We analyze data on bone mineral density (BMD) and age for white females age 20+ in the third National Health and Nutrition Examination Survey. For the sample the age of each individual is known, but some individuals did not have their BMD measured, mainly because they did not show up in the mobile examination centers. We have data from 35 counties, the small areas. We use two types of models to analyze the data. In the ignorable nonresponse model, BMD does not depend on whether an individual responds or not. In the nonignorable nonresponse model, BMD is related to whether he/she responds. We incorporate this relationship in our model by using a Bayesian approach. We further divide these two types of models into continuous and categorical data models. Our nonignorable nonresponse models have one important feature: They are ``close' to the ignorable nonresponse model thereby reducing the effects of the untestable assumptions so common in nonresponse models. In the continuous data models, because the age of all nonrespondents are known and there is a relation between BMD and age, age is used as a covariate. In the categorical data models BMD has three levels (normal, osteopenia, osteoporosis) and age has two levels (younger than 50 years, at least 50 years). Thus, age is a supplemental margin for the $2 imes 3$ categorical table. Our research on the categorical models is much deeper than on the continuous models. Our models are hierarchical, a feature that allows a ``borrowing of strength' across the counties. Individual inference for most of the counties is unreliable because there is large variation. This ``borrowing of strength' is therefore necessary because it permits a substantial reduction in variation. The joint posterior density of the parameters for each model is complex. Thus, we fit each model using Markov chain Monte Carlo methods to obtain samples from the posterior density. These samples are used to make inference about BMD and age, and the relation between BMD and age. For the continuous data models, we show that there is an important relation between BMD and age by using a deviance measure, and we show that the nonignorable nonresponse models are to be preferred. For the categorical data models, we are able to estimate the proportion of individuals in each BMD and age cell of the categorical table, and we can assess the relation between BMD and age using the Bayes factor. A sensitivity analysis shows that there are differences, typically small, in inference that permits different levels of association between BMD and age. A simulation study shows that there is not much difference in inference between the ignorable nonresponse models and the nonignorable nonresponse models. As expected, BMD depends on age and this inference can be obtained for some small counties. For the data we use, there are virtually no young individuals with osteoporosis. The nonignorable nonresponse models generalize the ignorable nonresponse models, and therefore, allow broader inference.

Cette thèse est consacrée à l'utilisation d'information auxiliaire en théorie des sondages à l'étape de l'échantillonnage et à l'étape de l'estimation. Dans le chapitre 2, on donne une présentation des principales notions de la théorie des sondages. Au chapitre 3, on propose une extension de la famille des estimateurs par calage reposant sur l'emploi de paramètres de calage complexes. Au chapitre 4 et 5, on s'intéresse à la correction simultanée des erreurs d'échantillonnage et de non-réponse au moyen d'un calage unique. On montre qu'en dépit du fait que le calage n'utilise pas explicitement les probabilités de réponse, il est nécessaire d'écrire le modèle de réponse afin de choisir correctement la fonction de calage. A défaut, on s'expose à des estimateurs biaisés dont le biais peut dépasser le biais de l'estimateur non-ajusté. En particulier, dans le cas du calage généralisé, la variance et le biais sont amplifiés pour des variables de calage faiblement corrélées aux variables instrumentales. Au chapitre 6, on montre qu'une approche conditionnelle, par rapport au plan de sondage, permet de construire des estimateurs plus robustes aux valeurs extrêmes et aux "sauteurs de strates". Au chapitre 7, on met en évidence que la méthode du tirage réjectif de Fuller conduit un estimateur par la régression qui peut être biaisé lorsque la variable d'intérêt ne suit pas un modèle de régression linéaire en fonction des variables d'équilibrage. / This thesis is devoted to the use of auxiliary information in sampling theory at the sampling stage and estimation stage. In Chapter 2, we give an overview of the key concepts of sampling theory. In Chapter 3, we propose an extension of the family of calibration estimators based on the use of complex parameters. In Chapter 4 and 5, we are interested in the simultaneous correction of sampling errors and nonresponse using a single calibration. It shows that despite the fact that the calibration does not explicitly use the response probabilities, it is necessary to write the response model to correctly select the calibration function. Otherwise, we run the risk of biased estimators whose bias can exceed the bias of the unadjusted estimator. In particular, in the case of generalized calibration, the variance and bias are amplified for calibration variables weakly correlated with the instrumental variables. In Chapter 6, we show that a conditional approach, based on the design, leads to estimators more robust to outliers and "jumpers strata. In Chapter 7, we highlight that the Fuller rejective sampling yield to a regression estimator which can be biased when the variable of interest does not follow a linear regression with the balancing variables.

