population is shrinking (like Jilin), the demand for teachers may not drop as sharply as the student count. This is due to the "China Education Modernization 2035" goal, which aims for lower student-to-teacher ratios to improve educational quality.
  • The 3-Year Buffer: There is a predictable lag between birth rate changes and preschool demand, allowing for a "forward-looking layout" of resources if governments use predictive modeling.

Key Results

  • Guangdong Province: Predicted to see an overall upward trend. The preschool population is expected to grow from ~4.99 million in 2023 to over 5.00 million by 2035, with a corresponding 0.29% increase in teacher demand.
  • Jilin Province: Predicted to follow a "wave-like" downward trend. The preschool population is expected to drop from ~415,000 in 2023 to 406,000 by 2035, a 2.19% decline.
  • Teacher-to-Student Ratios: By 2035, ratios are projected to be 14.68:1 in Guangdong and 12.71:1 in Jilin. While these meet China's basic standard (15:1), they still lag behind the "high standard" of 10:1 often seen in developed nations, suggesting a persistent "overcrowding" issue.

Limitations and Future Work

The authors acknowledge several hurdles:

  • Sample Size: The study only compares two provinces. Future research should expand to a panel database covering all provinces to improve generalizability.
  • Missing Variables: The current model does not explicitly include GDP growth rates, specific migration flow data, or the exact number of women of childbearing age, which could refine accuracy.
  • Policy Lag: Because the three-child policy was only introduced in 2021, its full impact isn't yet fully reflected in the historical training data. Continuous updates to the model will be necessary as more "post-policy" data becomes available.

Summary updated

Abstract

Our findings provide data support for local governments to formulate education policies that are coordinated with the changes in the school-age population and the demand for teachers. It also has reference significance for foreign countries.
BibTeX
@inproceedings{xu2025predicting,
  author = {Dongyuan Xu and Zhichao Wang and Jiwei Zhang},
  title = {{Predicting the demand for preschool teachers in China under the three-child policy: a comparative analysis between Guangdong Province and Jilin Province}},
  booktitle = {Frontiers in Psychology},
  year = {2025},
  doi = {10.3389/fpsyg.2025.1536476},
}
Figure