Inference and foundations

Filter results by

Search Help
Currently selected filters that can be removed

Keyword(s)

Geography

1 facets displayed. 0 facets selected.

Survey or statistical program

2 facets displayed. 0 facets selected.

Content

1 facets displayed. 0 facets selected.
Sort Help
entries

Results

All (82)

All (82) (20 to 30 of 82 results)

  • Articles and reports: 12-001-X201100211602
    Description:

    This article attempts to answer the three questions appearing in the title. It starts by discussing unique features of complex survey data not shared by other data sets, which require special attention but suggest a large variety of diverse inference procedures. Next a large number of different approaches proposed in the literature for handling these features are reviewed with discussion on their merits and limitations. The approaches differ in the conditions underlying their use, additional data required for their application, goodness of fit testing, the inference objectives that they accommodate, statistical efficiency, computational demands, and the skills required from analysts fitting the model. The last part of the paper presents simulation results, which compare the approaches when estimating linear regression coefficients from a stratified sample in terms of bias, variance, and coverage rates. It concludes with a short discussion of pending issues.

    Release date: 2011-12-21

  • Surveys and statistical programs – Documentation: 12-001-X201100211603
    Description:

    In many sample surveys there are items requesting binary response (e.g., obese, not obese) from a number of small areas. Inference is required about the probability for a positive response (e.g., obese) in each area, the probability being the same for all individuals in each area and different across areas. Because of the sparseness of the data within areas, direct estimators are not reliable, and there is a need to use data from other areas to improve inference for a specific area. Essentially, a priori the areas are assumed to be similar, and a hierarchical Bayesian model, the standard beta-binomial model, is a natural choice. The innovation is that a practitioner may have much-needed additional prior information about a linear combination of the probabilities. For example, a weighted average of the probabilities is a parameter, and information can be elicited about this parameter, thereby making the Bayesian paradigm appropriate. We have modified the standard beta-binomial model for small areas to incorporate the prior information on the linear combination of the probabilities, which we call a constraint. Thus, there are three cases. The practitioner (a) does not specify a constraint, (b) specifies a constraint and the parameter completely, and (c) specifies a constraint and information which can be used to construct a prior distribution for the parameter. The griddy Gibbs sampler is used to fit the models. To illustrate our method, we use an example on obesity of children in the National Health and Nutrition Examination Survey in which the small areas are formed by crossing school (middle, high), ethnicity (white, black, Mexican) and gender (male, female). We use a simulation study to assess some of the statistical features of our method. We have shown that the gain in precision beyond (a) is in the order with (b) larger than (c).

    Release date: 2011-12-21

  • Articles and reports: 12-001-X201100111446
    Description:

    Small area estimation based on linear mixed models can be inefficient when the underlying relationships are non-linear. In this paper we introduce SAE techniques for variables that can be modelled linearly following a non-linear transformation. In particular, we extend the model-based direct estimator of Chandra and Chambers (2005, 2009) to data that are consistent with a linear mixed model in the logarithmic scale, using model calibration to define appropriate weights for use in this estimator. Our results show that the resulting transformation-based estimator is both efficient and robust with respect to the distribution of the random effects in the model. An application to business survey data demonstrates the satisfactory performance of the method.

    Release date: 2011-06-29

  • Articles and reports: 12-001-X201100111451
    Description:

    In the calibration method proposed by Deville and Särndal (1992), the calibration equations take only exact estimates of auxiliary variable totals into account. This article examines other parameters besides totals for calibration. Parameters that are considered complex include the ratio, median or variance of auxiliary variables.

    Release date: 2011-06-29

  • Surveys and statistical programs – Documentation: 12-001-X201000111250
    Description:

    We propose a Bayesian Penalized Spline Predictive (BPSP) estimator for a finite population proportion in an unequal probability sampling setting. This new method allows the probabilities of inclusion to be directly incorporated into the estimation of a population proportion, using a probit regression of the binary outcome on the penalized spline of the inclusion probabilities. The posterior predictive distribution of the population proportion is obtained using Gibbs sampling. The advantages of the BPSP estimator over the Hájek (HK), Generalized Regression (GR), and parametric model-based prediction estimators are demonstrated by simulation studies and a real example in tax auditing. Simulation studies show that the BPSP estimator is more efficient, and its 95% credible interval provides better confidence coverage with shorter average width than the HK and GR estimators, especially when the population proportion is close to zero or one or when the sample is small. Compared to linear model-based predictive estimators, the BPSP estimators are robust to model misspecification and influential observations in the sample.

