The Use of Sample Weights in Multivariate Multilevel Models with an Application to Income Data Collected by Using a Rotating Panel Survey

Abstract

Summary Longitudinal data from labour force surveys permit the investigation of income dynamics at the individual level. However, the data often originate from surveys with a complex multistage sampling scheme. In addition, the hierarchical structure of the data that is imposed by the different stages of the sampling scheme often represents the natural grouping in the population. Motivated by how income dynamics differ between the formal and informal sectors of the Brazilian economy and the data structure of the Brazilian Labour Force Survey, we extend the probability‐weighted iterative generalized least squares estimation method. Our method is used to fit multivariate multilevel models to the Brazilian Labour Force Survey data where the covariance structure between occasions at the individual level is modelled. We conclude that there are significant income differentials and that incorporating the weights in the parameter estimation has some effect on the estimated coefficients and standard errors.

The Australian Census Longitudinal Dataset: using record linkage to create a longitudinal sample from a series of cross‐sections

“…This treatment of sequential linkage errors is similar to sequential modelling of response propensities for drop‐out in standard longitudinal surveys, where the probability of responding at time t + T is the product of a set of conditional response probabilities at each time point. See for example Veiga, Smith & Brown (), which follows the approach of Lepkowski (). This allows the most flexible use of information for modelling non‐response at each time point.…”

Fitting Multivariate Multilevel Models under Informative Sampling

“…To date, however, most literature has focused on MLMs for univariate responses, with the notable exception of Pfeffermann and Sverchkov (2019), who considered a multivariate hierarchical model, in the context of estimating multiple characteristics at small area level under nonignorable (NMAR) nonresponse. Veiga et al (2014) also extended the approach of Pfeffermann et al (1998) to multivariate MLMs in the context of longitudinal complex survey data. Multivariate responses emerge both as a means to handle longitudinal measurements on the same units, as considered by Veiga et al (2014), and in the context of educational assessment studies where students are assessed using two or more proficiency tests, with each test providing a proficiency score.…”

Latent Variable Model for Multivariate Data with Measure-specific Sample Weights and Its Application in Hospital Compare

“…[7] studied the volume related weights which are subject-specific via the hierarchical logistic models. [9] applied subject-specific weights in multivariate multilevel models to the longitudinal data. [5] gave an approach that based on the likelihood to generalize the overall score.…”