Sort Help
entries

Results

All (16)

All (16) (0 to 10 of 16 results)

  • Articles and reports: 12-001-X202100200003
    Description:

    Calibration weighting is a statistically efficient way for handling unit nonresponse. Assuming the response (or output) model justifying the calibration-weight adjustment is correct, it is often possible to measure the variance of estimates in an asymptotically unbiased manner. One approach to variance estimation is to create jackknife replicate weights. Sometimes, however, the conventional method for computing jackknife replicate weights for calibrated analysis weights fails. In that case, an alternative method for computing jackknife replicate weights is usually available. That method is described here and then applied to a simple example.

    Release date: 2022-01-06

  • Articles and reports: 12-001-X201900200007
    Description:

    When fitting an ordered categorical variable with L > 2 levels to a set of covariates onto complex survey data, it is common to assume that the elements of the population fit a simple cumulative logistic regression model (proportional-odds logistic-regression model). This means the probability that the categorical variable is at or below some level is a binary logistic function of the model covariates. Moreover, except for the intercept, the values of the logistic-regression parameters are the same at each level. The conventional “design-based” method used for fitting the proportional-odds model is based on pseudo-maximum likelihood. We compare estimates computed using pseudo-maximum likelihood with those computed by assuming an alternative design-sensitive robust model-based framework. We show with a simple numerical example how estimates using the two approaches can differ. The alternative approach is easily extended to fit a general cumulative logistic model, in which the parallel-lines assumption can fail. A test of that assumption easily follows.

    Release date: 2019-06-27

  • Articles and reports: 12-001-X201800154928
    Description:

    A two-phase process was used by the Substance Abuse and Mental Health Services Administration to estimate the proportion of US adults with serious mental illness (SMI). The first phase was the annual National Survey on Drug Use and Health (NSDUH), while the second phase was a random subsample of adult respondents to the NSDUH. Respondents to the second phase of sampling were clinically evaluated for serious mental illness. A logistic prediction model was fit to this subsample with the SMI status (yes or no) determined by the second-phase instrument treated as the dependent variable and related variables collected on the NSDUH from all adults as the model’s explanatory variables. Estimates were then computed for SMI prevalence among all adults and within adult subpopulations by assigning an SMI status to each NSDUH respondent based on comparing his (her) estimated probability of having SMI to a chosen cut point on the distribution of the predicted probabilities. We investigate alternatives to this standard cut point estimator such as the probability estimator. The latter assigns an estimated probability of having SMI to each NSDUH respondent. The estimated prevalence of SMI is the weighted mean of those estimated probabilities. Using data from NSDUH and its subsample, we show that, although the probability estimator has a smaller mean squared error when estimating SMI prevalence among all adults, it has a greater tendency to be biased at the subpopulation level than the standard cut point estimator.

    Release date: 2018-06-21

  • Articles and reports: 12-001-X201700254872
    Description:

    This note discusses the theoretical foundations for the extension of the Wilson two-sided coverage interval to an estimated proportion computed from complex survey data. The interval is shown to be asymptotically equivalent to an interval derived from a logistic transformation. A mildly better version is discussed, but users may prefer constructing a one-sided interval already in the literature.

    Release date: 2017-12-21

  • Articles and reports: 12-001-X201500114172
    Description:

    When a random sample drawn from a complete list frame suffers from unit nonresponse, calibration weighting to population totals can be used to remove nonresponse bias under either an assumed response (selection) or an assumed prediction (outcome) model. Calibration weighting in this way can not only provide double protection against nonresponse bias, it can also decrease variance. By employing a simple trick one can estimate the variance under the assumed prediction model and the mean squared error under the combination of an assumed response model and the probability-sampling mechanism simultaneously. Unfortunately, there is a practical limitation on what response model can be assumed when design weights are calibrated to population totals in a single step. In particular, the choice for the response function cannot always be logistic. That limitation does not hinder calibration weighting when performed in two steps: from the respondent sample to the full sample to remove the response bias and then from the full sample to the population to decrease variance. There are potential efficiency advantages from using the two-step approach as well even when the calibration variables employed in each step is a subset of the calibration variables in the single step. Simultaneous mean-squared-error estimation using linearization is possible, but more complicated than when calibrating in a single step.