    Release date: 2010-06-29

  • Articles and reports: 11-536-X200900110806
    Description:

    Recent work using a pseudo empirical likelihood (EL) method for finite population inferences with complex survey data focused primarily on a single survey sample, non-stratified or stratified, with considerable effort devoted to computational procedures. In this talk we present a pseudo empirical likelihood approach to inference from multiple surveys and multiple-frame surveys, two commonly encountered problems in survey practice. We show that inferences about the common parameter of interest and the effective use of various types of auxiliary information can be conveniently carried out through the constrained maximization of joint pseudo EL function. We obtain asymptotic results which are used for constructing the pseudo EL ratio confidence intervals, either using a chi-square approximation or a bootstrap calibration. All related computational problems can be handled using existing algorithms on stratified sampling after suitable re-formulation.

    Release date: 2009-08-11

  • Articles and reports: 12-001-X200800110606
    Description:

    Data from election polls in the US are typically presented in two-way categorical tables, and there are many polls before the actual election in November. For example, in the Buckeye State Poll in 1998 for governor there are three polls, January, April and October; the first category represents the candidates (e.g., Fisher, Taft and other) and the second category represents the current status of the voters (likely to vote and not likely to vote for governor of Ohio). There is a substantial number of undecided voters for one or both categories in all three polls, and we use a Bayesian method to allocate the undecided voters to the three candidates. This method permits modeling different patterns of missingness under ignorable and nonignorable assumptions, and a multinomial-Dirichlet model is used to estimate the cell probabilities which can help to predict the winner. We propose a time-dependent nonignorable nonresponse model for the three tables. Here, a nonignorable nonresponse model is centered on an ignorable nonresponse model to induce some flexibility and uncertainty about ignorabilty or nonignorability. As competitors we also consider two other models, an ignorable and a nonignorable nonresponse model. These latter two models assume a common stochastic process to borrow strength over time. Markov chain Monte Carlo methods are used to fit the models. We also construct a parameter that can potentially be used to predict the winner among the candidates in the November election.

    Release date: 2008-06-26

  • Articles and reports: 11-522-X200600110392
    Description:

    We use a robust Bayesian method to analyze data with possibly nonignorable nonresponse and selection bias. A robust logistic regression model is used to relate the response indicators (Bernoulli random variable) to the covariates, which are available for everyone in the finite population. This relationship can adequately explain the difference between respondents and nonrespondents for the sample. This robust model is obtained by expanding the standard logistic regression model to a mixture of Student's distributions, thereby providing propensity scores (selection probability) which are used to construct adjustment cells. The nonrespondents' values are filled in by drawing a random sample from a kernel density estimator, formed from the respondents' values within the adjustment cells. Prediction uses a linear spline rank-based regression of the response variable on the covariates by areas, sampling the errors from another kernel density estimator; thereby further robustifying our method. We use Markov chain Monte Carlo (MCMC) methods to fit our model. The posterior distribution of a quantile of the response variable is obtained within each sub-area using the order statistic over all the individuals (sampled and nonsampled). We compare our robust method with recent parametric methods

    Release date: 2008-03-17

  • Articles and reports: 11-522-X200600110398
    Description:

    The study of longitudinal data is vital in terms of accurately observing changes in responses of interest for individuals, communities, and larger populations over time. Linear mixed effects models (for continuous responses observed over time) and generalized linear mixed effects models and generalized estimating equations (for more general responses such as binary or count data observed over time) are the most popular techniques used for analyzing longitudinal data from health studies, though, as with all modeling techniques, these approaches have limitations, partly due to their underlying assumptions. In this review paper, we will discuss some advances, including curve-based techniques, which make modeling longitudinal data more flexible. Three examples will be presented from the health literature utilizing these more flexible procedures, with the goal of demonstrating that some otherwise difficult questions can be reasonably answered when analyzing complex longitudinal data in population health studies.

