An Interview with Terry Reese

Abstract

In July 2005, Bonnie Parks spoke with Terry Reese, digital production unit head at Oregon State University’s Valley Library (Corvallis). In this interview, Reese shares his views on topics ranging from electronic journal management to the impact Google Scholar might have on the way libraries provide access to information.     

Scatterplots

A pair of axes, a few data points liberally distributed around, click on the mouse and your software will sort you out a scatterplot. Or will it? R. Allan Reese can tell you better.     

Scatterplots revisited

How can a diagram present three or more variables at once? Allan Reese recommends keeping scatterplots two-dimensional—even if they represent higher dimensions of knowledge.     

Databases and documenting data

Codebooks are usually associated with spies and concealed information. Allan Reese proposes statisticians should adopt codebooks but for the opposite purpose, making data more understandable. Without a codebook, you are working in the dark.     

Darwin, Mendel and the evolution of evolution

The history of science is full of myths. Darwin has his fair share; but Gregor Mendel, his fellow scientist and contemporary, has suffered even more. R. Allan Reese disentangles what we like to believe about Mendel from what we should believe—and finds a modern species whose origin was not by conventional evolution.     

Line plots

You can see line plots or line graphs in any quality newspaper most days, presenting business data. Even the downmarket papers use them occasionally to illustrate trends. They are so common that we might assume that they are well understood. Most spreadsheets and statistical packages can produce them with ease. But all is not as it seems. R. Allan Reese explains the pitfalls.     

Bay! Bar charts

Allan Reese has often asserted that authors and editors appear less able to proofread graphs than text. It is hard to prove such a claim except by parading examples that, had they been expressed in text, would fall into the category that Private Eye calls "Colemanballs"—things you would never say with brain fully engaged. Here, he provides a few such examples.     

Does significance matter?

The name of this magazine was the subject of fierce debate. Although "significance" is a word often associated with statistics, it is very often misapplied or misunderstood. Its technical implications differ from the everyday use of the word and competing schools of thought within statistics have varying views on the correct usage. As the magazine is launched, R. Allan Reese examines some of these ideas from the viewpoint of a pragmatic data ­analyst.     

截面相关面板数据的部分异质结构的识别

在大维面板数据中,截面之间很可能呈现出部分异质的特征,即参数在截面间具有组群效应,同组参数相同而不同组参数相异。如果忽略部分异质性而采用完全异质或同质的方法,可能导致估计的不一致性以及统计推断无效性。鉴于已有的部分异质性的研究要么限定截面独立,要么局限于强因子情形,本文尝试在Reese和Westerlund（2015）[1]提出的允许强因子或非强因子存在的较一般的框架下探讨面板数据部分异质结构的识别问题。采用Pesaran（2006）[2] CCE （Common Correlated Effects）方法处理不同强弱的共同因子,并借鉴Su et al.,（2016）[3]的C-Lasso （Classifier- Least Absolute Shrinkage and Selection Operator）方法,对CCE变化后的方程构造带有加法-乘法惩罚项的惩罚最小二乘,优化后以同步地实现分组和参数的估计。理论分析表明,在强因子或半强因子情形中,本文所提方法在分组方面具有渐近一致性,即所有个体被正确分组的概率随着 而趋于1。同时,参数的Lasso估计和事后Lasso估计均具有渐近正态性。另外分析结果也表明,因子的强弱不会影响分组的一致性但会影响以上两种估计量的渐近正态性,因子越强,两种估计量收敛得越快。模拟结果则表明有限样本下,本文所提的方法在分组、参数估计和分组数确定方面均具有良好的表现。具体的,在强因子和不同的半强因子情形中,随着N,T的增加,分组和分组数正确率很快地上升到100%,而两种参数估计的均方根误差和偏差则明显地降低。最后,利用本文所提的方法,研究了人力资本对经济增长影响的部分异质性。     

Predicting Home Run Production in Major League Baseball Using a Bayesian Semiparametric Model

This article attempts to predict home run hitting performance of Major League Baseball players using a Bayesian semiparametric model. Following Berry, Reese and Larkey we include in the model effects for era of birth, season of play, and home ball park. We estimate performance curves for each player using orthonormal quartic polynomials. We use a Dirichlet process prior on the unknown distribution for the coefficients of the polynomials, and parametric priors for the other effects. Dirichlet process priors are useful in prediction for two reasons: (1) an increased probability of obtaining more precise prediction comes with the increased flexibility of the prior specification, and (2) the clustering inherent in the Dirichlet process provides the means to share information across players. Data from 1871 to 2008 were used to fit the model. Data from 2009 to 2016 were used to test the predictive ability of the model. A parametric model was also fit to compare the predictive performance of the models. We used what we called “pure performance” curves to predict future performance for 22 players. The nonparametric method provided superior predictive performance.     

Calibration Intervals in Linear Regression Models

Many of the existing methods of finding calibration intervals in simple linear regression rely on the inversion of prediction limits. In this article, we propose an alternative procedure which involves two stages. In the first stage, we find a confidence interval for the value of the explanatory variable which corresponds to the given future value of the response. In the second stage, we enlarge the confidence interval found in the first stage to form a confidence interval called, calibration interval, for the value of the explanatory variable which corresponds to the theoretical mean value of the future observation. In finding the confidence interval in the first stage, we have used the method based on hypothesis testing and percentile bootstrap. When the errors are normally distributed, the coverage probability of resulting calibration interval based on hypothesis testing is comparable to that of the classical calibration interval. In the case of non normal errors, the coverage probability of the calibration interval based on hypothesis testing is much closer to the target value than that of the calibration interval based on percentile bootstrap.     

