In this talk, I will present an introduction to multilevel (mixed) models and give a demonstration of fitting them in Stata. I will show how to fit both random-intercept and random-coefficient models, as well as crossed models and models with special residual structures. While I will primarily work with linear (Gaussian) models, I will briefly consider generalized linear models to show that they are syntactically identical and hence no harder to fit.
Presentation will be conducted in English.
In this paper, I investigate the statistical pattern of human growth. As is well known, the human body growth has two stages. The first stage is realized by its ability to stand and walk and the development of various senses, including seeing, hearing, and smelling. Extending neural networks enable human beings to think and communicate with language. The second stage is concentrated mainly in reproductive function. In this paper, I use panel data covering individuals from infancy to 10 years of age. This covers the period of the first growth pattern, and I try to identify a turning point of how the growth pattern changes from the first to the second stages.
We estimate the term structure of the probability of default (PD) of Japanese public companies. In addition to accounting variables for each company, we use macroeconomic factors such as stock price index, general economic indicators, and oil price as explanatory variables. We also compute the expected loss and unexpected loss of a hypothetical loan portfolio based on the estimated term structure of PD, with and without such macroeconomic factors. We find that considering macroeconomic factors affects the expected loss more than the unexpected loss.
In the present study, the author predicts that the effect of extraversion on personal network size is different among the right (upper) and left (lower) part of the distribution and tests the prediction with quantile regression. The result shows that extraversion positively correlates with the 70th to 90th percentile of personal network size to a greater extent than that of the 30th to 10th percentile.
Providing evidence-based standard care is an important aspect of the quality of medical care. Although medical record abstraction is desirable when examining the care provided, it is sometimes prohibitively labor intensive. An alternative way is to use existing data collected for other purposes, such as cancer registry and insurance claims data. To capture detailed clinical situations from these electronic data, one must construct elaborate algorithms that define target patients and recommended care for such patients. We found that Stata was useful in this process.
Propensity-score analysis is a relatively new statistical methodology to estimate an intervention effect (for example, treatment) in observational data when randomized controlled trials, the gold standard of clinical effectiveness research, are not feasible or ethical. Propensity-score analysis uses a two-step process. I will show multiple Stata commands for creating propensity scores and estimating the effect of the intervention within groups of patients.
LightStone Corporation, the official distributor of Stata in Japan.