Analysis

To determine and measure the impact of informativeness we compare design-based and model-based variances of estimated parameters, as well as the estimated parameters themselves, in a logistic model under the assumption that the postulated model is true. An approach for assessing the impact of informativeness is given. In order to address the additional complexity of the impact of informativeness on power, we propose a new approximation for a linear combination of non-central chi-square distributions, using generalized design effects. A large simulation study, based on generating a population under the postulated model, using parameter estimates derived from the NPHS, allows us to detect and to measure the informativeness, and to compare the robustness of studied approaches.

There are a number of asymptotically equivalent procedures for deriving the Taylor series approximation of variances for complex statistics. In Binder and Patak (1994) the theoretical justification for one class of methods was derived. However, many of these methods can be derived for practical examples using straightforward techniques that are not clearly described in Binder and Patak. In this paper we give a “cookbook” approach that can be used for many examples, and that has been shown to have good finite sample properties. Normally the method of choice becomes clear through arguments such as model-assisted methods or linearizing the jackknife; however, using our approach yields the desired results more directly. As well, we present new results on the application of these techniques to two-phase samples.

We summarize some salient aspects of the theory of estimation functions for finite populations. In particular, we discuss the problem of estimation of means and totals and extend this theory to estimating functions. We then apply this estimating functions framework to the problem of estimating measures of income inequality. The resulting statistics are nonlinear functions of the observations. Some of them depend on the order of observations or quantiles. Consequently, the mean squared errors of these estimates are inexpressible by simple formulae and cannot be estimated by conventional variance estimation methods. We show that within the estimating function framework this problem can be resolved using the Taylor linearization method. Finally, we illustrate the proposed methodology using income data from Canadian Survey of Consumer Finance and comparing it to the ‘delete-one-cluster’ jackknifing method.

The Canadian Health and Disability Survey, administered as a supplement to the Canadian Labour Force Survey in October 1983, collected data on potentially disabled persons by means of a screening questionnaire and a follow-up questionnaire for those screened-in. The data from the screening questionnaire, consisting of a set of activities of daily living, were used to group respondents according to identifiable characteristics. A description of the groups of respondents is provided along with an evaluation of the methods used in their determination. An incompletely ordered severity scale is proposed.

Univariate statistical models, linear regression models and generalized linear models are briefly reviewed. Examples of a two-way analysis of variance, a three-way analysis of variance and logistic regression for a three way layout are given.

Statistics Canada, Canada’s central statistical agency, has been compiling national mortality statistics, including those on cancer mortality since 1921. Also, cancer incidence data are available from 1969.

The data quality of these files may be assessed in a variety of ways. Ratios of cancer mortality to incidence give some information on coverage errors. Micro-data matches between incidence and mortality files give an indication of misclassifications. As well, multiple registrations for cancer incidence may be duplicates. Completeness and availability of data items are also important for special studies.

In this paper, the feasibility of using these measures of data quality and the implications of these measures are discussed.

The problem of specifying and estimating the variance of estimated parameters based on complex sample designs from finite populations is considered. The results of this paper are particularly useful when the parameter estimators cannot be defined explicitly as a function of other statistics from the sample. It is shown how these results can be applied to linear regression, logistic regression and log linear contingency table models.

To determine and measure the impact of informativeness we compare design-based and model-based variances of estimated parameters, as well as the estimated parameters themselves, in a logistic model under the assumption that the postulated model is true. An approach for assessing the impact of informativeness is given. In order to address the additional complexity of the impact of informativeness on power, we propose a new approximation for a linear combination of non-central chi-square distributions, using generalized design effects. A large simulation study, based on generating a population under the postulated model, using parameter estimates derived from the NPHS, allows us to detect and to measure the informativeness, and to compare the robustness of studied approaches.

There are a number of asymptotically equivalent procedures for deriving the Taylor series approximation of variances for complex statistics. In Binder and Patak (1994) the theoretical justification for one class of methods was derived. However, many of these methods can be derived for practical examples using straightforward techniques that are not clearly described in Binder and Patak. In this paper we give a “cookbook” approach that can be used for many examples, and that has been shown to have good finite sample properties. Normally the method of choice becomes clear through arguments such as model-assisted methods or linearizing the jackknife; however, using our approach yields the desired results more directly. As well, we present new results on the application of these techniques to two-phase samples.

We summarize some salient aspects of the theory of estimation functions for finite populations. In particular, we discuss the problem of estimation of means and totals and extend this theory to estimating functions. We then apply this estimating functions framework to the problem of estimating measures of income inequality. The resulting statistics are nonlinear functions of the observations. Some of them depend on the order of observations or quantiles. Consequently, the mean squared errors of these estimates are inexpressible by simple formulae and cannot be estimated by conventional variance estimation methods. We show that within the estimating function framework this problem can be resolved using the Taylor linearization method. Finally, we illustrate the proposed methodology using income data from Canadian Survey of Consumer Finance and comparing it to the ‘delete-one-cluster’ jackknifing method.

The Canadian Health and Disability Survey, administered as a supplement to the Canadian Labour Force Survey in October 1983, collected data on potentially disabled persons by means of a screening questionnaire and a follow-up questionnaire for those screened-in. The data from the screening questionnaire, consisting of a set of activities of daily living, were used to group respondents according to identifiable characteristics. A description of the groups of respondents is provided along with an evaluation of the methods used in their determination. An incompletely ordered severity scale is proposed.

Univariate statistical models, linear regression models and generalized linear models are briefly reviewed. Examples of a two-way analysis of variance, a three-way analysis of variance and logistic regression for a three way layout are given.

Statistics Canada, Canada’s central statistical agency, has been compiling national mortality statistics, including those on cancer mortality since 1921. Also, cancer incidence data are available from 1969.

The data quality of these files may be assessed in a variety of ways. Ratios of cancer mortality to incidence give some information on coverage errors. Micro-data matches between incidence and mortality files give an indication of misclassifications. As well, multiple registrations for cancer incidence may be duplicates. Completeness and availability of data items are also important for special studies.

In this paper, the feasibility of using these measures of data quality and the implications of these measures are discussed.

The problem of specifying and estimating the variance of estimated parameters based on complex sample designs from finite populations is considered. The results of this paper are particularly useful when the parameter estimators cannot be defined explicitly as a function of other statistics from the sample. It is shown how these results can be applied to linear regression, logistic regression and log linear contingency table models.