Qualité, Lionel

2021-10-6, Chevalier, Clément, Qualité, Lionel

La pondération de l'enquête suisse sur la structure des salaires 2018 est révisée. L'objectif principal de cette révision est d'utiliser l'information sur les revenus des salariés, mise à disposition de l'Office fédéral de la statistique par les caisses de compensation. Cette information avait déjà été utilisée pour réviser le plan de sondage de l'enquête et doit permettre de réaliser des gains de précision. Ce travail a également été l'occasion de revoir toutes les étapes de la préparation des données et de la pondération.

2009, Qualité, Lionel, Tillé, Yves

Ce document est constitué de deux parties. Dans la première partie, nous nous intéressons à certains plans de sondage à probabilités inégales, et dans la deuxième partie nous étudions le problème des enquêtes répétées. Bien que les sujets développés dans ces deux parties semblent entièrement différents, ils sont en fait reliés. La première partie est principalement consacrée à l’étude des propriétés de deux plans de sondage de taille fixe. Dans un premier chapitre, il est démontré que le plan de sondage à entropie maximale et de taille fixe est plus efficace que le sondage avec remise. Dans le second chapitre, nous montrons que le sondage systématique est un plan à support minimal. Nous donnons aussi quelques résultats sur la variance de l’estimateur de Horvitz-Thompson pour les plans à entropie maximale et pour les plans à support minimal. La deuxième partie débute par une étude de cas sur l’estimation de précision des évolutions dans le panel suisse sur la valeur ajoutée. Dans le chapitre suivant, nous proposons un estimateur de covariance pour les panels rotatifs à probabilités inégales. Enfin, nous présentons un système de coordination d’échantillons poissoniens développé pour l’Office Fédéral de la Statistique Suisse., This document is divided into two parts. The first part revolves around the properties of some unequal probability survey sampling designs, and the second part deals with repeated surveys. While the topics developed in these two parts appear to be largely different, they are in fact related. The first part is devoted to the study of properties of two sampling designs with fixed size. In a first chapter we show that maximum entropy sampling with fixed size is more efficient than sampling with replacement. In a second chapter we prove that systematic sampling is a minimum support design. We also give some results on the variance of the Horvitz-Thompson estimator for maximum entropy and for minimum support designs. The second part begins with a case study of the estimation of variance of evolutions in the Swiss panel on value added. In a second chapter, we give covariance estimators for rotating panels with unequal inclusion probabilities. Finally, we describe a coordination method of maximum entropy samples that was developed for the Swiss Federal Statistical Office.

, Nedyalkova, Desislava, Tillé, Yves, Qualité, Lionel

Coordination of probabilistic samples is a challenging theoretical problem faced by statistical institutes. One of their aims is to obtain good estimates for each wave while spreading the response burden across the entire population. There is a collection of existing solutions that try to attend to these needs. These solutions, which were developed independently, are integrated in a general framework and their corresponding longitudinal designs are computed. The properties of these longitudinal designs are discussed. It is also noted that there is an antagonism between a good rotation and control over the cross-sectional sampling design. A compromise needs to be reached between the quality of the sample coordination, which appears to be optimal for a systematic longitudinal sampling design, and the freedom of choice of the cross-sectional design. In order to reach such a compromise, an algorithm that uses a new method of longitudinal sampling is proposed

2021-6-15, Qualité, Lionel

The Swiss Federal Statistical Office undertook to revise the sampling design of its biennial Earnings Structure Survey (ESS) for 2018. The new design uses administrative data on incomes, collected by the Swiss Compensation Office (SCO) for social insurance purposes, as proxy variables. The ESS provides information on wages and salaries paid by businesses in relation to the jobs and individual characteristics of employees. Its main products are median standardized wages for a range of population or business domains and total rewards within activity sections. The sampling design revision aimed at improving efficiency by using the SCO incomes and adapting the allocation procedure to the survey objectives: multiple medians and totals. The sampling design used for previous surveys targeted best precision on the overall average earnings under a cost constraint. A linearization technique allows replacing the estimation variance of a median with that of an estimator of total in the allocation problem. We compared results of different linearization procedures using simulations on the SCO incomes matched with the business register. The multi-objective nature of the problem required to scale up the allocation procedure for the case of several hundred interest variables. This is done in two steps: limit values for the coefficient of variation of each interest estimator in the ESS are computed taking into account the maximum achievable precision under full census and a non-response scenario. These limits are set to the publication thresholds of 3% or 5% when possible. Then, a new allocation procedure of ‘sample sizes within strata’, adapted to the ESS sampling design, aims for the minimum overall sample size under constraint that no estimator coefficient of variation exceeds its assigned limit value. The result is a sample size reduced by 10% (approx. 5’000 businesses) for the ESS2018 over the ESS2016 sample and a better control on precisions of the most relevant estimates.

, Qualité, Lionel, Tillé, Yves

We propose a method for estimating the variance of estimators of changes over time, a method that takes account of all the components of these estimators: the sampling design, treatment of non-response, treatment of large companies, correlation of non- response from one wave to another, the effect of using a panel, robustification, and calibration using a ratio estimator. This method, which serves to determine the confidence intervals of changes over time, is then applied to the Swiss survey of value added.

2019-3-1, Qualité, Lionel

Starting in 2009, the Swiss Federal Statistical Office has been using a coordinated sampling procedure for its business surveys - and for its population and household surveys since 2010. The method, developed in Qualité (2009), is an extension of Brewer et al.’s (1972) procedure. It is based on the use of permanent random numbers and allows selecting, at each sampling occasion, a Poisson sample with chosen positive or negative coordination with respect to all past surveys. A priority needs to be assigned to these coordination requirements, as they may be inconsistent. By the end of 2018, 35 business surveys composed of 95 sub-samples were successfully selected using this system. During this period, the sampling frame was updated twice-yearly using extracts of our business register. The pragmatic solution that we adopted consisted in transmitting history of ancient units to new units based on events recorded by register managers. However, an automatic treatment that matched completely the typology of demographic events of Eurostat (2010) was not possible: some information was missing in our records (e.g. in case of a Takeover, which business takes over which), and arbitration is needed for “n-to-m” events. A challenging issue that arose with our population and household surveys is that of the sustainability of the coordinated sampling system. Indeed, the amount of necessary computations and stored data increase with the number of selected samples. While our tests showed that the business sampling system is able to accommodate hundreds more selections, a solution eventually has to be found for the very long run. References: Brewer, K., Early, L., and Joyce, S. (1972). Selecting several samples from a single population. Australian Journal of Statistics, 14 (3), 231-239. Qualité, L. (2009). Unequal probability sampling and repeated surveys. PhD thesis, University of Neuchâtel.

, Qualité, Lionel, Tillé, Yves

Nous proposons une méthode pour estimer la variance des estimateurs des évolutions qui prend en compte toutes les composantes de ceux-ci : le plan de sondage, le traitement des non-réponses, le traitement des grosses entreprises, la corrélation de la non-réponse d’une vague à l’autre, l’effet dû à l’utilisation d’un panel, la robustification et le calage au moyen d’un estimateur par le ratio. Cette méthode, qui permet la détermination d’intervalles de confiance des évolutions, est ensuite appliquée à l’enquête suisse sur la valeur ajoutée.