    Release date: 2008-03-17

  • Articles and reports: 11-522-X200600110419
    Description:

    Health services research generally relies on observational data to compare outcomes of patients receiving different therapies. Comparisons of patient groups in observational studies may be biased, in that outcomes differ due to both the effects of treatment and the effects of patient prognosis. In some cases, especially when data are collected on detailed clinical risk factors, these differences can be controlled for using statistical or epidemiological methods. In other cases, when unmeasured characteristics of the patient population affect both the decision to provide therapy and the outcome, these differences cannot be removed using standard techniques. Use of health administrative data requires particular cautions in undertaking observational studies since important clinical information does not exist. We discuss several statistical and epidemiological approaches to remove overt (measurable) and hidden (unmeasurable) bias in observational studies. These include regression model-based case-mix adjustment, propensity-based matching, redefining the exposure variable of interest, and the econometric technique of instrumental variable (IV) analysis. These methods are illustrated using examples from the medical literature including prediction of one-year mortality following heart attack; the return to health care spending in higher spending U.S. regions in terms of clinical and financial benefits; and the long-term survival benefits of invasive cardiac management of heart attack patients. It is possible to use health administrative data for observational studies provided careful attention is paid to addressing issues of reverse causation and unmeasured confounding.

    Release date: 2008-03-17
Data (0)

Data (0) (0 results)

No content available at this time.

Analysis (69)

Analysis (69) (60 to 70 of 69 results)

  • Articles and reports: 12-001-X199200214487
    Description:

    This paper reviews the idea of robustness for randomisation and model-based inference for descriptive and analytic surveys. The lack of robustness for model-based procedures can be partially overcome by careful design. In this paper a robust model-based approach to analysis is proposed based on smoothing methods.

    Release date: 1992-12-15

  • Articles and reports: 12-001-X199200214488
    Description:

    In many finite population sampling problems the design that is optimal in the sense of minimizing the variance of the best linear unbiased estimator under a particular working model is bad in the sense of robustness - it leaves the estimator extremely vulnerable to bias if the working model is incorrect. However there are some important models under which one design provides both efficiency and robustness. We present a theorem that identifies such models and their optimal designs.

    Release date: 1992-12-15

  • Articles and reports: 12-001-X199100214504
    Description:

    Simple or marginal quota surveys are analyzed using two methods: (1) behaviour modelling (superpopulation model) and prediction estimation, and (2) sample modelling (simple restricted random sampling) and estimation derived from the sample distribution. In both cases the limitations of the theory used to establish the variance formulas and estimates when measuring totals are described. An extension of the quota method (non-proportional quotas) is also briefly described and analyzed. In some cases, this may provide a very significant improvement in survey precision. The advantages of the quota method are compared with those of random sampling. The latter remains indispensable in the case of large scale surveys within the framework of Official Statistics.

    Release date: 1991-12-16

  • Articles and reports: 12-001-X199100114521
    Description:

    Marginal and approximate conditional likelihoods are given for the correlation parameters in a normal linear regression model with correlated errors. This general likelihood approach is applied to obtain marginal and approximate conditional likelihoods for the correlation parameters in sampling on successive occasions under both simple random sampling on each occasion and more complex surveys.

    Release date: 1991-06-14

  • Articles and reports: 12-001-X199000114560
    Description:

    Early developments in sampling theory and methods largely concentrated on efficient sampling designs and associated estimation techniques for population totals or means. More recently, the theoretical foundations of survey based estimation have also been critically examined, and formal frameworks for inference on totals or means have emerged. During the past 10 years or so, rapid progress has also been made in the development of methods for the analysis of survey data that take account of the complexity of the sampling design. The scope of this paper is restricted to an overview and appraisal of some of these developments.

    Release date: 1990-06-15

  • Articles and reports: 12-001-X198900214568
    Description:

    The paper describes a Monte Carlo study of simultaneous confidence interval procedures for k > 2 proportions, under a model of two-stage cluster sampling. The procedures investigated include: (i) standard multinomial intervals; (ii) Scheffé intervals based on sample estimates of the variances of cell proportions; (iii) Quesenberry-Hurst intervals adapted for clustered data using Rao and Scott’s first and second order adjustments to X^2; (iv) simple Bonferroni intervals; (v) Bonferroni intervals based on transformations of the estimated proportions; (vi) Bonferroni intervals computed using the critical points of Student’s t. In several realistic situations, actual coverage rates of the multinomial procedures were found to be seriously depressed compared to the nominal rate. The best performing intervals, from the point of view of coverage rates and coverage symmetry (an extension of an idea due to Jennings), were the t-based Bonferroni intervals derived using log and logit transformations. Of the Scheffé-like procedures, the best performance was provided by Quesenberry-Hurst intervals in combination with first-order Rao-Scott adjustments.

