Fertility modeling in low fertility of Thai society

Abstract

Modeling of fertility has been widely studied in the literature to understand the past, current and future trend of population and to identify the suitable population policy and target population to sustain fertility level. The current studies show using Poisson regression models are relevant as fertility outcome rather than ordinary least square model. This paper concerns the suitable model to predict the number of children ever born and the desired number of additional children of ever-married women aged 18-49 in 2009 and compared the result from the linear regression model. The results showed that 1) the variance of number of children ever born and the desired number of additional children are similar to the mean, while most of sample reported “zero child-number intention”; 2) The Poisson regression model is relevant as number of children ever born, while Poisson zero-inflated regression is statistically appropriate for the modeling the desired number of additional children than Poisson regression in low fertility of Thailand; 3) compared with results from linear regression models show the statistically significant and the direction of relationships are slightly different with the appropriate model. Regarding an academics recommendations, it depends on an objective of research. If the research aims to analyses association between explanatory variables and count data variable, result of linear regression model in case concordance of analysis result was tested. Should the research aims to create a forecast equation of count data or focuses on magnitude of difference between the population subgroup. Analysis Result of regression model which is suitable for distribution of count data needs to be presented, namely Poisson regression model, Negative-binomial regression model, Zero-inflated Poisson regression model and Zero-inflated Negative-binomial regression model.