Document Type

Dissertation - Open Access

Award Date

2024

Degree Name

Doctor of Philosophy (PhD)

Department / School

Psychology, Sociology and Rural Studies

First Advisor

Weiwei Zhang

Abstract

Forecasting population by educational attainment not only benefit government planning on allocating educational resources, labor market demand, and long-term human capital and overall well-being of society (Lutz et al. 2008), but also help predicting size and structure of future population (Lutz and KC 2011). To seek effective method to produce precise results with limited resources, this dissertation compared the Cohort Component Method (CCM) and Hamilton Perry (HP) method to forecast population that are 25 years and older with Associate’s degree and above, and evaluated how factors may impact the accuracy, and how factors interact with each other to influence the accuracy. The results reveal that differences in methods, lengths of predicting period, educational attainment levels, forecast years (year before COVID-19 and year after), measurements, geography levels and characteristics are related to different levels of forecasting accuracy, and these factors may interact with each other to impact the pattern of accuracy. This dissertation found that the HP method is generally more accurate than the CCM method to forecast population 25 years and older and with Associate’s degree and above overall at the national level and the CCM method is more accurate than the HP method at the state level in Florida and South Dakota for 1-year forecast in 2019. Longer predicting period is likely to have less accurate forecasts regardless of choices of tested methods and geographies, but with adjustment of measurements, forecasting for longer predicting period may have comparable accuracy result. Educational attainment levels had different preferences with the methods depending on the geography. The impact of COVID-19 pandemic on forecasting accuracy also depends on the choice of method and measurements in different geographies. This dissertation suggests considering the HP method for forecasts with longer period of forecasting, larger population groups, larger geographies, population group with clear trend patterns; and considering the CCM forecast for smaller population groups, smaller geographies, groups with no clear change patterns, when groups in the geography have reliable data sources for estimating population components (Birth, Death, Migration) changes. The dissertation emphasized the uniqueness of each forecasting project with the combination of different predicting features and elements (different population groups with distinct demographic attributes, geographies, lengths of prediction period) and discussed the importance of the evaluation process of each project to help selections on methods and measurements. The dissertation also discussed data and resource limitations, and the importance of data quality. With the rapid development of Machine Learning and Artificial Intelligence techniques, more effectively and efficiently toolboxes and software may be developed and applied in the demography field in the near future.

Publisher

South Dakota State University

Download

Share

COinS
 

Rights Statement

In Copyright