The two-sample problem for interval estimation of the difference of two binomial p:s is considered. A new interval, the (AN) Andersson–Nerman, is presented and compared with the standard Wald interval and another good alternative, the Agresti interval. The AN interval turns out not to be without defects, but in its construction, where we seek to compensate for the correlation between the point estimator and its standard error, we will get an insight as to why the Wald interval does not work well. This seems not to have been previously taken into account for the two-sample case.

Numerous approximate confidence intervals in closed form have been suggested over the years for the parameter p in the bino-mial distribution. One of the oldest and still most advocated is the Wilson (score) interval, which in this article will be comparedwith the less well-known Andersson–Nerman (henceforth AN) interval. These intervals are quite closely related, but it will beshown analytically and illustrated by examples that the coverage probability of the AN interval always equals or exceeds that ofthe Wilson interval, while also having uniformly larger expected length. Asymptotic expressions for the coverage probability andexpected length of the AN interval are provided, using Edgeworth and Taylor expansions, respectively. The well-behaved Wilsonand AN pivots are furthermore contrasted with the problematic Wald pivot. The latter gives rise to the Wald interval, which isprobably one of the best known and most used procedures altogether in the history of statistical inference. Unfortunately, this in-terval performs poorly, but even to this day it is to be found in many current textbooks. Therefore, it is still of relevance to searchfor attractive alternatives based on sound statistical principles.

The Wald confidence interval for a binomial p should be completely gone by now from teaching as the interval to present and recommend. In reality, this unfortunately not so. This article makes a strong case for instead introducing the Wilson interval even in a first course in statistical inference.

When teaching we usually not only demonstrate/discuss how a certain method works, but, not less important, why it works. In contrast, the Wald confidence interval for a binomial p constitutes an excellent example of a case where we might be interested in why a method does not work. It has been in use for many years and, sadly enough, it is still to be found in many textbooks in mathematical statistics/statistics. The reasons for not using this interval are plentiful and this fact gives us a good opportunity to discuss all of its deficiencies and draw conclusions which are of more general interest. We will mostly use already known results and bring them together in a manner appropriate to the teaching situation. The main purpose of this article is to show how to stimulate students to take a more critical view of simplifications and approximations. We primarily aim for master’s students who previously have been confronted with the Wilson (score) interval, but parts of the presentation may as well be suitable for bachelor’s students. 

The problem of constructing a reasonably simple yet wellbehaved confidence interval for a binomial parameter p is old but still fascinating and surprisingly complex. During the last century, many alternatives to the poorly behaved standard Wald interval have been suggested. It seems though that the Wald interval is still much in use in spite of many efforts over the years through publications to point out its deficiencies. This paper constitutes yet another attempt to provide an alternative and it builds on a special case of a general technique for adjusted intervals primarily based on Wald type statistics. The main idea is to construct an approximate pivot with uncorrelated, or nearly uncorrelated, components. The resulting AN (Andersson–Nerman) interval, as well as a modification thereof, is compared with the well-renowned Wilson and AC (Agresti–Coull) intervals and the subsequent discussion will in itself hopefully shed some new light on this seemingly elementary interval estimation situation. Generally, an alternative to the Wald interval is to be judged not only by performance, its expression should also indicate why we will obtain a better behaved interval. It is argued that the well-behaved AN interval meets this requirement.

The Poisson distribution is here used to illustrate Bayesian inference concepts with the ultimate goal to construct credible intervals for a mean. The evaluation of the resulting intervals is in terms of mismatched priors and posteriors. The discussion is in the form of an imaginary dialog between a teacher and a student, who have met earlier, discussing and evaluating the Wald and score confidence intervals, as well as confidence intervals based on transformation and bootstrap techniques. From the perspective of the student the learning process is akin to a real research situation. The student is supposed to have studied mathematical statistics for at least two semesters.

High nonresponse is a very common problem in sample surveys today. In statistical terms we are worried about increased bias and variance of estimators for population quantities such as totals or means. Different methods have been suggested in order to compensate for this phenomenon. We can roughly divide them into imputation and calibration and it is the latter approach we will focus on here. A wide spectrum of possibilities is included in the class of calibration estimators. We explore linear calibration, where we suggest using a nonresponse version of the design-based optimal regression estimator. Comparisons are made between this estimator and a GREG type estimator. Distance measures play a very important part in the construction of calibration estimators. We show that an estimator of the average response propensity (probability) can be included in the “optimal” distance measure under nonresponse, which will help to reduce the bias of the resulting estimator. To illustrate empirically the theoretically derived results for the suggested estimators, a simulation study has been carried out. The population is called KYBOK and consists of clerical municipalities in Sweden, where the variables include financial as well as size measurements. The results are encouraging for the “optimal” estimator in combination with the estimated average response propensity, where the bias was reduced for most of the Poisson sampling cases in the study.

The Poisson distribution is here used to illustrate Bayesian inference concepts with the ultimate goal to construct credible intervals for a mean. The evaluation of the resulting intervals is in terms of potential negative effects of mismatched priors and posteriors. The discussion is in the form of an imaginary dialogue between a teacher and a student, who have met earlier, discussing and evaluating the Wald and score confidence intervals, as well as confidence intervals based on transformation and bootstrap techniques. From the perspective of the student the learning process is akin to a real research situation. By this time the student  is supposed to have studied mathematical statistics for at least two semesters.

High nonresponse is a very common problem in sample surveys today. In statistical terms we are worried about increased bias and variance of estimators for population quantities such as totals or means. Different methods have been suggested in order to compensate for this phenomenon. We can roughly divide them into imputation and calibration and it is the latter approach we will focus on here. A wide spectrum of possibilities is included in the class of calibration estimators. We explore linear calibration, where we suggest using a nonresponse version of the design-based optimal regression estimator. Comparisons are made between this estimator and a GREG type estimator. Distance measures play a very important part in the construction of calibration estimators. We show that an estimator of the average response propensity (probability) can be included in the "optimal" distance measure under nonresponse, which will help reducing the bias of the resulting estimator.  To illustrate empirically the theoretically derived results for the suggested estimators, a simulation study has been carried out. The population is called KYBOK and consists of clerical municipalities in Sweden, where the variables include financial as well as size measurements. The  results are encouraging for the "optimal" estimator in combination with the estimated average response propensity, where the bias was highly reduced for the Poisson sampling cases in the study.