    Release date: 2015-06-29

  • Articles and reports: 12-001-X201200111689
    Description:

    When there is unit (whole-element) nonresponse in a survey sample drawn using probability-sampling principles, a common practice is to divide the sample into mutually exclusive groups in such a way that it is reasonable to assume that each sampled element in a group were equally likely to be a survey nonrespondent. In this way, unit response can be treated as an additional phase of probability sampling with the inverse of the estimated probability of unit response within a group serving as an adjustment factor when computing the final weights for the group's respondents. If the goal is to estimate the population mean of a survey variable that roughly behaves as if it were a random variable with a constant mean within each group regardless of the original design weights, then incorporating the design weights into the adjustment factors will usually be more efficient than not incorporating them. In fact, if the survey variable behaved exactly like such a random variable, then the estimated population mean computed with the design-weighted adjustment factors would be nearly unbiased in some sense (i.e., under the combination of the original probability-sampling mechanism and a prediction model) even when the sampled elements within a group are not equally likely to respond.

    Release date: 2012-06-27

  • Articles and reports: 11-536-X200900110813
    Description:

    The National Agricultural Statistics Service (NASS) has increasingly been using a delete-a-group (DAG) jackknife to estimate variances. In surveys where this technique is used, each sampled element is given 16 weights: the element's actual sampling weight after incorporating all nonresponse and calibration adjustments and 15 jackknife replicate weights. NASS recommends constructing confidence intervals for univariate statistics assuming its DAG jackknife has 14 degrees of freedom. This paper discusses methods of modifying the DAG jackknife to reduce the potential finite-sample bias. It also describes a method of measuring the effective degrees of freedom in situations where the NASS recommendation of 14 may be too generous.

    Release date: 2009-08-11

  • Articles and reports: 12-001-X20060029547
    Description:

    Calibration weighting can be used to adjust for unit nonresponse and/or coverage errors under appropriate quasi-randomization models. Alternative calibration adjustments that are asymptotically identical in a purely sampling context can diverge when used in this manner. Introducing instrumental variables into calibration weighting makes it possible for nonresponse (say) to be a function of a set of characteristics other than those in the calibration vector. When the calibration adjustment has a nonlinear form, a variant of the jackknife can remove the need for iteration in variance estimation.

    Release date: 2006-12-21

  • Articles and reports: 11-522-X20030017601
    Description:

    This paper proposes the use of sequential-interval-Poisson (SIP) sampling across many surveys in order to limit the number of times an individual farm is enumerated for a National Agricultural Statistics Service (NASS) survey.

    Release date: 2005-01-26

  • Articles and reports: 11-522-X20020016714
    Description:

    In this highly technical paper, we illustrate the application of the delete-a-group jack-knife variance estimator approach to a particular complex multi-wave longitudinal study, demonstrating its utility for linear regression and other analytic models. The delete-a-group jack-knife variance estimator is proving a very useful tool for measuring variances under complex sampling designs. This technique divides the first-phase sample into mutually exclusive and nearly equal variance groups, deletes one group at a time to create a set of replicates and makes analogous weighting adjustments in each replicate to those done for the sample as a whole. Variance estimation proceeds in the standard (unstratified) jack-knife fashion.

    Our application is to the Chicago Health and Aging Project (CHAP), a community-based longitudinal study examining risk factors for chronic health problems of older adults. A major aim of the study is the investigation of risk factors for incident Alzheimer's disease. The current design of CHAP has two components: (1) Every three years, all surviving members of the cohort are interviewed on a variety of health-related topics. These interviews include cognitive and physical function measures. (2) At each of these waves of data collection, a stratified Poisson sample is drawn from among the respondents to the full population interview for detailed clinical evaluation and neuropsychological testing. To investigate risk factors for incident disease, a 'disease-free' cohort is identified at the preceding time point and forms one major stratum in the sampling frame.

    We provide proofs of the theoretical applicability of the delete-a-group jack-knife for particular estimators under this Poisson design, paying needed attention to the distinction between finite-population and infinite-population (model) inference. In addition, we examine the issue of determining the 'right number' of variance groups.

    Release date: 2004-09-13
Stats in brief (0)

Stats in brief (0) (0 results)

No content available at this time.

