Weighting for Unequal Selection Probabilities in Multilevel Models

Pfeffermann, Danny; Skinner, Chris J.; Holmes, David; Goldstein, Harvey; Rasbash, Jon

doi:10.1111/1467-9868.00106

Cited by 330 publications

(393 citation statements)

References 11 publications

(10 reference statements)

Supporting

Mentioning

383

Contrasting

Unclassified

Order By: Relevance

“…The estimation process was performed using the average weights per pool for simplicity, which implied that the pools should be randomly formed inside each cluster. Our results are in line with those reported by Pfeffermann et al (1998Pfeffermann et al ( , 2006 (for a normal response), by Grilli and Pratesi (2004) and Rabe-Hesketh and Skrondal (2006) (for binary outcomes in the non-grouptesting context). We found that when the sampling process is informative, weights at both levels should be included.…”

Section: Discussionsupporting

confidence: 93%

See 1 more Smart Citation

A regression model for pooled data in a two-stage survey under informative sampling with application for detecting and estimating the presence of transgenic corn

Montesinos‐López¹,

Eskridge²,

Montesinos‐López³

et al. 2016

Seed Sci. Res.

View full text Add to dashboard Cite

Group-testing regression methods are effective for estimating and classifying binary responses and can substantially reduce the number of required diagnostic tests. However, there is no appropriate methodology when the sampling process is complex and informative. In these cases, researchers often ignore stratification and weights that can severely bias the estimates of the population parameters. In this paper, we develop group-testing regression models for analysing two-stage surveys with unequal selection probabilities and informative sampling. Weights are incorporated into the likelihood function using the pseudo-likelihood approach. A simulation study demonstrates that the proposed model reduces the bias in estimation considerably compared to other methods that ignore the weights. Finally, we apply the model for estimating the presence of transgenic corn in Mexico and we give the SAS code used for the analysis.

show abstract

Section: Discussionsupporting

confidence: 93%

“…However, these 'raw' weights need to be scaled to be used under a mixed model approach to avoid significant bias in the parameter estimates (Pfeffermann et al, 1998). For this reason, some scaling methods have been proposed.…”

Section: Probability Of Selectionmentioning

confidence: 99%

A regression model for pooled data in a two-stage survey under informative sampling with application for detecting and estimating the presence of transgenic corn

Montesinos‐López¹,

Eskridge²,

Montesinos‐López³

et al. 2016

Seed Sci. Res.

View full text Add to dashboard Cite

show abstract

“…MLM on survey data is an active area of research in applied statistics. The existing literature focuses on appropriate choices for weights of level-1 units when these units are selected with unequal probabilities (Pfeffermann et al 1998;Carle 2009). Level-1 weights were not an issue in this study because rooms were not randomly selected.…”

Section: Modelingmentioning

confidence: 99%

Source contributions of lead in residential floor dust and within-home variability of dust lead loading

Lucas

Bellanger

Strat

et al. 2014

Science of The Total Environment

View full text Add to dashboard Cite

Evidence of the impact of exposure to low levels of lead on children's health is increasing.Residential floor dust is the assumed origin of lead exposure by young children. In this study, we estimate the contribution of different lead sources to household interior floor dust contamination. We also estimate the within-home variability of interior floor dust lead loadings. A multilevel model was developed based on data collected in a French survey in 2008-2009 (484 housing units, 1834 rooms). Missing data were handled by multiple imputation using chained equations. The intra-home correlation between interior floor Log dust lead loadings was approximately 0.6. Dust lead from the landing of an apartment, mostly originating outside the building, was the major contributor to interior floor dust lead.Secondary contributors included the lead-based paint on exterior railings, track-in of the exterior soil of the children's play area into the dwelling, smoking inside the home, demolition of nearby old buildings and sites of pollution in the vicinity. Interior lead-based paint contaminated interior floor dust only in old and non-renovated dwellings. To reduce interior floor dust lead levels in the general population of dwellings, common areas should be maintained, and track-in from the outside should be limited as much as possible.

show abstract

“…Simulation studies have revealed that the regression coefficients are biased for small cluster sizes [7] [8]. In the literature a cluster size greater than 20 is considered as reasonably large cluster size for 2 level data for weighted multilevel analyses [1].…”

Section: Introductionmentioning

confidence: 99%

Multilevel Modeling of Binary Outcomes with Three-Level Complex Health Survey Data

Rozi¹,

Mahmud²,

Lancaster³

et al. 2017

OJEpi

View full text Add to dashboard Cite

Complex survey designs often involve unequal selection probabilities of clusters or units within clusters. When estimating models for complex survey data, scaled weights are incorporated into the likelihood, producing a pseudo likelihood. In a 3-level weighted analysis for a binary outcome, we implemented two methods for scaling the sampling weights in the National Health Survey of Pakistan (NHSP). For NHSP with health care utilization as a binary outcome we found age, gender, household (HH) goods, urban/rural status, community development index, province and marital status as significant predictors of health care utilization (p-value < 0.05). The variance of the random intercepts using scaling method 1 is estimated as 0.0961 (standard error 0.0339) for PSU level, and 0.2726 (standard error 0.0995) for household level respectively. Both estimates are significantly different from zero (p-value < 0.05) and indicate considerable heterogeneity in health care utilization with respect to households and PSUs. The results of the NHSP data analysis showed that all three analyses, weighted (two scaling methods) and un-weighted, converged to almost identical results with few exceptions. This may have occurred because of the large number of 3rd and 2nd level clusters and relatively small ICC. We performed a simulation study to assess the effect of varying prevalence and intra-class correlation coefficients (ICCs) on bias of fixed effect parameters and variance components of a multilevel pseudo maximum likelihood (weighted) analysis. The simulation results showed that the performance of the scaled weighted estimators is satisfactory for both scaling methods. Incorporating simulation into the analysis of complex multilevel surveys allows the integrity of the results to be tested and is recommended as good practice.

show abstract

Weighting for Unequal Selection Probabilities in Multilevel Models

Cited by 330 publications

References 11 publications

A regression model for pooled data in a two-stage survey under informative sampling with application for detecting and estimating the presence of transgenic corn

A regression model for pooled data in a two-stage survey under informative sampling with application for detecting and estimating the presence of transgenic corn

Source contributions of lead in residential floor dust and within-home variability of dust lead loading

Multilevel Modeling of Binary Outcomes with Three-Level Complex Health Survey Data

Contact Info

Product

Resources

About