Journal of survey statistics and methodology, Jan 7, 2019
The most widespread method of computing confidence intervals (CIs) in complex surveys is to add a... more The most widespread method of computing confidence intervals (CIs) in complex surveys is to add and subtract the margin of error (MOE) from the point estimate, where the MOE is the estimated standard error multiplied by the suitable Gaussian quantile. This Wald-type interval is used by the American Community Survey (ACS), the largest US household sample survey. For inferences on small proportions with moderate sample sizes, this method often results in marked under-coverage and lower CI endpoint less than 0. We assess via simulation the coverage and width, in complex sample surveys, of seven alternatives to the Wald interval for a binomial proportion with sample size replaced by the 'effective sample size,' that is, the sample size divided by the design effect. Building on previous work by the present authors, our simulations address the impact of clustering, stratification, different stratum sampling fractions, and stratumspecific proportions. We show that all intervals undercover when there is clustering and design effects are computed from a simple design-based
In 2009, two major surveys in the Governments Division of the Census Bureau were redesigned to re... more In 2009, two major surveys in the Governments Division of the Census Bureau were redesigned to reduce sample size, save resources, and improve the precision of the estimates. We developed a new decision-based estimation method in which the collapsing of strata either by state and government type or by small and large size was determined by a series of hypothesis tests of the equality of fitted coefficients in linear relationships between attributes and their values in the previous census year. In this research, we study design-based variance estimation by a bootstrap method, for the new decision-based stratified regression estimates applied to the Annual Survey of Public Employment and Payroll. The bootstrap method, which goes beyond available theory, is validated through a small simulation study. We use the data from the 2007 Census of Governments Employment to illustrate our methods.
This article proposes a new random field model to describe the spatial variation of rainfall amou... more This article proposes a new random field model to describe the spatial variation of rainfall amounts accumulated over short periods of time. The model is intended to satisfy a set of desiderata motivated by the understanding of rainfall generating mechanisms and exploratory data analysis of datasets of this type. First and second order properties of the proposed model are derived, including the mean and covariance functions, as well as the families of marginal and bivariate distributions. Properties of the proposed model are shown by a mix of analytical derivations and numerical exploration that use Gauss-Hermite quadrature to approximate the required integrals. The proposed model also satisfies a stochastic dominance property, which is argued to be sensible and consistent with most rainfall data of this type. A study of identifiability is carried out, which strongly suggests all model parameters are identifiable. The generalized method of moments is proposed to estimate the parameters, and the properties of these estimators are explored based on simulated data.
Statistical theory and related fields, Jul 3, 2019
A recent paper [Thibaudeau, Slud, and Gottschalck (2017). Modeling log-linear conditional probabi... more A recent paper [Thibaudeau, Slud, and Gottschalck (2017). Modeling log-linear conditional probabilities for estimation in surveys. The Annals of Applied Statistics, 11, 680-697] proposed a 'hybrid' method of survey estimation combining coarsely cross-classified design-based survey-weighted totals in a population with loglinear or generalised-linear model-based conditional probabilities for cells in a finer cross-classification. The models were compared in weighted and unweighted forms on data from the US Survey of Income and Program Participation (SIPP), a large national longitudinal survey. The hybrid method was elaborated in a book-chapter [Thibaudeau, Slud, & Cheng (2019). Small-area estimation of cross-classified gross flows using longitudinal survey data. In P. Lynn (Ed.), Methodology of longitudinal surveys II. Wiley] about estimating gross flows in (two-period) longitudinal surveys, by considering fixed versus mixed effect versions of the conditional-probability models and allowing for 3 or more outcomes in the later-period categories used to define gross flows within generalised logistic regression models. The methodology provided for point and interval small-area estimation, specifically area-level two-period labour-status gross-flow estimation, illustrated on a US Current Population Survey (CPS) dataset of survey respondents in two successive months in 16 states. In the current paper, that data analysis is expanded in two ways: (i) by analysing the CPS dataset in greater detail, incorporating multiple random effects (slopes as well as intercepts), using predictive as well as likelihood metrics for model quality, and (ii) by showing how Bayesian computation (MCMC) provides insights concerning fixed-versus mixed-effect model predictions. The findings from fixed-effect analyses with state effects, from corresponding models with state random effects, and fom Bayes analysis of posteriors for the fixed state-effects with other model coefficients fixed, all confirm each other and support a model with normal random state effects, independent across states.
