Hitotsubashi University Auditorium
After reviewing the potential-outcome approach to estimating treatment effects from observational data, this talk discusses new estimators in Stata 14 for estimating average treatment effects from survival-time data and estimators for average treatments from endogenous treatment designs. The talk also covers new research on estimating quantile treatment effects.
One of the advantages of using Stata is not only its statistical capability but also its practical features for various data management tasks. In particular, using user-written programs, we can report estimation results in various formats. In my presentation, I will present some practical commands for data management and some techniques that enable us to efficiently summarize the estimation results.
The purpose of this study is to analyze the estimation bias due to attrition in panel data. Panel data have larger amounts of information and higher availability than cross-section data or time-series data. But the data are not without a problem. One of the problems is that respondents refuse to answer the question at some point of time.
Because panel data, by their nature, are follow-up research typically done annually basis, it is not possible to carry out the investigations with other subjects. It is not a problem as long as respondents drop off at random, but attrition may be influenced by observable and unobservable variables.
In such cases, we should not overlook the impact of attrition on estimation. In this presentation, I investigate the feature of attrition and use the information of attrition to execute the inverse-probability-weighting method and the bounds estimates that relax the assumption of estimation. Using various estimation methods, I consider how to deal with the estimation bias due to attriton.
An increasing number of developing countries have expanded health insurance programs for poor and disadvantaged groups to improve their access to health care facilities. This demand-driven approach, however, causes congestion in health care facilities. In some cases, these facilities discriminate against users, making them unwilling to use insurance. In this presentation, I show that in the Thai Nguyen province in Vietnam, health insurance nonutilization is highest at the central public hospital, which provides superior medical service but is hence most crowded. To overcome the endogeneity problem caused by the facility choice of the patient, which would be partly based on the prospect of insurance utilization, I use the imperfect instrumental-variables approach to estimate the lower bound of the parameter. I also introduce some Stata tips such as using macros and producing LaTeX output by which the estimation results are automatically updated during the presentation.
There have been many theoretical and empirical studies on the effects of income distribution on economic growth. This presentation uses Japanese prefectural panel data to empirically analyze how income distribution affects economic growth. Four measures of the income distribution are used in the system GMM estimations using Stata. The Gini indices, the income share of the third quintile, and the ratio of the income share of the top decile and the fifth decile show that income equality has positive effects on growth. The ratio of the income share of the bottom decile and the fifth decile does not have statistically significant effects.
Therefore, the estimation results show that the increased income inequality in Japan decreased the economic growth.
This study presents the benchmark model and a practical implementation of the macro stress test. Since the emergence of complicated instruments such as securitized products, the macro stress test has become prominent as both an internal risk management tool for financial institutions and a way for supervisors to maintain financial stability. However, there is currently no practical model for transforming the macro stress scenarios into the risk parameters of an institution's internal model. We develop a model for assessing a company's default risk using the generalized linear mixed-effects modeling functions of Stata as follows: First, we simultaneously estimate the impact of the company-specific, macroeconomic, and sector-specific risk factors using panel and time-series data. Second, we estimate the correlations among latent sector-specific factors. Third, we use logistic function as a link because it is much more popular than the probit function. Fourth, we provide a forecast procedure for both the probability of default and the stressed value at risk.
The prediction model that estimates the probability of the presence or the prognosis of disease is used as a supporting tool for medical personnel and patients. In this study, we constructed a prediction model that estimates the three-year rupture risk of unruptured cerebral aneurysms using data from Japanese cohorts. The model was constructed by survival analyses such as Cox proportional hazard analysis and then validated with independent external data. We employed a simple scoring system to present the final model with rounded predictor coefficients so that it can be easily used in a clinical setting. The TROPOD statement, which aims to improve the transparency of the reporting of a prediction model study announced this year, will also be referred to.
In the first part of this presentation, I report how Stata is used in and around Kyushu University. A few Stata users in Kyushu University heavily use Stata in clinical trials, observational epidemiology, genomic epidemiology, and social epidemiology. The users have published more than 100 papers in the past 10 years using Stata as a statistical analysis tool. Stata is used for designing clinical studies, for data management, and for drawing figures.
In the second part, I will illustrate its usage in conceiving and designing studies, doing statistical analyses for clinical trials, and drawing figures and constructing tables.
Yukinobu Kitamura, Hitotsubashi University
LightStone Corp., the official distributor of Stata in Japan.