The hierarchical age-period-cohort (HAPC) model, a class of mixed or hierarchical linear models, is now widely used in cohort analysis. This model treats the levels of the period and cohort factors as the groups to which each subject belongs and treats period and cohort effects as random effects. The HAPC model, however, has been criticized as being likely to produce subtle cohort effects even when definite effects are expected. This study elucidates why the HAPC model has such a serious deficiency. To overcome this problem, we revisit the Bayesian age-period-cohort (BAPC) model and focus on its method of taking the first-order differences in successive effect parameters. Like the HAPC model, the BAPC model can be interpreted as a mixed model, but it takes different approaches to the identification problem in cohort analysis. We demonstrate both models’ estimates for trends in the proportion of male college graduates in Japan, which seem to be dominantly affected by cohort. The results show that the HAPC model estimates near-zero cohort effects, whereas the BAPC model clearly detects positive cohort effects.
Causal inference means strategy that meets assumptions for identifying causal effects. Enormous theories and methods have already been introduced in causal inference literature. However, the theory and method imaged at the time of causality inference might differ between fields because causal inference does not necessarily correspond to partucular methods. In this paper, we systematically review the theory and method of causality inference while paying attention to the viewpoints of various fields such as epidemiology as well as social science. We also introduce the controversy continues among schools today as regards how to define and interpret causality.
Although statistical causal inference has become one of the major methods in quantitative research, systematic discussions of the meaning and impacts of it on quantitative sociology has not yet been found. In this paper, models of causal inference and related estimation methods are organized via the concept of heterogeneity. Above this arrangement, it is discussed, by using the sociological application of the multilevel analysis as a demonstration, that quantitative sociology has a tendency to treat heterogeneity in a different way from other quantitative research fields. As the conclusion, it is argued that there is a substantive difference between the approach of causal inference, which uses intervention or discontinuity from social processes, and that of the quantitative sociology, which usually refers to ordinary conceptual association in explaining social processes.
Analytical sociology has gained much popularity in recent years and thus, it is worthwhile revisiting its theoretical foundation and considering how it can contribute to our understanding of social mechanism. In this study, we focus on the concept of causality in analytical sociology and its relation to statistical causal inference. We propose two distinct model of the relationship between mechanism-based explanations and statistical causal inference-- discontinuity model and continuity model -- and argue that the latter should be adopted. Furthermore, we claim that the project of analytical sociology whose aim is to decompose macro social facts into individual action should be justified not by the "understandability" of action but by generalizability and transportability of the grasp of causality in action.
This study reviews research on causal analysis in sociology of education. In particular, it focuses on the following studies related to both sociology of education and social stratification: (1) educational inequality, (2) school and tracking effects, (3) neighborhood effects, (4) causal effects and heterogeneous causal effects of education, (5) social mobility and education, (6) single parenthood, and (7) multigenerational mobility. These studies have dealt with the difficulty of identifying causal effects and applied several methods of causal inference developed in statistics and other social science disciplines to sociological studies. Moreover, the studies have focused not only on identifying and estimating the causal effects, but also on understanding other structural sources such as confounding and selection processes that produce the associations. Based on this literature review, this paper highlights the future tasks of causal inference using observational data and discusses the usefulness of the causal inference framework in comprehending social and sociological phenomena.
This paper studies causal inference with a single-treated case using the synthetic control method (SCM) with both empirical data and Monte Carlo Simulations. SCM identifies the causal effects of a single-treated case by constructing a synthetic control, a weighted average of controls, that resembles a treated unit on observables and pre-treatment outcomes. As an empirical illustration, we examine the effects of the Great Hanshin Earthquake on poverty in Hyogo prefecture. Several years after the earthquake, poverty (measured as proportions of welfare recipients) increased in Hyogo prefecture, and it continues to be higher than in counterfactual groups 15 years after the quake. We then compare performances between SCM and a Difference-in-Differences (DD) estimator. Results from Monte Carlo Simulations indicate that when unit-specific time trends are assumed to be absent, SCM and DD perform similarly. However, when such trends are assumed to be present, biases produced by DD are greater than those produced by SCM. This suggests that SCM is suitable for identifying causal effects with a single-treated case when there are unobserved time trends.
The purpose of this paper is to organize the APC models of the previous studies systematically by Bayesian statistical modeling. A constrain is essential to solve the identification problem derived from linear dependency between age, period and cohort, but the framework of APC analysis is not summarized at present. In this paper, we focus on minimization of parameters and show that each model can be expressed by assuming normal distribution as prior distribution. The ridge regression model equivalent the Intrinsic Estimator is the method to obtain the estimates corresponding to “average of all constrained solutions” by grasping the rank deficient of design matrix as a purely mathematical phenomenon and minimizing the square norm of all parameters. The Random walk model known as the Bayesian cohort model is the constraint that minimizes the weighted sum of squares about first-order differences in successive parameters and presume a time series structure to cope with the identification problem of APC. In addition to introducing equality constraint models and random effects models, we also investigated the mathematical mechanism about the estimates of each model by simulation.
The shape of the population distribution of political attitudes regarding maintenance- innovation was investigated. Previously, this shape has been analyzed by using a single item political ideology scale. However, it has been pointed out that the shape was strongly influenced by response difficulty and reaction bias. Therefore, we proposed two methodological solutions to solve these problems. Firstly, we estimated the shape of the population distribution of attitudes using the Generalized Graded Unfolding Model (GGUM) with Skewed Generalized Error Distribution (SGED) as the prior distribution of political attitudes. Secondly, we adopted a statistical model to correct for the reaction bias of response data. We also examined how the shape changes based on the degree of political knowledge. Results indicated that the distribution of attitudes converged in the middle before removing the reaction bias, whereas the distribution approached a normal distribution after correcting the bias. Moreover, political knowledge affected the shape of the distribution of attitudes. The distribution of people with low political knowledge converged in middle, whereas the distribution of people with high political knowledge had a large variance and small kurtosis. Based on these results, we have discussed the possibility that mechanisms of political attitude formation could be inferred differently depending on the degree of political knowledge.
The purpose of this study is to build a Bayesian model for the income distribution generating process. Mathematical models of income distribution have been developed in the social sciences field; however, these models lack empirical validity. Human capital approaches have been developed to estimate the effect of individual investment on earnings, but those approaches lack rigorous mathematical consistency with the probability distribution of income. There is no appropriate probability model for testing the empirical validity of the theory that can explain the genesis of the distribution through human capital. To solve the problem, we built a generative income distribution model, expressed as a stochastic model, which formally represents human capital theory and a rigorous micro-macro linkage. Using nationwide survey data in Japan, we estimate the posterior distributions of the parameters of the probabilistic toy model using Markov chain Monte Carlo method. Moreover, we try to check the predictive accuracy of the models using the widely appreciable information criteria and the leave-one-out cross-validation. As a result, we conjecture that the predictive accuracy of the theory-based model is as good as that of the generalized linear model and provides interesting information about latent parameters.
Japan is experiencing the depopulation of many rural areas due to the outmigration of younger residents, who seek higher wages than those accrued from local small- and medium-sized enterprises (SMEs). However, some rural residents appear to be relatively satisfied with their daily lives. Few studies have investigated variations among young residents’ attitudes toward their jobs and personal lives and willingness to remain living in or leave their residential rural areas. This study used open interviews with a group of rural SME workers as the basis for the creation and administration of social surveys among a sample of 220 young workers in Iide, Yamagata Prefecture, Northern Japan. Analyses of the survey results demonstrated that although most respondents were dissatisfied with their work salaries and positions, they expressed satisfaction with their residential communities. Participants who were more willing to remain in their residential rural area had a higher satisfaction from their leisure and progress at work, but also reported a lower satisfaction from their coworkers. Rural SMEs can retain more young workers in rural areas by ensuring a better work-life balance and empowering employees with a greater sense of responsibility.