Motivated by the problem of `quality filtering' of estimated counts in American Community Survey ... more Motivated by the problem of `quality filtering' of estimated counts in American Community Survey (ACS) tables, and of reporting small-domain coverage results from the 2010 decennial-census Post-Enumeration Survey (PES), this talk describes methods for ** Refreshments will be served at @2:50pm in Room 502 Hill Center **
Background: Count data derived from high-throughput deoxy-ribonucliec acid (DNA) sequencing is fr... more Background: Count data derived from high-throughput deoxy-ribonucliec acid (DNA) sequencing is frequently used in quantitative molecular assays. Due to properties inherent to the sequencing process, unnormalized count data is compositional, measuring relative and not absolute abundances of the assayed features. This compositional bias confounds inference of absolute abundances. Commonly used count data normalization approaches like library size scaling/rarefaction/subsampling cannot correct for compositional or any other relevant technical bias that is uncorrelated with library size. Results: We demonstrate that existing techniques for estimating compositional bias fail with sparse metagenomic 16S count data and propose an empirical Bayes normalization approach to overcome this problem. In addition, we clarify the assumptions underlying frequently used scaling normalization methods in light of compositional bias, including scaling methods that were not designed directly to address it. Conclusions: Compositional bias, induced by the sequencing machine, confounds inferences of absolute abundances. We present a normalization technique for compositional bias correction in sparse sequencing count data, and demonstrate its improved performance in metagenomic 16s survey data. Based on the distribution of technical bias estimates arising from several publicly available large scale 16s count datasets, we argue that detailed experiments specifically addressing the influence of compositional bias in metagenomics are needed.
bioRxiv (Cold Spring Harbor Laboratory), Nov 8, 2021
Background Clinically relevant health outcomes are often accompanied by changes in the diversity ... more Background Clinically relevant health outcomes are often accompanied by changes in the diversity of associated microbial communities. For instance, decreased gut microbiome diversity accompanies childhood diarrhea [1] and enteric infections [2] and has been
In many settings, multiple data collections and analyses on the same topic are summarised separat... more In many settings, multiple data collections and analyses on the same topic are summarised separately through statistical estimators of parameters and variances, and yet there are scientific reasons for sharing some statistical parameters across these different studies. This paper summarises what is known from large-sample theory about when estimators of a common structural parameter from several independent samples can be combined functionally, or more specifically linearly, to obtain an asymptotically efficient estimator from the combined sample. The main idea is that such combination can be done when the separate-sample nuisance parameters, if any exist, vary freely and independently of one another. The issues are illustrated using data from a multi-centre lung cancer clinical trial. Examples are presented to show that separate estimators cannot always be combined in this way, and that the functionally combined separate estimators may have low or 0 efficiency compared to the unified analysis that could be performed by pooling the datasets.
This article reflects the views of the authors and should not be construed to be those of the US ... more This article reflects the views of the authors and should not be construed to be those of the US Food and Drug Administration.
and Program Participation (SIPP) is a survey with a longitudinal structure and complex nonignorab... more and Program Participation (SIPP) is a survey with a longitudinal structure and complex nonignorable design, for which correct estimation requires using the weights. The longitudinal setting also suggests conditional-independence relations between survey variables and early-versus late-wave employment classifications. We state original assumptions justifying an extension of the partially model-based approach of Pfeffermann, Skinner and Humphreys [J. Roy. Statist. Soc. Ser. A 161 (1998) 13-32], accounting for the design of SIPP and similar longitudinal surveys. Our assumptions support the use of log-linear models of longitudinal survey data. We highlight the potential they offer for simultaneous bias-control and reduction of sampling error relative to direct methods when applied to small subdomains and cells. Our assumptions allow us to innovate by showing how to rigorously use only a longitudinal survey to estimate a complex log-linear longitudinal association structure and embed it in cross-sectional totals to construct estimators that can be more efficient than direct estimators for small cells.
We survey the problem of choosing an orthonormal basis for representing waveforms observed under ... more We survey the problem of choosing an orthonormal basis for representing waveforms observed under varying experimental conditions. When the distri- bution of waveform ordinates is Gaussian, this is a variant of classic Principal Components and Factor Analysis sta- tistical models. Interesting mathematical issues arise in • formulating models in terms of identifiable param- eters, which may lie in manifolds rather than open Euclidean-space regions, • finding an appropriate asymptotic framework, due to the comparable numbers of cross-classifying variables and replicate waveforms within each category, and • numerically computing model estimates. These problems will be illustrated in terms of real data involving transverse cross-sectional ultrasound pictures of the human tongue during speech.
