Bayesian modeling of joint and conditional distributions

In this paper, we study a Bayesian approach to flexible modeling of conditional distributions. The approach uses a flexible model for the joint distribution of the dependent and independent variables and then extracts the conditional distributions of interest from the estimated joint distribution. We use a finite mixture of multivariate normals (FMMN) to estimate the joint distribution. The conditional distributions can then be assessed analytically or through simulations. The discrete variables are handled through the use of latent variables. The estimation procedure employs an MCMC algorithm. We provide a characterization of the Kullback–Leibler closure of FMMN and show that the joint and conditional predictive densities implied by the FMMN model are consistent estimators for a large class of data generating processes with continuous and discrete observables. The method can be used as a robust regression model with discrete and continuous dependent and independent variables and as a Bayesian alternative to semi- and non-parametric models such as quantile and kernel regression. In experiments, the method compares favorably with classical nonparametric and alternative Bayesian methods.     

Sampling-based Randomised Designs for Causal Inference under the Potential Outcomes Framework

We establish the inferential properties of the mean-difference estimator for the average treatment effect in randomised experiments where each unit in a population is randomised to one of two treatments and then units within treatment groups are randomly sampled. The properties of this estimator are well understood in the experimental design scenario where first units are randomly sampled and then treatment is randomly assigned but not for the aforementioned scenario where the sampling and treatment assignment stages are reversed. We find that the inferential properties of the mean-difference estimator under this experimental design scenario are identical to those under the more common sample-first-randomise-second design. This finding will bring some clarifications about sampling-based randomised designs for causal inference, particularly for settings where there is a finite super-population. Finally, we explore to what extent pre-treatment measurements can be used to improve upon the mean-difference estimator for this randomise-first-sample-second design. Unfortunately, we find that pre-treatment measurements are often unhelpful in improving the precision of average treatment effect estimators under this design, unless a large number of pre-treatment measurements that are highly associative with the post-treatment measurements can be obtained. We confirm these results using a simulation study based on a real experiment in nanomaterials.     

Using nonparametric methods in social surveys: an empirical study

The most common form of data for socio-economic studies comes from survey sampling. Often the designs of such surveys are complex and use stratification as a method for selecting sample units. A parametric regression model is widely employed for the analysis of such survey data. However the use of a parametric model to represent the relationship between the variables can be inappropriate. A natural alternative is to adopt a nonparametric approach. In this article we address the problem of estimating the finite population mean under stratified sampling. A new stratified estimator based on nonparametric regression is proposed for stratification with proportional allocation, optimum allocation and post-stratification. We focus on an educational and labor-related context with natural populations to test the proposed nonparametric method. Simulated populations have also been considered to evaluate the practical performance of the proposed method.     

Optimal estimates and optimal predictors of finite population characteristics in the presence of nonresponse

Summary When samples from a finite population are studied, for instance by interview, there will usually be some units from which no response is obtained. In this paper optimal predictors of finite population characteristics, when nonresponse is present, are studied. The predictors are studied under simple regression superpopulation models. The optimal predictors are connected to the classical weighted sample estimates which are shown to be maximum likelihood estimates, provided the probability function is fully described by the sampling design. The predictors are compared with respects to their efficiencies for some simple models and a possible explanation to the fact that the poststratification estimate which compensate for nonresponse does no better than the simple estimate, is pointed out.     

Asymptotics for Autocovariances and Integrated Periodograms for Linear Processes Observed at Lower Frequencies

One of the most frequently used class of processes in time series analysis is the one of linear processes. For many statistical quantities, among them sample autocovariances and sample autocorrelations, central limit theorems are available in the literature. We investigate classical linear processes under a nonstandard observation pattern; namely, we assume that we are only able to observe the linear process at a lower frequency. It is shown that such observation pattern destroys the linear structure of the observations and leads to substantially different asymptotic results for standard statistical quantities. Central limit theorems are given for sample autocovariances and sample autocorrelations as well as more general integrated periodograms and ratio statistics. Moreover, for specific autoregressive processes, the possibilities to estimate the parameters of the underlying autoregression from lower frequency observations are addressed. Finally, we suggest for autoregressions of order 2 a valid bootstrap procedure. A small simulation study demonstrates the performance of the bootstrap proposal for finite sample size.     

Polygonal designs with blocks of size k≤10

The use of polygonal designs is motivated by the desire to avoid the selection of contiguous units in a sample from an ordered finite population. However very little is known about polygonal designs that have block size 5 or more. We present new polygonal designs with blocks of sizes 5 through 10, including the first designs with block sizes 9 and 10. For block sizes 5 through 7, we identify, with one possible exception, all values for the number of varieties for which a polygonal design exists. Received January 2001     

Bootstrap unit root tests in panels with cross-sectional dependency

We apply bootstrap methodology to unit root tests for dependent panels with N cross-sectional units and T time series observations. More specifically, we let each panel be driven by a general linear process which may be different across cross-sectional units, and approximate it by a finite order autoregressive integrated process of order increasing with T. As we allow the dependency among the innovations generating the individual series, we construct our unit root tests from the estimation of the system of the entire N cross-sectional units. The limit distributions of the tests are derived by passing T to infinity, with N fixed. We then apply bootstrap method to the approximated autoregressions to obtain critical values for the panel unit root tests, and establish the asymptotic validity of such bootstrap panel unit root tests under general conditions. The proposed bootstrap tests are indeed quite general covering a wide class of panel models. They in particular allow for very general dynamic structures which may vary across individual units, and more importantly for the presence of arbitrary cross-sectional dependency. The finite sample performance of the bootstrap tests is examined via simulations, and compared to that of commonly used panel unit root tests. We find that our bootstrap tests perform relatively well, especially when N is small.     

Survey Estimates by Calibration on Complex Auxiliary Information

In the last decade, calibration estimation has developed into an important field of research in survey sampling. Calibration is now an important methodological instrument in the production of statistics. Several national statistical agencies have developed software designed to compute calibrated weights based on auxiliary information available in population registers and other sources. This paper reviews some recent progress and offers some new perspectives. Calibration estimation can be used to advantage in a range of different survey conditions. This paper examines several situations, including estimation for domains in one‐phase sampling, estimation for two‐phase sampling, and estimation for two‐stage sampling with integrated weighting. Typical of those situations is complex auxiliary information, a term that we use for information made up of several components. An example occurs when a two‐stage sample survey has information both for units and for clusters of units, or when estimation for domains relies on information from different parts of the population. Complex auxiliary information opens up more than one way of computing the final calibrated weights to be used in estimation. They may be computed in a single step or in two or more successive steps. Depending on the approach, the resulting estimates do differ to some degree. All significant parts of the total information should be reflected in the final weights. The effectiveness of the complex information is mirrored by the variance of the resulting calibration estimator. Its exact variance is not presentable in simple form. Close approximation is possible via the corresponding linearized statistic. We define and use automated linearization as a shortcut in finding the linearized statistic. Its variance is easy to state, to interpret and to estimate. The variance components are expressed in terms of residuals, similar to those of standard regression theory. Visual inspection of the residuals reveals how the different components of the complex auxiliary information interact and work together toward reducing the variance.     

Estimation of symmetric functions of a finite population

Summary Estimation of symmetric functions under the assumptions of noninformativeness of labels in a finite population of distinguishable units has been examined. The primitive strategies of srs, sample mean, sample variance etc. are found to play important roles.     

Estimation under two stage random permutation models

Summary Estimation of the population mean under assumptions of non-informativeness of labels in a two stage finite population of distinguishable units has been studied. Under the random permutation model, for the two stage set up, sample mean, the natural estimator, is found to be the best.     

Comparing Inference Methods for Non‐probability Samples

Social and economic scientists are tempted to use emerging data sources like big data to compile information about finite populations as an alternative for traditional survey samples. These data sources generally cover an unknown part of the population of interest. Simply assuming that analyses made on these data are applicable to larger populations is wrong. The mere volume of data provides no guarantee for valid inference. Tackling this problem with methods originally developed for probability sampling is possible but shown here to be limited. A wider range of model‐based predictive inference methods proposed in the literature are reviewed and evaluated in a simulation study using real‐world data on annual mileages by vehicles. We propose to extend this predictive inference framework with machine learning methods for inference from samples that are generated through mechanisms other than random sampling from a target population. Describing economies and societies using sensor data, internet search data, social media and voluntary opt‐in panels is cost‐effective and timely compared with traditional surveys but requires an extended inference framework as proposed in this article.     

Efficient tests for the presence of a pair of complex conjugate unit roots in real time series

We introduce a framework which allows us to draw a clear parallel between the test for the presence of seasonal unit roots and that for unit root at frequency 0 (or π$\pi$). It relies on the properties of the complex conjugate integrated of order one processes which are implicitly at work in the real time series. In the same framework as that of Phillips and Perron (Biometrica 75 (1988) 335), we derive tests for the presence of a pair of conjugate complex unit roots. The asymptotic distribution we obtain are formally close to those derived by these authors but expressed with complex Wiener processes. We then introduce sequences of near-integrated processes which allow us to study the local-to-unity asymptotic of the above test statistics. We state a result on the weak convergence of the partial sum of complex near-random walks which leads to complex Orstein–Uhlenbeck processes. Drawing on Elliott et al. (Econometrica 64 (1996) 813) we then study the design of point-optimal invariant test procedures and compute their envelope employing local-to-unity asymptotic approximations. This leads us to introduce new feasible and near efficient seasonal unit root tests. Their finite sample properties are investigated and compared with the different test procedures already available (J. Econometrics 44 (1991) 215; 62 (1994) 415; 85 (1998) 269) and those introduced in the first part of the paper.     

Variable selection and functional form uncertainty in cross-country growth regressions

Regression analyses of cross-country economic growth data are complicated by two main forms of model uncertainty: the uncertainty in selecting explanatory variables and the uncertainty in specifying the functional form of the regression function. Most discussions in the literature address these problems independently, yet a joint treatment is essential. We present a new framework that makes such a joint treatment possible, using flexible nonlinear models specified by Gaussian process priors and addressing the variable selection problem by means of Bayesian model averaging. Using this framework, we extend the linear model to allow for parameter heterogeneity of the type suggested by new growth theory, while taking into account the uncertainty in selecting explanatory variables. Controlling for variable selection uncertainty, we confirm the evidence in favor of parameter heterogeneity presented in several earlier studies. However, controlling for functional form uncertainty, we find that the effects of many of the explanatory variables identified in the literature are not robust across countries and variable selections.     

Estimation Techniques for Ordinal Data in Multiple Frame Surveys with Complex Sampling Designs

Surveys usually include questions where individuals must select one in a series of possible options that can be sorted. On the other hand, multiple frame surveys are becoming a widely used method to decrease bias due to undercoverage of the target population. In this work, we propose statistical techniques for handling ordinal data coming from a multiple frame survey using complex sampling designs and auxiliary information. Our aim is to estimate proportions when the variable of interest has ordinal outcomes. Two estimators are constructed following model‐assisted generalised regression and model calibration techniques. Theoretical properties are investigated for these estimators. Simulation studies with different sampling procedures are considered to evaluate the performance of the proposed estimators in finite size samples. An application to a real survey on opinions towards immigration is also included.     

One-Sided Coverage Intervals for a Proportion Estimated from a Stratified Simple Random Sample

Using an Edgeworth expansion to speed up the asymptotics, we develop one-sided coverage intervals for a proportion based on a stratified simple random sample. To this end, we assume the values of the population units are generated from independent random variables with a common mean within each stratum. These stratum means, in turn, may either be free to vary or are assumed to be equal. The more general assumption is equivalent to a model-free randomization-based framework when finite population correction is ignored. Unlike when an Edgeworth expansion is used to construct one-sided intervals under simple random sampling, it is necessary to estimate the variance of the estimator for the population proportion when the stratum means are allowed to differ. As a result, there may be accuracy gains from replacing the normal  z -score in the Edgeworth expansion with a  t -score.     

Optimum allocation in stratified sampling with replacement

In a stratified sample, when sampling is done with replacement in each stratum a better estimate of the population mean can be achieved by considering the distinct units only. An explicit expression for the variance for the mean, of a stratified sample based on the distinct units only, is obtained. Then the optimum allocation for the different stratum are obtained by minimizing this variance subject to (i) total sample size being fixed, or (ii) the expected number of distinct units being fixed. Neyman’s solutions are obtained as special cases. The solutions finally arrived at are algebraically complex, hence, numerical methods are applied. In all examples, the variance of the estimates obtained by this method are smaller than the variances obtained by Neyman’s allocation. A part of this work was supported by the Office of the Ordinance Research; U.S.A. Grant (DA-AROL(D)-31-124-G83) when the author was at University of California, Berkeley.     

Combining random sampling and census strategies - Justification of inclusion probabilities equal to 1

Very often values of a size variable are known for the elements of a population we want to sample. For example, the elements may be clusters, the size variable denoting the number of units in a cluster. Then, it is quite usual to base the selection of elements on inclusion probabilities which are proportionate to the size values. To estimate the total of all values of an unknown variable for the units in the population of interest (i.e. for the units contained in the clusters) we may use weights, e.g. inverse inclusion probabilities. We want to clarify these ideas by the minimax principle. Especially, we will show that the use of inclusion probabilities equal to 1 is recommendable for units with high values of the size measure. AMS Classification 2000: Primary 62D05. Secondary 62C20     

A consistent nonparametric test of parametric regression functional form in fixed effects panel data models

We propose a consistent test for a linear functional form against a nonparametric alternative in a fixed effects panel data model. We show that the test has a limiting standard normal distribution under the null hypothesis, and show that the test is a consistent test. We also establish the asymptotic validity of a bootstrap procedure which is used to better approximate the finite sample null distribution of the test statistic. Simulation results show that the proposed test performs well for panel data with a large number of cross-sectional units and a finite number of observations across time.     

Mixing of direct, ratio, and product method estimators

In a paper by S rivenkataramana T racy [4], four methods of estimating a population total Y with the use of an auxiliary variable were introduced, given a random sample without replacement from that population. These methods were "built around the idea that estimating the population total is essentially equivalent to estimating the total corresponding to the non-sample units, since that corresponding to the sample units is known once the sample is drawn and measurements are made on it."
However, in the case of small sampling fractions the nonsample units constitute most of the population and no great improvement over the traditional estimators is to be expected. Therefore the methods are compared with the existing estimators and it turns out that they are special cases of the "mixing estimators", introduced in this paper. The latter estimators can be made asymptotically equivalent to the regression estimator and are therefore asymptotically superior to all other estimators. An exact comparison is carried out on the artificial example given in [4]. The statement in this paper that "the proposed estimators are to be preferred to the regression estimator for., superiority of performance in the case of small samples" is evidently misleading. Finally a comparison is made with other "mixing-type" estimators, that can prove very useful in practice.