Literature Index

Subjects were given an experimental task in which they had to play the role of a quality-control researcher for a company. They had to consider a hypothetical experiment that involves testing a sample of batteries from a truck load, which may or may not be substandard. In the main experiment, subjects were given information about the prior probability of substandard truck loads (base rate), the degree of variability of battery life, and the mean difference between standard and substandard batteries, all of which are formally relevant to the decision, and they were also told the number of batteries in the truck (population size) that is formally irrelevant. The task was to decide (intuitively) how many batteries to test to achieve a specified error rate using a specified decision rule. In a second study, subjects were given a similar scenario, but asked to rate which pieces of information would be relevant to the decision. Subjects showed themselves to be sensitive to the effects of sample variability and base rate when making intuitive design decisions, though an odd effect of the mean difference factor was observed. There is also clear confirmation of a bias-to-weight sample size by population size as reported in earlier research using different kinds of judgment tasks.

The estimation of proportions is a subject which cannot be circumvented in a first survey sampling course. Estimating the proportion of voters in favour of a political party, based on a political opinion survey, is just one concrete example of this procedure. However, another important issue in survey sampling concerns the proper use of auxiliary information, which typically comes from external sources, such as administrative records or past surveys. Very often, an efficient insertion of the auxiliary information available will improve the precision of the estimations of the mean or the total when a regression estimator is used. Conceptually, it is difficult to justify using a regression estimator for estimating proportions. A student might want to know how the estimation of proportions can be improved when auxiliary information is available. In this article, I present estimators for a proportion which use the logistic regression estimator. Based on logistic models, this estimator efficiently facilitates a good modelling of survey data. The paper's second objective is to estimate a proportion using various sampling plans (such as a Bernoulli sampling and stratified designs). In survey sampling, each sample possesses its own probability and for a given unit, the inclusion probability denotes the probability that the sample will contain that particular unit. Bernoulli sampling may have an important pedagogical value, because students often have trouble with the concept of the inclusion probability. Stratified sampling plans may provide more insight and more precision. Some empirical results derived from applying four sampling plans to a real data base show that estimators of proportions may be made more efficient by the proper use of auxiliary information and that choosing a more satisfactory model may give additional precision. The paper also contains computer code written in S-Plus and a number of exercises.

In this article, I attempt to explicate the ethical prinicples of data analysis, to suggest some characteristics of research and researchers that give rise to ethical difficulties, and to provide recommendations for improved practice.

The author presents guidelines for the selection of statistical analysis software given to graduate students to work independently. Criteria for a good teaching program are delineated. Several software programs are evaluated: STATMASTER, Statistics and Probability, Monte Carlo Simulations, Survey Sampling, KEYSTAT, CAPSAS: Computer Assisted Program for the Selection of Appropriate Statistics, EDA: Exploratory Data Analysis, INTROSTAT, Statistics With Finesse, STATPAC, GANOVA: Generalized Analysis of Variance, Speedstat, Micro-DSS/Analysis, and Micro-TSP. Several statistics packages are briefly reviewed, including STATPRO, A-STAT, Computer Models for Management Science, Multiple Factor Analysis, a General Correlation Program, and Test Construction Package. An appendix lists criteria for evaluating software. Demonstration pages prepared with a TEXPRINT printer interface card in the APPLE II+ computer are included. (DWH)