This course was originally developed jointly with Benjamin Kedem and Paul Smith. It consists of m... more This course was originally developed jointly with Benjamin Kedem and Paul Smith. It consists of modules as indicated on the Course Syllabus. These fall roughly into three main headings: (A). R (& SAS) language elements and functionality, including computer-science ideas; (B). Numerical analysis ideas and implementation of statistical algorithms, primarily in R; and (C). Data analysis and statistical applications of (A)-(B). The object of the course is to reach a point where students have some facility in generating statistically meaningful models and outputs. Wherever possible, the use of R and numerical-analysis concepts is illustrated in the context of analysis of real or simulated data. The assigned homework problems will have the same flavor. The course formerly introduced Splus, where now we emphasize the use of R. The syntax is very much the same for the two packages, but R costs nothing and by now has much greater capabilities. Also, in past terms SAS has been introduced prim...
Scientists often need to know whether pairs of entities tend to occur together or independently. ... more Scientists often need to know whether pairs of entities tend to occur together or independently. Standard approaches to this issue use co-occurrence indices such as Jaccard, Sørensen-Dice, and Simpson. We show that these indices are sensitive to the prevalences of the entities they describe and that this invalidates their interpretability. We propose an index, α, that is insensitive to prevalences. Published datasets reanalyzed with both α and Jaccard’s index ( J ) yield profoundly different biological inferences. For example, a published analysis using J contradicted predictions of the island biogeography theory finding that community stability increased with increasing physical isolation. Reanalysis of the same dataset with the estimator α ˆ reversed that result and supported theoretical predictions. We found similarly marked effects in reanalyses of antibiotic cross-resistance and human disease biomarkers. Our index α is not merely an improvement; its use changes data interpretat...
Experiments embedded in surveys are a valuable way to test potential survey improvements; however... more Experiments embedded in surveys are a valuable way to test potential survey improvements; however, inference from these experiments can be difficult because one must account for the survey design aspects. Operational issues can restrict possible experimental design choices. In this paper, we discuss a design-based and a randomization-based approach to inference for experiments in an embedded survey. Additionally, some surveys are ongoing and present the option of utilizing data from previous time periods as well as the data from the test period. We discuss and compare an estimator that utilizes data only from the test period with one that compares outcome differences between the study period and a prior period across study treatment groups. Using an example from the American Community Survey (ACS) as well as a simulation, we compare the two inferential approaches and estimators. We find that the permutation-based inferential approach is a flexible and robust method for analyzing emb...
In 2009, two major surveys in the Governments Division of the U.S. Census Bureau were redesigned ... more In 2009, two major surveys in the Governments Division of the U.S. Census Bureau were redesigned to reduce sample size, save resources, and improve the precision of the estimates (Cheng, Corcoran, Barth and Hogue 2009). The new design divides each of the traditional state by government-type strata with sufficiently many units into two sub-strata according to each governmental unit’s total payroll, in order to sample less from the sub-stratum with small size units. The model-assisted approach is adopted in estimating population totals.
We consider the dual problems of choosing between competing small area models and validating mode... more We consider the dual problems of choosing between competing small area models and validating model assumptions in an area-level model. Many classes of small area models result in an estimate that is a convex combination of the direct and the marginal estimate for a given area. Therefore, competing models may share the same direct estimates, but give different marginal estimates as well as relative weight on the estimates. We discuss diagnostics to choose between competing models and parametric bootstrap methods to check for model validity and goodness of fit. We use the example of small area models related to the Voting Rights Act Section 203(b), which are used to estimate the number of limited English proficient and illiterate persons in certain language minority groups within jurisdictions using 5-year data from the American Community Survey.
The notion of respondent contact burden in sample surveys is defined, and a multi-stage process t... more The notion of respondent contact burden in sample surveys is defined, and a multi-stage process to develop policies for curtailing nonresponse follow-up is described with the goal of reducing this burden on prospective survey respondents. The method depends on contact history paradata containing information about contact attempts both for respondents and for sampled nonrespondents. By analysis of past data, policies to stop case follow-up based on control variables measured in paradata can be developed by calculating propensities to respond for paradata-defined subgroups of sampled cases. Competing policies can be assessed by comparing outcomes (lost interviews, numbers of contacts, patterns of reluctant participation, or refusal to participate) as if these stopping policies had been followed in past data. Finally, embedded survey experiments may be used to assess contact-burden reduction policies when these are implemented in the field. The multi-stage method described here abstrac...
Journal of survey statistics and methodology, Jan 7, 2019
The most widespread method of computing confidence intervals (CIs) in complex surveys is to add a... more The most widespread method of computing confidence intervals (CIs) in complex surveys is to add and subtract the margin of error (MOE) from the point estimate, where the MOE is the estimated standard error multiplied by the suitable Gaussian quantile. This Wald-type interval is used by the American Community Survey (ACS), the largest US household sample survey. For inferences on small proportions with moderate sample sizes, this method often results in marked under-coverage and lower CI endpoint less than 0. We assess via simulation the coverage and width, in complex sample surveys, of seven alternatives to the Wald interval for a binomial proportion with sample size replaced by the 'effective sample size,' that is, the sample size divided by the design effect. Building on previous work by the present authors, our simulations address the impact of clustering, stratification, different stratum sampling fractions, and stratumspecific proportions. We show that all intervals undercover when there is clustering and design effects are computed from a simple design-based
In 2009, two major surveys in the Governments Division of the Census Bureau were redesigned to re... more In 2009, two major surveys in the Governments Division of the Census Bureau were redesigned to reduce sample size, save resources, and improve the precision of the estimates. We developed a new decision-based estimation method in which the collapsing of strata either by state and government type or by small and large size was determined by a series of hypothesis tests of the equality of fitted coefficients in linear relationships between attributes and their values in the previous census year. In this research, we study design-based variance estimation by a bootstrap method, for the new decision-based stratified regression estimates applied to the Annual Survey of Public Employment and Payroll. The bootstrap method, which goes beyond available theory, is validated through a small simulation study. We use the data from the 2007 Census of Governments Employment to illustrate our methods.
This article proposes a new random field model to describe the spatial variation of rainfall amou... more This article proposes a new random field model to describe the spatial variation of rainfall amounts accumulated over short periods of time. The model is intended to satisfy a set of desiderata motivated by the understanding of rainfall generating mechanisms and exploratory data analysis of datasets of this type. First and second order properties of the proposed model are derived, including the mean and covariance functions, as well as the families of marginal and bivariate distributions. Properties of the proposed model are shown by a mix of analytical derivations and numerical exploration that use Gauss-Hermite quadrature to approximate the required integrals. The proposed model also satisfies a stochastic dominance property, which is argued to be sensible and consistent with most rainfall data of this type. A study of identifiability is carried out, which strongly suggests all model parameters are identifiable. The generalized method of moments is proposed to estimate the parameters, and the properties of these estimators are explored based on simulated data.
Statistical theory and related fields, Jul 3, 2019
A recent paper [Thibaudeau, Slud, and Gottschalck (2017). Modeling log-linear conditional probabi... more A recent paper [Thibaudeau, Slud, and Gottschalck (2017). Modeling log-linear conditional probabilities for estimation in surveys. The Annals of Applied Statistics, 11, 680-697] proposed a 'hybrid' method of survey estimation combining coarsely cross-classified design-based survey-weighted totals in a population with loglinear or generalised-linear model-based conditional probabilities for cells in a finer cross-classification. The models were compared in weighted and unweighted forms on data from the US Survey of Income and Program Participation (SIPP), a large national longitudinal survey. The hybrid method was elaborated in a book-chapter [Thibaudeau, Slud, & Cheng (2019). Small-area estimation of cross-classified gross flows using longitudinal survey data. In P. Lynn (Ed.), Methodology of longitudinal surveys II. Wiley] about estimating gross flows in (two-period) longitudinal surveys, by considering fixed versus mixed effect versions of the conditional-probability models and allowing for 3 or more outcomes in the later-period categories used to define gross flows within generalised logistic regression models. The methodology provided for point and interval small-area estimation, specifically area-level two-period labour-status gross-flow estimation, illustrated on a US Current Population Survey (CPS) dataset of survey respondents in two successive months in 16 states. In the current paper, that data analysis is expanded in two ways: (i) by analysing the CPS dataset in greater detail, incorporating multiple random effects (slopes as well as intercepts), using predictive as well as likelihood metrics for model quality, and (ii) by showing how Bayesian computation (MCMC) provides insights concerning fixed-versus mixed-effect model predictions. The findings from fixed-effect analyses with state effects, from corresponding models with state random effects, and fom Bayes analysis of posteriors for the fixed state-effects with other model coefficients fixed, all confirm each other and support a model with normal random state effects, independent across states.
Motivated by the problem of `quality filtering' of estimated counts in American Community Survey ... more Motivated by the problem of `quality filtering' of estimated counts in American Community Survey (ACS) tables, and of reporting small-domain coverage results from the 2010 decennial-census Post-Enumeration Survey (PES), this talk describes methods for ** Refreshments will be served at @2:50pm in Room 502 Hill Center **
Background: Count data derived from high-throughput deoxy-ribonucliec acid (DNA) sequencing is fr... more Background: Count data derived from high-throughput deoxy-ribonucliec acid (DNA) sequencing is frequently used in quantitative molecular assays. Due to properties inherent to the sequencing process, unnormalized count data is compositional, measuring relative and not absolute abundances of the assayed features. This compositional bias confounds inference of absolute abundances. Commonly used count data normalization approaches like library size scaling/rarefaction/subsampling cannot correct for compositional or any other relevant technical bias that is uncorrelated with library size. Results: We demonstrate that existing techniques for estimating compositional bias fail with sparse metagenomic 16S count data and propose an empirical Bayes normalization approach to overcome this problem. In addition, we clarify the assumptions underlying frequently used scaling normalization methods in light of compositional bias, including scaling methods that were not designed directly to address it. Conclusions: Compositional bias, induced by the sequencing machine, confounds inferences of absolute abundances. We present a normalization technique for compositional bias correction in sparse sequencing count data, and demonstrate its improved performance in metagenomic 16s survey data. Based on the distribution of technical bias estimates arising from several publicly available large scale 16s count datasets, we argue that detailed experiments specifically addressing the influence of compositional bias in metagenomics are needed.
bioRxiv (Cold Spring Harbor Laboratory), Nov 8, 2021
Background Clinically relevant health outcomes are often accompanied by changes in the diversity ... more Background Clinically relevant health outcomes are often accompanied by changes in the diversity of associated microbial communities. For instance, decreased gut microbiome diversity accompanies childhood diarrhea [1] and enteric infections [2] and has been
In many settings, multiple data collections and analyses on the same topic are summarised separat... more In many settings, multiple data collections and analyses on the same topic are summarised separately through statistical estimators of parameters and variances, and yet there are scientific reasons for sharing some statistical parameters across these different studies. This paper summarises what is known from large-sample theory about when estimators of a common structural parameter from several independent samples can be combined functionally, or more specifically linearly, to obtain an asymptotically efficient estimator from the combined sample. The main idea is that such combination can be done when the separate-sample nuisance parameters, if any exist, vary freely and independently of one another. The issues are illustrated using data from a multi-centre lung cancer clinical trial. Examples are presented to show that separate estimators cannot always be combined in this way, and that the functionally combined separate estimators may have low or 0 efficiency compared to the unified analysis that could be performed by pooling the datasets.
This article reflects the views of the authors and should not be construed to be those of the US ... more This article reflects the views of the authors and should not be construed to be those of the US Food and Drug Administration.
and Program Participation (SIPP) is a survey with a longitudinal structure and complex nonignorab... more and Program Participation (SIPP) is a survey with a longitudinal structure and complex nonignorable design, for which correct estimation requires using the weights. The longitudinal setting also suggests conditional-independence relations between survey variables and early-versus late-wave employment classifications. We state original assumptions justifying an extension of the partially model-based approach of Pfeffermann, Skinner and Humphreys [J. Roy. Statist. Soc. Ser. A 161 (1998) 13-32], accounting for the design of SIPP and similar longitudinal surveys. Our assumptions support the use of log-linear models of longitudinal survey data. We highlight the potential they offer for simultaneous bias-control and reduction of sampling error relative to direct methods when applied to small subdomains and cells. Our assumptions allow us to innovate by showing how to rigorously use only a longitudinal survey to estimate a complex log-linear longitudinal association structure and embed it in cross-sectional totals to construct estimators that can be more efficient than direct estimators for small cells.
We survey the problem of choosing an orthonormal basis for representing waveforms observed under ... more We survey the problem of choosing an orthonormal basis for representing waveforms observed under varying experimental conditions. When the distri- bution of waveform ordinates is Gaussian, this is a variant of classic Principal Components and Factor Analysis sta- tistical models. Interesting mathematical issues arise in • formulating models in terms of identifiable param- eters, which may lie in manifolds rather than open Euclidean-space regions, • finding an appropriate asymptotic framework, due to the comparable numbers of cross-classifying variables and replicate waveforms within each category, and • numerically computing model estimates. These problems will be illustrated in terms of real data involving transverse cross-sectional ultrasound pictures of the human tongue during speech.
