Optimal allocation for linear combinations of linear predictors in generalized linear models

The problem of the allocation of experimental units to experimental groups is studied within the context of generalized linear models. Optimal designs for the estimation of linear combinations of linear predictors are characterized, using concepts from the theory of optimal design. If there is only one linear combination of interest, then the D-optimal allocation is equivalent to the well-known Neyman allocation of subsamples in stratified sampling. However, if the number of linear combinations equals the number of design points, or experimental groups, then the equal replication of all design points is D-optimal. For cases in between, there are no easily accessible general solutions to the problem, although some particular cases are solved, including: i estimation of the n- 1 possible comparisons with a control group in an n-point, one-factor design; and ii estimation of 2 one or two of the four natural parameters of a 2 factorial design. The A-optimal allocations are determined in general.     

Optimal allocations in sequential tests involving two populations with covariates

We consider the problem of testing which of two normally distributed treatments has the largest mean, when the tested populations incorporate a covariate. From the class of procedures using the invariant sequential probability ratio test we derive an optimal allocation that minimizes, in a continuous time setting, the expected sampling costs. Simulations show that this procedure reduces the number of observations from the costlier treatment and categories while maintaining an overall sample size closer to the “pairwise” procedure. A randomized trial example is given.     

Optimal experimental plan for multi-level stress testing with Weibull regression under progressive Type-II extremal censoring

In the design of constant-stress life-testing experiments, the optimal allocation in a multi-level stress test with Type-I or Type-II censoring based on the Weibull regression model has been studied in the literature. Conventional Type-I and Type-II censoring schemes restrict our ability to observe extreme failures in the experiment and these extreme failures are important in the estimation of upper quantiles and understanding of the tail behaviors of the lifetime distribution. For this reason, we propose the use of progressive extremal censoring at each stress level, whereas the conventional Type-II censoring is a special case. The proposed experimental scheme allows some extreme failures to be observed. The maximum likelihood estimators of the model parameters, the Fisher information, and asymptotic variance–covariance matrices of the maximum likelihood estimates are derived. We consider the optimal experimental planning problem by looking at four different optimality criteria. To avoid the computational burden in searching for the optimal allocation, a simple search procedure is suggested. Optimal allocation of units for two- and four-stress-level situations is determined numerically. The asymptotic Fisher information matrix and the asymptotic optimal allocation problem are also studied and the results are compared with optimal allocations with specified sample sizes. Finally, conclusions and some practical recommendations are provided.     

D-optimal minimax design criterion for two-level fractional factorial designs

A D-optimal minimax design criterion is proposed to construct two-level fractional factorial designs, which can be used to estimate a linear model with main effects and some specified interactions. D-optimal minimax designs are robust against model misspecification and have small biases if the linear model contains more interaction terms. When the D-optimal minimax criterion is compared with the D-optimal design criterion, we find that the D-optimal design criterion is quite robust against model misspecification. Lower and upper bounds derived for the loss functions of optimal designs can be used to estimate the efficiencies of any design and evaluate the effectiveness of a search algorithm. Four algorithms to search for optimal designs for any run size are discussed and compared through several examples. An annealing algorithm and a sequential algorithm are particularly effective to search for optimal designs.     

On parallel policies for ranking and selection problems

In this paper we develop and test experimental methodologies for selection of the best alternative among a discrete number of available treatments. We consider a scenario where a researcher sequentially decides which treatments are assigned to experimental units. This problem is particularly challenging if a single measurement of the response to a treatment is time-consuming and there is a limited time for experimentation. This time can be decreased if it is possible to perform measurements in parallel. In this work we propose and discuss asynchronous extensions of two well-known Ranking & Selection policies, namely, Optimal Computing Budget Allocation (OCBA) and Knowledge Gradient (KG) policy. Our extensions (Asynchronous Optimal Computing Budget Allocation (AOCBA) and Asynchronous Knowledge Gradient (AKG), respectively) allow for parallel asynchronous allocation of measurements. Additionally, since the standard KG method is sequential (it can only allocate one experiment at a time) we propose a parallel synchronous extension of KG policy – Synchronous Knowledge Gradient (SKG). Computer simulations of our algorithms indicate that our parallel KG-based policies (AKG, SKG) outperform the standard OCBA method as well as AOCBA, if the number of evaluated alternatives is small or the computing/experimental budget is limited. For experimentations with large budgets and big sets of alternatives, both the OCBA and AOCBA policies are more efficient.     

An easy method of constructing latin square designs balanced for the immediate residual and other order effects

Bradley (1958) proposed a very simple procedure for constructing latin square designs to counterbalance the immediate sequential effect for an even number of treatments. When the number of treatments is odd, balance in a single latin square is not possible. In the present note we have developed an analogous method for the construction of such designs which may be used for an even or odd number of treatments. A proof has also been offered to assure the general validity of the procedure.     