This study is concerned in reducing high dimensionality problem of auxiliary variables in the calibration estimation with the presence of nonresponse. The calibration estimation is a weighting method assists to compensate for the nonresponse in the survey analysis. Calibration estimation using principal components (PCs) is new idea in the literatures. Principal component analysis (PCA) is used in reduction dimension of the auxiliary variables. PCA in calibration estimation is presented as an alternative method for choosing the auxiliary variables. In this study, simulation on the real data is used and nonresponse mechanism is applied on the sampled data. The calibration estimator is compared using different criteria such as varying the nonresponse rate and increasing the sample size. From the results, although the calibration estimation based on the principal components have reasonable outputs to use instead of the whole auxiliary variables for the means, the variance is very large compared with based on original auxiliary variables. Finally, we identified the principal component analysis is not efficient in the reduction of high dimensionality problem of auxiliary variables in the calibration estimation for large sample sizes.

No description available.

La non-réponse, rencontrée dans la plupart des enquêtes épidémiologiques, est génératrice de biais de sélection (qui, dans ce cas est un biais de non-réponse) lorsqu’elle est liée aux variables d’intérêt. En surveillance épidémiologique, dont un des objectifs est d’estimer des prévalences, on a souvent recours à des enquêtes par sondage. On est alors confronté à la non-réponse totale et on peut utiliser des méthodes issues de la statistique d’enquête pour la corriger. Le biais de non-réponse peut être exprimé comme le produit de l’inverse du taux de réponse et de la covariance entre la probabilité de réponse et la variable d’intérêt. Ainsi, deux types de solution peuvent généralement être envisagés pour diminuer ce biais. La première consiste à chercher à augmenter le taux de réponse au moment de la planification de l’enquête. Cependant, la maximisation du taux de réponse peut entraîner d’autres types de biais, comme des biais de mesure. Dans la seconde, après avoir recueilli les données, on utilise des informations liées a priori aux variables d’intérêt et à la probabilité de réponse, et disponibles à la fois pour les répondants et les non-répondants pour calculer des facteurs correctifs. Cette solution nécessite donc de disposer d’informations sur l'ensemble de l'échantillon tiré au sort (que les personnes aient répondu ou non) ; or ces informations sont en général peu nombreuses. Les possibilités récentes d'accès aux bases médico-administratives (notamment celles de l'assurance maladie) ouvrent de nouvelles perspectives sur cet aspect.Les objectifs de ce travail, qui sont centrés sur les biais de non-réponse, étaient d’étudier l’apport de données supplémentaires (enquête complémentaire auprès de non-répondants et bases médico-administratives) et de discuter l’influence du taux de réponse sur l’erreur de non-réponse et l’erreur de mesure.L'analyse était centrée sur la surveillance épidémiologique des risques professionnels via l’exploitation des données de la phase pilote de la cohorte Coset-MSA à l’inclusion. Dans cette enquête, en plus des données recueillies par questionnaire (enquête initiale et enquête complémentaire auprès de non-répondants), des informations auxiliaires issues de bases médico-administratives (SNIIR-AM et MSA) étaient disponibles pour les répondants mais aussi pour les non-répondants à l’enquête par questionnaire.Les résultats montrent que les données de l’enquête initiale, qui présentait un taux de réponse de 24%, corrigées pour la non-réponse avec des informations auxiliaires directement liées à la thématique de l’enquête (la santé et le travail) fournissent des estimations de prévalence en général proches de celles obtenues grâce à la combinaison des données de l’enquête initiale et de l’enquête complémentaire (dont le taux de réponse atteignait 63%) après correction de la non réponse par ces mêmes informations auxiliaires. La recherche d'un taux de réponse maximal à l’aide d’une enquête complémentaire n’apparait donc pas nécessaire pour diminuer le biais de non réponse. Cette étude a néanmoins mis en avant l’existence de potentiels biais de mesure plus importants pour l’enquête initiale que pour l’enquête complémentaire. L’étude spécifique du compromis entre erreur de non-réponse et erreur de mesure montre que, pour les variables qui ont pu être étudiées, après correction de la non-réponse, la somme de l’erreur de non-réponse de l’erreur de mesure est équivalente dans l’enquête initiale et dans les enquêtes combinées (enquête initiale et complémentaire).Ce travail a montré l’intérêt des bases médico-administratives pour diminuer l’erreur de non-réponse et étudier les erreurs de mesure dans une enquête de surveillance épidémiologique. / Nonresponse occurs in most epidemiologic surveys and may generate selection bias (which is, in this case, a nonresponse bias) when it is linked to outcome variables. In epidemiologic surveillance, whose one of the purpose is to estimate prevalences, it is usual to use survey sampling. In this case, unit nonresponse occurs and it is possible to use methods coming from survey sampling to correct for nonresponse. Nonresponse bias can be expressed as the product of the inverse of the response rate and the covariance between the probability of response and the outcome variable. Thus, two options are available to reduce the effect of nonresponse. The first is to increase the response rate by developing appropriate strategies at the study design phase. However, the maximization of the response rate can prompt other kinds of bias, such as measurement bias. In the second option, after data collection, information associated with both nonresponse and the outcome variable, and available for both respondents and nonrespondents, can be used to calculate corrective factors. This solution requires having information on the complete random sample (respondents and nonrespondents); but this information is rarely sufficient. Recent possibilities to access administrative databases (particularly those pertaining to health insurance) offer new perspectives on this aspect.The objectives of this work focused on the nonresponse bias were to study the contribution of supplementary data (administrative databases and complementary survey among nonrespondents) and to discuss the influence of the response rate on the nonresponse error and the measurement error. The analyses focused on occupational health epidemiologic surveillance, using data (at inclusion) from the Coset-MSA cohort pilot study. In this study, in addition to the data collected by questionnaire (initial and complementary survey among nonrespondents), auxiliary information from health and occupational administrative databases was available for both respondents and nonrespondents.Results show that the data from the initial survey (response rate : 24%), corrected for nonresponse with information directly linked to the study subject (health and work) produce estimations of prevalence close to those obtained by combining data from the initial survey and the complementary survey (response rate : 63%), after nonresponse adjustment on the same auxiliary information. Using a complementary survey to attain a maximal response rate does not seem to be necessary in order to decrease nonresponse bias. Nevertheless, this study highlights potential measurement bias which could be more consequential for the initial survey than for the complementary survey. The specific study of the trade-off between nonresponse error and measurement error shows that, for the studied variables and after correction for nonresponse, the sum of the nonresponse error and the measurement error is equivalent in the initial survey and in the combined surveys (initial plus complementary survey). This work illustrated the potential of administrative databases for decreasing the nonresponse error and for evaluating measurement error in an epidemiologic surveillance survey.

Survey research is a popular methodology used to gather data on a myriad of phenomena. Self-report victim surveys administered by the Federal government are used to substantially broaden our understanding of the nature and extent of crime. A potential source of nonsampling error, respondent fatigue is thought to manifest in contemporary victim surveys, as respondents become "test wise" after repeated exposure to survey instruments. Using a special longitudinal data file, the presence and influence of respondent fatigue in national self-report victim surveys is examined from three perspectives. Collectively, results provide a comprehensive look at how respondent fatigue may impact crime estimates produced by national self-report victim surveys.

Disertacijoje nagrinemos problemos, iškylancios ieškant geriausios mažų sričių vertinimo strategijos. Ieškant geriausios mažų sričių vertinimo strategijos susiduriama su modelio parinkimo, imties plano ir ivertinio konstravimo, neatsakymu vertinimo ir papildomos informacijos panaudojimo problemomis. / In the dissertation special problems that may be encountered in finding optimal estimation strategy for small area estimation, in particular, model diagnostics for small area models, constrained estimation, sample design selection, nonresponse adjustment and borrowing strength across both small areas and time are considered.

Thesis (Ph.D.)--Ohio University, August, 2006. / Title from PDF t.p. Includes bibliographical references.

No description available.

Indiana University-Purdue University Indianapolis (IUPUI) / The emergence of the global SARS-CoV-2 pandemic created tremendous impact on humanity beginning in late 2019. Public health researchers at Indiana University Richard M. Fairbanks School of Public Health responded by conducting research into the etiological profile of the virus, including a large Indiana state-wide population-based prevalence study in early 2020. Methods Data on demographics, tobacco use, health status, and reasons for participating in the population prevalence study were used to conduct three retrospective cross-sectional studies. The first study assessed the association of self-reported health and tobacco behaviors with COVID-19 infection (n=8,241). The second study used successive wave analysis to assess nonresponse bias (n=3,658). Finally, participants demographics were characterized by who responded to text, email, phone calls, or postcards and by the number of prompts needed to elicit participation (n= 3,658). Results The first study found self-identified health status of those reporting “poor, “fair” or good” had a higher risk of past or current infections compared to “very good” or “excellent” health status (P <0.02). Positive smoking status was inversely associated with SARS-CoV-2 infection (p <0.001). When assessing the sample for non-response bias (n=3,658), 40.9% responded in wave 1 of recruitment, 34.1% in wave 2 and 25.0% in wave 3 for an overall participation rate of 23.6%. There were no significant differences in response by waves and demographics, being recently exposed or reasons for participating. In the final study, compared to males, females made up 54.6% of the sample and responded at a higher rate to postcards (8.2% vs. 7.5%) and text/emails (28.1 vs. 24.6%, 2= 7.43, p 0.025); and responded at a higher percentage after 1 contact (21.4 vs. 17.9%, 2 = 7.6, p 0.023). Conclusion This research contributed to the scientific understanding of the etiological picture of SARS-CoV-2. Additionally, the current study used a novel method that public health practitioners can easily implement to detect non-response bias in primary data collection without advanced statistical methods. Finally, the current study allows researchers to focus not only on the modality of inviting participants, but the frequency of invitations needed to secure specific populations, reducing time and resources.