This course was originally developed jointly with Benjamin Kedem and Paul Smith. It consists of m... more This course was originally developed jointly with Benjamin Kedem and Paul Smith. It consists of modules as indicated on the Course Syllabus. These fall roughly into three main headings: (A). R (& SAS) language elements and functionality, including computer-science ideas; (B). Numerical analysis ideas and implementation of statistical algorithms, primarily in R; and (C). Data analysis and statistical applications of (A)-(B). The object of the course is to reach a point where students have some facility in generating statistically meaningful models and outputs. Wherever possible, the use of R and numerical-analysis concepts is illustrated in the context of analysis of real or simulated data. The assigned homework problems will have the same flavor. The course formerly introduced Splus, where now we emphasize the use of R. The syntax is very much the same for the two packages, but R costs nothing and by now has much greater capabilities. Also, in past terms SAS has been introduced prim...
Scientists often need to know whether pairs of entities tend to occur together or independently. ... more Scientists often need to know whether pairs of entities tend to occur together or independently. Standard approaches to this issue use co-occurrence indices such as Jaccard, Sørensen-Dice, and Simpson. We show that these indices are sensitive to the prevalences of the entities they describe and that this invalidates their interpretability. We propose an index, α, that is insensitive to prevalences. Published datasets reanalyzed with both α and Jaccard’s index ( J ) yield profoundly different biological inferences. For example, a published analysis using J contradicted predictions of the island biogeography theory finding that community stability increased with increasing physical isolation. Reanalysis of the same dataset with the estimator α ˆ reversed that result and supported theoretical predictions. We found similarly marked effects in reanalyses of antibiotic cross-resistance and human disease biomarkers. Our index α is not merely an improvement; its use changes data interpretat...
Experiments embedded in surveys are a valuable way to test potential survey improvements; however... more Experiments embedded in surveys are a valuable way to test potential survey improvements; however, inference from these experiments can be difficult because one must account for the survey design aspects. Operational issues can restrict possible experimental design choices. In this paper, we discuss a design-based and a randomization-based approach to inference for experiments in an embedded survey. Additionally, some surveys are ongoing and present the option of utilizing data from previous time periods as well as the data from the test period. We discuss and compare an estimator that utilizes data only from the test period with one that compares outcome differences between the study period and a prior period across study treatment groups. Using an example from the American Community Survey (ACS) as well as a simulation, we compare the two inferential approaches and estimators. We find that the permutation-based inferential approach is a flexible and robust method for analyzing emb...
In 2009, two major surveys in the Governments Division of the U.S. Census Bureau were redesigned ... more In 2009, two major surveys in the Governments Division of the U.S. Census Bureau were redesigned to reduce sample size, save resources, and improve the precision of the estimates (Cheng, Corcoran, Barth and Hogue 2009). The new design divides each of the traditional state by government-type strata with sufficiently many units into two sub-strata according to each governmental unit’s total payroll, in order to sample less from the sub-stratum with small size units. The model-assisted approach is adopted in estimating population totals.
We consider the dual problems of choosing between competing small area models and validating mode... more We consider the dual problems of choosing between competing small area models and validating model assumptions in an area-level model. Many classes of small area models result in an estimate that is a convex combination of the direct and the marginal estimate for a given area. Therefore, competing models may share the same direct estimates, but give different marginal estimates as well as relative weight on the estimates. We discuss diagnostics to choose between competing models and parametric bootstrap methods to check for model validity and goodness of fit. We use the example of small area models related to the Voting Rights Act Section 203(b), which are used to estimate the number of limited English proficient and illiterate persons in certain language minority groups within jurisdictions using 5-year data from the American Community Survey.
The notion of respondent contact burden in sample surveys is defined, and a multi-stage process t... more The notion of respondent contact burden in sample surveys is defined, and a multi-stage process to develop policies for curtailing nonresponse follow-up is described with the goal of reducing this burden on prospective survey respondents. The method depends on contact history paradata containing information about contact attempts both for respondents and for sampled nonrespondents. By analysis of past data, policies to stop case follow-up based on control variables measured in paradata can be developed by calculating propensities to respond for paradata-defined subgroups of sampled cases. Competing policies can be assessed by comparing outcomes (lost interviews, numbers of contacts, patterns of reluctant participation, or refusal to participate) as if these stopping policies had been followed in past data. Finally, embedded survey experiments may be used to assess contact-burden reduction policies when these are implemented in the field. The multi-stage method described here abstrac...
Uploads
Papers by Eric Slud