Reply

Selection processes that are inherent in litigation complicate policy-oriented research of medical malpractice tort reforms. In regard to their deterrent impact, the range of potential inferences based on analyses of claim frequency is limited because plaintiffs only file a subset of potential claims. In regard to their impact on litigation costs, researchers often analyze data on claim disposition, but it is difficult to determine whether effects attributed to tort reforms reflect changes in litigant behavior or their influence on the selection of claims. In this article, we evaluate these problems and report results of our study of the effects of medical malpractice reforms on claim disposition.     

Some theoretical results on approximation methods 1

In this paper we give an approximation procedure to surfaces which are defined on a _p-dimensional region and are observable (disturbed with some noice) according to an experimental design. In this procedure we combine clustering methods, discriminant analysis and smoothing techniques.

In the second part of the paper is considered some investigations on statistical properties of linear smoothing procedures. We assume linear models and for a broad class of models we prove the consistence of the estimation of the expectation of observations after smoothing.

In the last section we give some results on efficiency.     

On a Characterization of Generalized Pareto Distribution Based on Generalized Order Statistics

In recent years, several attempts have been made to characterize the generalized Pareto distribution (GPD) based on the properties of order statistics and record values. In the present article, we give a characterization result on GPD based on the spacing of generalized order statistics.     

First and Second Derivatives in Time Series Classification Using DTW

In our previous work, we developed a new distance function based on a derivative and showed that our algorithm is effective. In contrast to well-known measures from the literature, our approach considers the general shape of a time series rather than standard distance of function (value) comparison. The new distance was used in classification with the nearest neighbor rule. Now we improve on our previous technique by adding the second derivative. In order to provide a comprehensive comparison, we conducted a set of experiments, testing effectiveness on 47 time series datasets from a wide variety of application domains. Our experiments show that this new method provides a significantly more accurate classification on the examined datasets.     

Comparing and generating Latin Hypercube designs in Kriging models

In Computer Experiments (CE), a careful selection of the design points is essential for predicting the system response at untried points, based on the values observed at tried points. In physical experiments, the protocol is based on Design of Experiments, a methodology whose basic principles are questioned in CE. When the responses of a CE are modeled as jointly Gaussian random variables with their covariance depending on the distance between points, the use of the so called space-filling designs (random designs, stratified designs and Latin Hypercube designs) is a common choice, because it is expected that the nearer the untried point is to the design points, the better is the prediction. In this paper we focus on the class of Latin Hypercube (LH) designs. The behavior of various LH designs is examined according to the Gaussian assumption with exponential correlation, in order to minimize the total prediction error at the points of a regular lattice. In such a special case, the problem is reduced to an algebraic statistical model, which is solved using both symbolic algebraic software and statistical software. We provide closed-form computation of the variance of the Gaussian linear predictor as a function of the design, in order to make a comparison between LH designs. In principle, the method applies to any number of factors and any number of levels, and also to classes of designs other than LHs. In our current implementation, the applicability is limited by the high computational complexity of the algorithms involved.     

Evidence-Based Assessment and Application of Prognostic Markers: The Long Way from Single Studies to Meta-Analysis

In recent years, several attempts have been made to characterize the generalized Pareto distributions (GPD) based on the properties of order statistics and record values. In the present article, we give some characterization results on GPD based on order statistics and generalized order statistics. Some characterizations of uniform distribution based on expectation of some functions of order statistics are also given.     

A new criterion of confidence set estimation: Improvement of the Neyman shortness

In past studies various criteria have been proposed for evaluating the performance of a confidence set. However, each of these criteria often causes some unsatisfactory results even for the standard models such as location model, scale model and multinormal model. In this article, we propose a new criterion so that the procedure of the confidence set estimation based on the criterion can lead to a desirable confidence set at least for the above models. The approach is on the basis of an improvement of the Neyman shortness according to two steps. The first step is some kind of theoretical improvement, referring to a proposal of Pratt. As a result, we get a solution to Pratt's paradox. In the second step, we adopt a kind of robust or minimax procedure without sticking to the uniform optimality. In conclusion, it is shown that the procedure based on our criterion produces a desirable and acceptable confidence set.     

Contrast function estimation for the drift parameter of ergodic jump diffusion process

In this paper, we consider an ergodic diffusion process with jumps whose drift coefficient depends on an unknown parameter. We suppose that the process is discretely observed. We introduce an estimator based on a contrast function, which is efficient without requiring any conditions on the rate at which the step discretization goes to zero, and where we allow the observed process to have nonsummable jumps. This extends earlier results where the condition on the step discretization was needed and where the process was supposed to have summable jumps. In general situations, our contrast function is not explicit and one has to resort to some approximation. In the case of a finite jump activity, we propose explicit approximations of the contrast function such that the efficient estimation of the drift parameter is feasible. This extends the results obtained by Kessler in the case of continuous processes.     

Models for papilloma multiplicity and regression: applications to transgenic mouse studies

In cancer studies that use transgenic or knockout mice, skin tumour counts are recorded over time to measure tumorigenicity. In these studies cancer biologists are interested in the effect of endogenous and/or exogenous factors on papilloma onset, multiplicity and regression. In this paper an analysis of data from a study conducted by the National Institute of Environmental Health Sciences on the effect of genetic factors on skin tumorigenesis is presented. Papilloma multiplicity and regression are modelled by using Bernoulli, Poisson and binomial latent variables, each of which can depend on covariates and previous outcomes. An EM algorithm is proposed for parameter estimation, and generalized estimating equations adjust for extra dependence between outcomes within individual animals. A Cox proportional hazards model is used to describe covariate effects on the onset of tumours.