Efficient estimators for adaptive stratified sequential sampling

In stratified sampling, methods for the allocation of effort among strata usually rely on some measure of within-stratum variance. If we do not have enough information about these variances, adaptive allocation can be used. In adaptive allocation designs, surveys are conducted in two phases. Information from the first phase is used to allocate the remaining units among the strata in the second phase. Brown et al. [Adaptive two-stage sequential sampling, Popul. Ecol. 50 (2008), pp. 239–245] introduced an adaptive allocation sampling design – where the final sample size was random – and an unbiased estimator. Here, we derive an unbiased variance estimator for the design, and consider a related design where the final sample size is fixed. Having a fixed final sample size can make survey-planning easier. We introduce a biased Horvitz–Thompson type estimator and a biased sample mean type estimator for the sampling designs. We conduct two simulation studies on honey producers in Kurdistan and synthetic zirconium distribution in a region on the moon. Results show that the introduced estimators are more efficient than the available estimators for both variable and fixed sample size designs, and the conventional unbiased estimator of stratified simple random sampling design. In order to evaluate efficiencies of the introduced designs and their estimator furthermore, we first review some well-known adaptive allocation designs and compare their estimator with the introduced estimators. Simulation results show that the introduced estimators are more efficient than available estimators of these well-known adaptive allocation designs.     

Optimum multivariate stratified double sampling design: Chebyshev's Goal Programming approach

In stratified sampling when strata weights are unknown a double sampling technique may be used to estimate them. A large simple random sample from the unstratified population is drawn and units falling in each stratum are recorded. A stratified random sample is then selected and simple random subsamples are obtained out of the previously selected units of the strata. This procedure is called double sampling for stratification. If the problem of non-response is there, then subsamples are divided into classes of respondents and non-respondents. A second subsample is then obtained out of the non-respondents and an attempt is made to obtain the information by increasing efforts, persuasion and call backs. In this paper, the problem of obtaining a compromise allocation in multivariate stratified random sampling is discussed when strata weights are unknown and non-response is present. The problem turns out to be a multiobjective non-linear integer programming problem. An approximation of the problem to an integer linear programming problem by linearizing the non-linear objective functions at their individual optima is worked out. Chebyshev's goal programming technique is then used to solve the approximated problem. A numerical example is also presented to exhibit the practical application of the developed procedure.     

Marginally restricted sequential D‐optimal designs

In many experiments, not all explanatory variables can be controlled. When the units arise sequentially, different approaches may be used. The authors study a natural sequential procedure for “marginally restricted” D‐optimal designs. They assume that one set of explanatory variables (x₁) is observed sequentially, and that the experimenter responds by choosing an appropriate value of the explanatory variable x₂. In order to solve the sequential problem a priori, the authors consider the problem of constructing optimal designs with a prior marginal distribution for x₁. This eliminates the influence of units already observed on the next unit to be designed. They give explicit designs for various cases in which the mean response follows a linear regression model; they also consider a case study with a nonlinear logistic response. They find that the optimal strategy often consists of randomizing the assignment of the values of x₂.     

A data-driven selection of the number of clusters in the Dirichlet allocation model via Bayesian mixture modelling

In this paper, we consider a Bayesian mixture model that allows us to integrate out the weights of the mixture in order to obtain a procedure in which the number of clusters is an unknown quantity. To determine clusters and estimate parameters of interest, we develop an MCMC algorithm denominated by sequential data-driven allocation sampler. In this algorithm, a single observation has a non-null probability to create a new cluster and a set of observations may create a new cluster through the split-merge movements. The split-merge movements are developed using a sequential allocation procedure based in allocation probabilities that are calculated according to the Kullback–Leibler divergence between the posterior distribution using the observations previously allocated and the posterior distribution including a ‘new’ observation. We verified the performance of the proposed algorithm on the simulated data and then we illustrate its use on three publicly available real data sets.     

Bayesian two-stage optimal design for mixture models

In this paper, a Bayesian two-stage D–D optimal design for mixture experimental models under model uncertainty is developed. A Bayesian D-optimality criterion is used in the first stage to minimize the determinant of the posterior variances of the parameters. The second stage design is then generated according to an optimalityprocedure that collaborates with the improved model from the first stage data. The results show that a Bayesian two-stage D–D-optimal design for mixture experiments under model uncertainty is more efficient than both the Bayesian one-stage D-optimal design and the non-Bayesian one-stage D-optimal design in most situations. Furthermore, simulations are used to obtain a reasonable ratio of the sample sizes between the two stages.     

