Survey Methodology

View the most recent version.

June 2007

Content note: At this moment, full content is available in PDF only.

To access the PDF publication, please use the "Full content in PDF" link on the sidebar (on the left-hand side of this page).

In this issue

This issue of Survey Methodology includes papers covering a variety of methodological subjects such as modeling and estimation, weighting and variance estimation, non-response and sampling.

In the first paper of the issue, Skinner and Vieira investigate the effect of clustered sampling on variance estimation in longitudinal surveys. They present theoretical arguments and empirical evidence of the effects of ignoring clustering in longitudinal analyses, and find that these effects tend to be larger than for corresponding cross-sectional analyses. They also compare traditional survey sampling based methods to account for clustering in variance estimation to a multi­level modeling approach.

Kovačević and Roberts compare three models for analyzing multiple spells arising from data collected through longitudinal surveys with complex survey designs, which can involve stratification and clustering. These models are variations of the Cox proportional hazards model along the same lines as those proposed in the literature by Lin and Wei (1989), Binder (1992) and Lin (2000). These three models are compared using data from Statistics Canada's Survey on Labour and Income Dynamics (SLID). This paper gives new insight into fitting Cox models to survey data containing multiple spells per individual, a situation that arises quite frequently. The paper also illustrates some of the challenges in fitting Cox models to survey data.

Elliott, in his paper, presents a method for balancing elevated variance due to extreme weights with potential bias using a Bayesian weight trimming method in generalized linear models. This is accomplished by using a stratified hierarchical Bayesian model in which strata are determined by the probabilities of inclusion or survey weights. He illustrates andevaluates the approach using simulations based on linear and logistic regression models, and an application using data from the Partners for Child Passenger Safety dataset.

Thepaper by Breidt, Opsomer, Johnson and Ranalli explores the use of semiparametric methods for the estimation of population means. In semiparametric estimation, some variables are assumed to be linearly related to the variable of interest while the other variables may have a complicated, unspecified relation to the variable of interest. The authors study theoretically the properties under the sampling design of the resulting estimators. In particular, they show the design-consistency and the asymptotic normality of their estimator. Their method is then applied to data from a survey of lakes in the northeastern United States.

Tanguay and Lavallée address the problem of estimating the depreciation of assets based on a database of price ratios. In their paper, the issue is that the ratios do not come from a random sample from the population of ratios. The authors argue that the distribution of ratios should converge to a Uniform distribution and propose a weighting scheme that will make the weighted empirical distribution function approximately uniform. The proposed method is illustrated by an example using data on the depreciation of automobiles.

Steel and Clark present an empirical and theoretical comparison of person­level generalized regression survey weights and integrated household­ level weights in the case of a simple random sample of households in which all household members are selected. They conclude that there is little or no loss in efficiency associated with integrated weighting.

Saigo, in his paper, proposes a bootstrap variance estimation procedure for two­ phase designs with high sampling fractions. The method uses common bootstrap techniques, but adjusts the values of the auxiliary variables for units that are selected in the first phase sample only. The proposed technique is illustrated using several commonly used estimators such as the ratio estimator, and estimators of the distribution function and quantiles. Results from a simulation study comparing the proposed method to several others are presented.

In the paper by Longford the problem of estimating the MSE of small area estimates is investigated. A composite estimator of the MSE of small area means is obtained by combining a model-based variance estimator and a naïve estimator of the MSE. The coefficient that combines the two estimators minimizes the expected MSE of the resulting composite estimator of the MSE. The proposed estimator is compared with existing estimators through several simulation studies.

Shao considers the problem of imputing for missing values when the nonresponse is nonignorable. In the situation where the nonresponse depends on a cluster level random effect, he shows that the commonly used mean imputed estimator is biased unless the mean of the cluster is used. For variance estimation, a jackknife variance estimation procedure for the proposed estimator is provided. The proposed estimator is compared with the mean imputed estimator by means of a simulation study.

In the final paper of this issue, Tiwari, Nigam and Pant make use of the idea of nearest proportional to size sampling designs to obtain optimal controlled sample designs where non­preferred samples have zero selection probabilities. The optimal controlled sampled designs are obtained by combining an initial inclusion probability proportional to size design and quadratic programming techniques to ensure that non­preferred samples have a zero selection probability. Their method is illustrated using several examples.

Harold Mantel, Deputy Editor

You need to use the free Adobe Reader to view PDF documents. To view (open) these files, simply click on the link. To download (save) them, right-click on the link. Note that if you are using Internet Explorer or AOL, PDF documents sometimes do not open properly. See Troubleshooting PDFs. PDF documents may not be accessible by some devices. For more information, visit the Adobe website or contact us for assistance.