    Release date: 1989-12-15

  • Articles and reports: 12-001-X198500114364
    Description:

    Conventional methods of inference in survey sampling are critically examined. The need for conditioning the inference on recognizable subsets of the population is emphasized. A number of real examples involving random sample sizes are presented to illustrate inferences conditional on the realized sample configuration and associated difficulties. The examples include the following: estimation of (a) population mean under simple random sampling; (b) population mean in the presence of outliers; (c) domain total and domain mean; (d) population mean with two-way stratification; (e) population mean in the presence of non-responses; (f) population mean under general designs. The conditional bias and the conditional variance of estimators of a population mean (or a domain mean or total), and the associated confidence intervals, are examined.

    Release date: 1985-06-14

  • Articles and reports: 12-001-X198400114351
    Description:

    Most sample surveys conducted by organizations such as Statistics Canada or the U.S. Bureau of the Census employ complex designs. The design-based approach to statistical inference, typically the institutional standard of inference for simple population statistics such as means and totals, may be extended to parameters of analytic models as well. Most of this paper focuses on application of design-based inferences to such models, but rationales are offered for use of model-based alternatives in some instances, by way of explanation for the author’s observation that both modes of inference are used in practice at his own institution.

    Within the design-based approach to inference, the paper briefly describes experience with linear regression analysis. Recently, variance computations for a number of surveys of the Census Bureau have been implemented through “replicate weighting”; the principal application has been for variances of simple statistics, but this technique also facilitates variance computation for virtually any complex analytic model. Finally, approaches and experience with log-linear models are reported.

    Release date: 1984-06-15

  • Articles and reports: 12-001-X198100214319
    Description:

    The problems associated with making analytical inferences from data based on complex sample designs are reviewed. A basic issue is the definition of the parameter of interest and whether it is a superpopulation model parameter or a finite population parameter. General methods based on a generalized Wald Statistics and its modification or on modifications of classical test statistics are discussed. More detail is given on specific methods-on linear models and regression and on categorical data analysis.

    Release date: 1981-12-15
Reference (16)

Reference (16) (0 to 10 of 16 results)

  • Surveys and statistical programs – Documentation: 11-522-X201300014259
    Description:

    In an effort to reduce response burden on farm operators, Statistics Canada is studying alternative approaches to telephone surveys for producing field crop estimates. One option is to publish harvested area and yield estimates in September as is currently done, but to calculate them using models based on satellite and weather data, and data from the July telephone survey. However before adopting such an approach, a method must be found which produces estimates with a sufficient level of accuracy. Research is taking place to investigate different possibilities. Initial research results and issues to consider are discussed in this paper.

    Release date: 2014-10-31

  • Surveys and statistical programs – Documentation: 12-001-X201300211887
    Description:

    Multi-level models are extensively used for analyzing survey data with the design hierarchy matching the model hierarchy. We propose a unified approach, based on a design-weighted log composite likelihood, for two-level models that leads to design-model consistent estimators of the model parameters even when the within cluster sample sizes are small provided the number of sample clusters is large. This method can handle both linear and generalized linear two-level models and it requires level 2 and level 1 inclusion probabilities and level 1 joint inclusion probabilities, where level 2 represents a cluster and level 1 an element within a cluster. Results of a simulation study demonstrating superior performance of the proposed method relative to existing methods under informative sampling are also reported.

    Release date: 2014-01-15

  • Surveys and statistical programs – Documentation: 12-001-X201200211758
    Description:

    This paper develops two Bayesian methods for inference about finite population quantiles of continuous survey variables from unequal probability sampling. The first method estimates cumulative distribution functions of the continuous survey variable by fitting a number of probit penalized spline regression models on the inclusion probabilities. The finite population quantiles are then obtained by inverting the estimated distribution function. This method is quite computationally demanding. The second method predicts non-sampled values by assuming a smoothly-varying relationship between the continuous survey variable and the probability of inclusion, by modeling both the mean function and the variance function using splines. The two Bayesian spline-model-based estimators yield a desirable balance between robustness and efficiency. Simulation studies show that both methods yield smaller root mean squared errors than the sample-weighted estimator and the ratio and difference estimators described by Rao, Kovar, and Mantel (RKM 1990), and are more robust to model misspecification than the regression through the origin model-based estimator described in Chambers and Dunstan (1986). When the sample size is small, the 95% credible intervals of the two new methods have closer to nominal confidence coverage than the sample-weighted estimator.