Articles and reports (16)

Articles and reports (16) (0 to 10 of 16 results)

  • Articles and reports: 12-001-X202100200003
    Description:

    Calibration weighting is a statistically efficient way for handling unit nonresponse. Assuming the response (or output) model justifying the calibration-weight adjustment is correct, it is often possible to measure the variance of estimates in an asymptotically unbiased manner. One approach to variance estimation is to create jackknife replicate weights. Sometimes, however, the conventional method for computing jackknife replicate weights for calibrated analysis weights fails. In that case, an alternative method for computing jackknife replicate weights is usually available. That method is described here and then applied to a simple example.

    Release date: 2022-01-06

  • Articles and reports: 12-001-X201900200007
    Description:

    When fitting an ordered categorical variable with L > 2 levels to a set of covariates onto complex survey data, it is common to assume that the elements of the population fit a simple cumulative logistic regression model (proportional-odds logistic-regression model). This means the probability that the categorical variable is at or below some level is a binary logistic function of the model covariates. Moreover, except for the intercept, the values of the logistic-regression parameters are the same at each level. The conventional “design-based” method used for fitting the proportional-odds model is based on pseudo-maximum likelihood. We compare estimates computed using pseudo-maximum likelihood with those computed by assuming an alternative design-sensitive robust model-based framework. We show with a simple numerical example how estimates using the two approaches can differ. The alternative approach is easily extended to fit a general cumulative logistic model, in which the parallel-lines assumption can fail. A test of that assumption easily follows.

    Release date: 2019-06-27

  • Articles and reports: 12-001-X201800154928
    Description:

    A two-phase process was used by the Substance Abuse and Mental Health Services Administration to estimate the proportion of US adults with serious mental illness (SMI). The first phase was the annual National Survey on Drug Use and Health (NSDUH), while the second phase was a random subsample of adult respondents to the NSDUH. Respondents to the second phase of sampling were clinically evaluated for serious mental illness. A logistic prediction model was fit to this subsample with the SMI status (yes or no) determined by the second-phase instrument treated as the dependent variable and related variables collected on the NSDUH from all adults as the model’s explanatory variables. Estimates were then computed for SMI prevalence among all adults and within adult subpopulations by assigning an SMI status to each NSDUH respondent based on comparing his (her) estimated probability of having SMI to a chosen cut point on the distribution of the predicted probabilities. We investigate alternatives to this standard cut point estimator such as the probability estimator. The latter assigns an estimated probability of having SMI to each NSDUH respondent. The estimated prevalence of SMI is the weighted mean of those estimated probabilities. Using data from NSDUH and its subsample, we show that, although the probability estimator has a smaller mean squared error when estimating SMI prevalence among all adults, it has a greater tendency to be biased at the subpopulation level than the standard cut point estimator.

    Release date: 2018-06-21

  • Articles and reports: 12-001-X201700254872
    Description:

    This note discusses the theoretical foundations for the extension of the Wilson two-sided coverage interval to an estimated proportion computed from complex survey data. The interval is shown to be asymptotically equivalent to an interval derived from a logistic transformation. A mildly better version is discussed, but users may prefer constructing a one-sided interval already in the literature.

    Release date: 2017-12-21

  • Articles and reports: 12-001-X201500114172
    Description:

    When a random sample drawn from a complete list frame suffers from unit nonresponse, calibration weighting to population totals can be used to remove nonresponse bias under either an assumed response (selection) or an assumed prediction (outcome) model. Calibration weighting in this way can not only provide double protection against nonresponse bias, it can also decrease variance. By employing a simple trick one can estimate the variance under the assumed prediction model and the mean squared error under the combination of an assumed response model and the probability-sampling mechanism simultaneously. Unfortunately, there is a practical limitation on what response model can be assumed when design weights are calibrated to population totals in a single step. In particular, the choice for the response function cannot always be logistic. That limitation does not hinder calibration weighting when performed in two steps: from the respondent sample to the full sample to remove the response bias and then from the full sample to the population to decrease variance. There are potential efficiency advantages from using the two-step approach as well even when the calibration variables employed in each step is a subset of the calibration variables in the single step. Simultaneous mean-squared-error estimation using linearization is possible, but more complicated than when calibrating in a single step.

    Release date: 2015-06-29

  • Articles and reports: 12-001-X201200111689
    Description:

    When there is unit (whole-element) nonresponse in a survey sample drawn using probability-sampling principles, a common practice is to divide the sample into mutually exclusive groups in such a way that it is reasonable to assume that each sampled element in a group were equally likely to be a survey nonrespondent. In this way, unit response can be treated as an additional phase of probability sampling with the inverse of the estimated probability of unit response within a group serving as an adjustment factor when computing the final weights for the group's respondents. If the goal is to estimate the population mean of a survey variable that roughly behaves as if it were a random variable with a constant mean within each group regardless of the original design weights, then incorporating the design weights into the adjustment factors will usually be more efficient than not incorporating them. In fact, if the survey variable behaved exactly like such a random variable, then the estimated population mean computed with the design-weighted adjustment factors would be nearly unbiased in some sense (i.e., under the combination of the original probability-sampling mechanism and a prediction model) even when the sampled elements within a group are not equally likely to respond.

    Release date: 2012-06-27

  • Articles and reports: 11-536-X200900110813
    Description:

    The National Agricultural Statistics Service (NASS) has increasingly been using a delete-a-group (DAG) jackknife to estimate variances. In surveys where this technique is used, each sampled element is given 16 weights: the element's actual sampling weight after incorporating all nonresponse and calibration adjustments and 15 jackknife replicate weights. NASS recommends constructing confidence intervals for univariate statistics assuming its DAG jackknife has 14 degrees of freedom. This paper discusses methods of modifying the DAG jackknife to reduce the potential finite-sample bias. It also describes a method of measuring the effective degrees of freedom in situations where the NASS recommendation of 14 may be too generous.

    Release date: 2009-08-11

  • Articles and reports: 12-001-X20060029547
    Description:

    Calibration weighting can be used to adjust for unit nonresponse and/or coverage errors under appropriate quasi-randomization models. Alternative calibration adjustments that are asymptotically identical in a purely sampling context can diverge when used in this manner. Introducing instrumental variables into calibration weighting makes it possible for nonresponse (say) to be a function of a set of characteristics other than those in the calibration vector. When the calibration adjustment has a nonlinear form, a variant of the jackknife can remove the need for iteration in variance estimation.

    Release date: 2006-12-21

  • Articles and reports: 11-522-X20030017601
    Description:

    This paper proposes the use of sequential-interval-Poisson (SIP) sampling across many surveys in order to limit the number of times an individual farm is enumerated for a National Agricultural Statistics Service (NASS) survey.

    Release date: 2005-01-26

  • Articles and reports: 11-522-X20020016714
    Description:

    In this highly technical paper, we illustrate the application of the delete-a-group jack-knife variance estimator approach to a particular complex multi-wave longitudinal study, demonstrating its utility for linear regression and other analytic models. The delete-a-group jack-knife variance estimator is proving a very useful tool for measuring variances under complex sampling designs. This technique divides the first-phase sample into mutually exclusive and nearly equal variance groups, deletes one group at a time to create a set of replicates and makes analogous weighting adjustments in each replicate to those done for the sample as a whole. Variance estimation proceeds in the standard (unstratified) jack-knife fashion.

    Our application is to the Chicago Health and Aging Project (CHAP), a community-based longitudinal study examining risk factors for chronic health problems of older adults. A major aim of the study is the investigation of risk factors for incident Alzheimer's disease. The current design of CHAP has two components: (1) Every three years, all surviving members of the cohort are interviewed on a variety of health-related topics. These interviews include cognitive and physical function measures. (2) At each of these waves of data collection, a stratified Poisson sample is drawn from among the respondents to the full population interview for detailed clinical evaluation and neuropsychological testing. To investigate risk factors for incident disease, a 'disease-free' cohort is identified at the preceding time point and forms one major stratum in the sampling frame.

    We provide proofs of the theoretical applicability of the delete-a-group jack-knife for particular estimators under this Poisson design, paying needed attention to the distinction between finite-population and infinite-population (model) inference. In addition, we examine the issue of determining the 'right number' of variance groups.

    Release date: 2004-09-13
Journals and periodicals (0)

Journals and periodicals (0) (0 results)

No content available at this time.

Date modified: