Thursday, Aug 10: 8:30 AM - 10:20 AM
1187
Invited Paper Session
Metro Toronto Convention Centre
Room: CC-715A
Applied
Yes
Main Sponsor
Survey Research Methods Section
Co Sponsors
Government Statistics Section
Indian Statistical Institute
Presentations
It is often of interest to combine available estimates of a similar quantity from multiple data sources. When the corresponding variances of each estimate are also available, a model should take into account the uncertainty of the estimates themselves as well as the uncertainty in the estimation of variances. In addition, if there exists a strong association between estimates and their variances, the correlation between these two quantities should also be considered. In this paper, we propose a bivariate hierarchical Bayesian model that jointly models the estimates and the associated estimated variances assuming a correlation between these two measures. We conduct simulations to explore the performance of the proposed bivariate Bayesian model and compare it to other commonly used methods under different correlation scenarios. The proposed bivariate Bayesian model has a wide range of applications. We illustrate its application in meta-analysis and small area estimation.
The United States Department of Agriculture's National Agricultural Statistics Service provides state estimates of the cash rent paid for various land-use categories based on the Cash Rents Survey (CRS). Some of the realized sample sizes are too small to support reliable direct estimates, and there are outliers. In addition, quantities of interest for geographically contiguous small areas in CRS display a spatial pattern. Statistical agencies are increasingly considering the use of small area models in the estimation process. These models can provide indirect but reliable estimates for small areas. Therefore, we propose a hierarchical Bayesian area-level two-component mixture model with spatial random effects to account for outliers and spatial correlation. The model incorporates two years of data and a discounting factor for the first year provides a prior for the hyperparameters that is not too tight. We assess the effectiveness of the spatial model based on a simulation study and a case study from 2022 CRS. The results show superior performance of the proposed model over the direct estimates and the original Fay-Herriot model.
Adaptive cluster sampling designs were proposed as a method that could
be used when sampling rare populations whose units tend to appear in clusters.
The resulting estimators are not based on any model assumptions and are design
unbiased and can have smaller variance than standard methods. Here we will argue
that when adaptive cluster sampling is appropriate the resulting inference problem is really
one of estimating a domain total. Moreover, the usual estimator does not take into account
all the available information in the design. Here we will present a quasi Bayesian approach
which incorporates information which is now ignored. We will see that the resulting
estimators are a significant improvement over current methods.
Understanding data from modern surveys and developing methodologies for data that may be subject to ordering, heterogeneity, or other biases are critical to data science. In this presentation, we present methods that utilize ordering information to provide greater power than traditional procedures and for valid, conservative estimation of population means in small or empty domains without additional model assumptions. We illustrate the applications using the new R package 'csurvey'. If time permits, we will also discuss our findings and advances toward strategies for minimizing biases in crowdsourcing. we will provide a sneak preview of our new statistical inference procedures for dealing with data subject to selection biases. This talk is dedicated to Joe Sedranks for his longtime contribution to inference from survey samples.
This talk will review some of the history of descriptive and analytic inference in survey sampling, from the standpoints of design-based-model-assisted and Bayesian approaches. Each of these approaches has been influenced by the existence and practices of the other, but they remain distinct and apart. Illustrations touch on the uses of randomization, design-model correspondence, survey weights, the role of likelihood, and multi-level models.