    Release date: 2012-12-19

  • Surveys and statistical programs – Documentation: 12-001-X201200111688
    Description:

    We study the problem of nonignorable nonresponse in a two dimensional contingency table which can be constructed for each of several small areas when there is both item and unit nonresponse. In general, the provision for both types of nonresponse with small areas introduces significant additional complexity in the estimation of model parameters. For this paper, we conceptualize the full data array for each area to consist of a table for complete data and three supplemental tables for missing row data, missing column data, and missing row and column data. For nonignorable nonresponse, the total cell probabilities are allowed to vary by area, cell and these three types of "missingness". The underlying cell probabilities (i.e., those which would apply if full classification were always possible) for each area are generated from a common distribution and their similarity across the areas is parametrically quantified. Our approach is an extension of the selection approach for nonignorable nonresponse investigated by Nandram and Choi (2002a, b) for binary data; this extension creates additional complexity because of the multivariate nature of the data coupled with the small area structure. As in that earlier work, the extension is an expansion model centered on an ignorable nonresponse model so that the total cell probability is dependent upon which of the categories is the response. Our investigation employs hierarchical Bayesian models and Markov chain Monte Carlo methods for posterior inference. The models and methods are illustrated with data from the third National Health and Nutrition Examination Survey.

    Release date: 2012-06-27

  • Surveys and statistical programs – Documentation: 12-001-X201100211603
    Description:

    In many sample surveys there are items requesting binary response (e.g., obese, not obese) from a number of small areas. Inference is required about the probability for a positive response (e.g., obese) in each area, the probability being the same for all individuals in each area and different across areas. Because of the sparseness of the data within areas, direct estimators are not reliable, and there is a need to use data from other areas to improve inference for a specific area. Essentially, a priori the areas are assumed to be similar, and a hierarchical Bayesian model, the standard beta-binomial model, is a natural choice. The innovation is that a practitioner may have much-needed additional prior information about a linear combination of the probabilities. For example, a weighted average of the probabilities is a parameter, and information can be elicited about this parameter, thereby making the Bayesian paradigm appropriate. We have modified the standard beta-binomial model for small areas to incorporate the prior information on the linear combination of the probabilities, which we call a constraint. Thus, there are three cases. The practitioner (a) does not specify a constraint, (b) specifies a constraint and the parameter completely, and (c) specifies a constraint and information which can be used to construct a prior distribution for the parameter. The griddy Gibbs sampler is used to fit the models. To illustrate our method, we use an example on obesity of children in the National Health and Nutrition Examination Survey in which the small areas are formed by crossing school (middle, high), ethnicity (white, black, Mexican) and gender (male, female). We use a simulation study to assess some of the statistical features of our method. We have shown that the gain in precision beyond (a) is in the order with (b) larger than (c).

    Release date: 2011-12-21

  • Surveys and statistical programs – Documentation: 12-001-X201000111250
    Description:

    We propose a Bayesian Penalized Spline Predictive (BPSP) estimator for a finite population proportion in an unequal probability sampling setting. This new method allows the probabilities of inclusion to be directly incorporated into the estimation of a population proportion, using a probit regression of the binary outcome on the penalized spline of the inclusion probabilities. The posterior predictive distribution of the population proportion is obtained using Gibbs sampling. The advantages of the BPSP estimator over the Hájek (HK), Generalized Regression (GR), and parametric model-based prediction estimators are demonstrated by simulation studies and a real example in tax auditing. Simulation studies show that the BPSP estimator is more efficient, and its 95% credible interval provides better confidence coverage with shorter average width than the HK and GR estimators, especially when the population proportion is close to zero or one or when the sample is small. Compared to linear model-based predictive estimators, the BPSP estimators are robust to model misspecification and influential observations in the sample.