Multistage bandit problems

Consider two or more treatments with dichotomous responses. The total number N of experimental units are to be allocated in a fixed number r of stages. The problem is to decide how many units to assign to each treatment in each stage. Responses from selections in previous stages are available and can be considered but responses in the current stage are not available until the next group of selections is made. Information is updated via the Bayes theorem after each stage. The goal is to maximize the overall expected number of successes in the N units.Two forms of prior information are considered: (i) All arms have beta priors, and (ii) prior distributions have continuous densities. Various characteristics of optimal decisions are presented. For example, in most cases of (i) and (ii), the rate of the optimal size of the first stage cannot be greater than √N when r = 2.     

D-optimal designs with minimal and nearly minimal number of units

D-optimal designs are identified in classes of connected block designs with fixed block size when the number of experimental units is one or two more than the minimal number required for the design to be connected. An application of one of these results is made to identify D-optimal designs in a class of minimally connected row-column designs. Graph-theoretic methods are employed to arrive at the optimality results.     

The Calibration for Two-Phase Randomized Response Estimator

This article presents the calibration procedure of the two-phase randomized response (RR) technique for surveying the sensitive characteristic. When the sampling scheme is two-phase or double sampling, auxiliary information known from the entire population can be used, but the auxiliary information should be information available from both the first and second phases of the sample. If there is auxiliary information available from both the first and second phases, then we can improve the ordinary two-phase RR estimator by incorporating this information in the estimation procedure. In this article, we used the new two-step Newton's method for computing unknown constants in the calibration procedure and compared the efficiency of the proposed estimator through some numerical study.     

A general adaptive allocation design for continuous responses in the presence of covariates

Several adaptive allocation designs are available in the clinical trial literature for allocating the entering patients among two competing treatments, having binary responses and skewing the allocation in favor of the better treatment. No adaptive design is available for continuous responses in the presence of prognostic factors, which is not model based. In the present paper, a general allocation design is introduced which assumes no specific regression model or distribution of responses. Some performance characteristics of the design are studied. Some related inference, following the allocation, is also studied. The proposed procedure is compared with some possible competitors. A real data set is used to illustrate the applicability of the proposed design.     

Sequential designs for a Poisson regression model

The locally D-optimal design for the two-variable interaction Poisson regression model is not robust to parameter misspecification. To deal with this issue, a sequential design approach is investigated. Simulation is used to evaluate the sequential design. It is found that the sequential design is very efficient and quite robust to parameter misspecification on both unrestricted and restricted design regions.     

Optimal designs for response functions with a downturn

In many toxicological assays, interactions between primary and secondary effects may cause a downturn in mean responses at high doses. In this situation, the typical monotonicity assumption is invalid and may be quite misleading. Prior literature addresses the analysis of response functions with a downturn, but so far as we know, this paper initiates the study of experimental design for this situation. A growth model is combined with a death model to allow for the downturn in mean doses. Several different objective functions are studied. When the number of treatments equals the number of parameters, Fisher information is found to be independent of the model of the treatment means and on the magnitudes of the treatments. In general, A- and D_A-optimal weights for estimating adjacent mean differences are found analytically for a simple model and numerically for a biologically motivated model. Results on c-optimality are also obtained for estimating the peak dose and the EC₅₀ (the treatment with response half way between the control and the peak response on the increasing portion of the response function). Finally, when interest lies only in the increasing portion of the response function, we propose composite D-optimal designs.     

THE SIMPLE LINEAR CALIBRATION PROBLEM AS AN OPTIMAL EXPERIMENTAL DESIGN

ABSTRACT

The aim of this paper is to consider the linear calibration problem through an optimal design approach, to evaluate the approximate variance of the calibrating value and provide approximate confidence intervals. The sequential approach through Stochastic Approximation is also applied to obtain a D-optimal design is considered.     

Some further results on the MV-optimality of block designs for comparing test treatments to a standard treatment

In this paper we further consider the problem of determining optimal block designs which can be used to compare v test treatments to a standard treatment in experimental situations where the available experimental units are to be arranged in b blocks of size k. A design is said to be MV-optimal in such an experimental setting it is minimizes the maximal variance with which treatment differences involving the standard treatment are estimated. In this paper we derive some further sufficient conditions for a design to be MV-optimal in an experimental situation such as described above.     

Discrimination between two binary data models: sequentially designed experiments

In this paper we propose a sequential procedure to design optimum experiments for discriminating between two binary data models. For the problem to be fully specified, not only the mode1link functions should be provided but also their associated linear predictor structures. Further, we suppose that one of the models is true, albeit it is not known which of them. Under these assumptions the procedure consists of making sequential choices of single experimental units to discriminate between the rival models as efficiently as possible. Depending on whether the models are nested or not, alternative methods are proposed.

To illustrate the procedure, a simulation study for the classical case of pro bit versus logit model is presented. It enables us to estimate the total sample sizes required to gain a certain power of discrimination and compare them to sample sizes for methods that were previously suggested in the literature.