    Release date: 2010-06-29

  • Surveys and statistical programs – Documentation: 12-002-X20040027035
    Description:

    As part of the processing of the National Longitudinal Survey of Children and Youth (NLSCY) cycle 4 data, historical revisions have been made to the data of the first 3 cycles, either to correct errors or to update the data. During processing, particular attention was given to the PERSRUK (Person Identifier) and the FIELDRUK (Household Identifier). The same level of attention has not been given to the other identifiers that are included in the data base, the CHILDID (Person identifier) and the _IDHD01 (Household identifier). These identifiers have been created for the public files and can also be found in the master files by default. The PERSRUK should be used to link records between files and the FIELDRUK to determine the household when using the master files.

    Release date: 2004-10-05

  • Surveys and statistical programs – Documentation: 13F0026M2001003
    Description:

    Initial results from the Survey of Financial Security (SFS), which provides information on the net worth of Canadians, were released on March 15 2001, in The daily. The survey collected information on the value of the financial and non-financial assets owned by each family unit and on the amount of their debt.

    Statistics Canada is currently refining this initial estimate of net worth by adding to it an estimate of the value of benefits accrued in employer pension plans. This is an important addition to any asset and debt survey as, for many family units, it is likely to be one of the largest assets. With the aging of the population, information on pension accumulations is greatly needed to better understand the financial situation of those nearing retirement. These updated estimates of the Survey of Financial Security will be released in late fall 2001.

    The process for estimating the value of employer pension plan benefits is a complex one. This document describes the methodology for estimating that value, for the following groups: a) persons who belonged to an RPP at the time of the survey (referred to as current plan members); b) persons who had previously belonged to an RPP and either left the money in the plan or transferred it to a new plan; c) persons who are receiving RPP benefits.

    This methodology was proposed by Hubert Frenken and Michael Cohen. The former has many years of experience with Statistics Canada working with data on employer pension plans; the latter is a principal with the actuarial consulting firm William M. Mercer. Earlier this year, Statistics Canada carried out a public consultation on the proposed methodology. This report includes updates made as a result of feedback received from data users.

    Release date: 2001-09-05

  • Surveys and statistical programs – Documentation: 13F0026M2001002
    Description:

    The Survey of Financial Security (SFS) will provide information on the net worth of Canadians. In order to do this, information was collected - in May and June 1999 - on the value of the assets and debts of each of the families or unattached individuals in the sample. The value of one particular asset is not easy to determine, or to estimate. That is the present value of the amount people have accrued in their employer pension plan. These plans are often called registered pension plans (RPP), as they must be registered with Canada Customs and Revenue Agency. Although some RPP members receive estimates of the value of their accrued benefit, in most cases plan members would not know this amount. However, it is likely to be one of the largest assets for many family units. And, as the baby boomers approach retirement, information on their pension accumulations is much needed to better understand their financial readiness for this transition.

    The intent of this paper is to: present, for discussion, a methodology for estimating the present value of employer pension plan benefits for the Survey of Financial Security; and to seek feedback on the proposed methodology. This document proposes a methodology for estimating the value of employer pension plan benefits for the following groups:a) persons who belonged to an RPP at the time of the survey (referred to as current plan members); b) persons who had previously belonged to an RPP and either left the money in the plan or transferred it to a new plan; c) persons who are receiving RPP benefits.

    Release date: 2001-02-07

  • Surveys and statistical programs – Documentation: 11-522-X19990015642
    Description:

    The Longitudinal Immigration Database (IMDB) links immigration and taxation administrative records into a comprehensive source of data on the labour market behaviour of the landed immigrant population in Canada. It covers the period 1980 to 1995 and will be updated annually starting with the 1996 tax year in 1999. Statistics Canada manages the database on behalf of a federal-provincial consortium led by Citizenship and Immigration Canada. The IMDB was created specifically to respond to the need for detailed and reliable data on the performance and impact of immigration policies and programs. It is the only source of data at Statistics Canada that provides a direct link between immigration policy levers and the economic performance of immigrants. The paper will examine the issues related to the development of a longitudinal database combining administrative records to support policy-relevant research and analysis. Discussion will focus specifically on the methodological, conceptual, analytical and privacy issues involved in the creation and ongoing development of this database. The paper will also touch briefly on research findings, which illustrate the policy outcome links the IMDB allows policy-makers to investigate.

    Release date: 2000-03-02
Date modified: