A simulation study of sample size for multilevel logistic regression models

Moineddin, Rahim; Matheson, Flora I.; Glazier, Richard H.

doi:10.1186/1471-2288-7-34

Cited by 328 publications

(269 citation statements)

References 17 publications

Supporting

Mentioning

253

Contrasting

Unclassified

Order By: Relevance

“…For example, children in districts further from central Shanghai were more likely to be from low-income, rural migrant families. We reported findings from the clustered robust standard errors instead of those from the multilevel modeling because the small number of districts (N = 7) may produce bias and inaccurate estimates in multilevel logistic regression models [34]. We present these results using multiply imputed data in Table 3.…”

Section: Analysis Planmentioning

confidence: 99%

Child Overweight and Obesity in Shanghai, China: Contextualizing Chinese Socioeconomic and Gender Differences

et al. 2017

View full text Add to dashboard Cite

Purpose Childhood overweight and obesity is on the rise in China and in Chinese cities in particular. The aim of this study is to explore the extent of income differences in childhood overweight in Shanghai, China, and examine demographic, social, and behavioral explanations for these differences. Methods Using the 2014 Child Well-Being Study of Shanghai, China-a survey that included extensive contextual information on children and their families in China's most populous city, prevalence rates and adjusted odds ratios of child overweight and obesity at age 7 were calculated by income tercile controlling for a wide variety of sociodemographic variables. Results District aggregate income increases the odds of child overweight/obesity, but only for boys. In contrast, rural hukou status was associated with lower odds of overweight/obesity for girls. Conclusions Boys at age 7 are more likely to be overweight and obese than girls. District income further increases this likelihood for boys, while rural hukou status decreases this likelihood for girls, suggesting that preferences for boys and thinness ideals for girls may play a role in the income patterning of childhood overweight and obesity.

show abstract

Section: Analysis Planmentioning

confidence: 99%

Child Overweight and Obesity in Shanghai, China: Contextualizing Chinese Socioeconomic and Gender Differences

et al. 2017

View full text Add to dashboard Cite

show abstract

“…However in our simulation this is not the case probably due to the large sample size. In addition, Moineddin et al [21] did not vary prevalence of the outcome and ICC simultaneously, whereas we varied these for 3 rd level and 2 nd level clusters simultaneously. Though we did not find any particular pattern in the bias with respect to varying prevalence of the outcome and ICCs, the bias of the variance component increased slightly when the ICC increased for both scaling methods.…”

Section: Discussionmentioning

confidence: 99%

“…However, this study did not involve a complex survey design with weights [21]. Large bias was observed for low prevalence (10%) of the outcome when the size of the cluster was small, that is 5.…”

Section: Discussionmentioning

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

“…For comparison, we ran a simulation program similar to Moineddin [13] with CS within subject correlation structure to confirm the results. For the purpose of comparing of simulation and our formula, we assume the sample size is 200 and without any drop outs.…”

Section: Examples Of Calculationsmentioning

confidence: 99%

“…This variance consists of two components: the G-side random effects and the R-side errors. The variance is formulated as (13) This can be rewritten as (14) by putting the first term into the sandwich-like error structure of so that G becomes transformed G*. We can further simplify equation (14) by extracting a combined variance parameter so the transformed R* matrix contains only the correlation parameters ρ*.…”

Section: Sample Size and Power Calculationsmentioning

confidence: 99%

Sample size and power calculations based on generalized linear mixed models with correlated binary outcomes

Dang

Mazumdar

Houck

2008

Computer Methods and Programs in Biomedicine

View full text Add to dashboard Cite

The generalized linear mixed model (GLIMMIX) provides a powerful technique to model correlated outcomes with different types of distributions. The model can now be easily implemented with SAS PROC GLIMMIX in version 9.1. For binary outcomes, linearization methods of penalized quasilikelihood (PQL) or marginal quasi-likelihood (MQL) provide relatively accurate variance estimates for fixed effects. Using GLIMMIX based on these linearization methods, we derived formulas for power and sample size calculations for longitudinal designs with attrition over time. We found that the power and sample size estimates depend on the within-subject correlation and the size of random effects. In this article, we present tables of minimum sample sizes commonly used to test hypotheses for longitudinal studies. A simulation study was used to compare the results. We also provide a Web link to the SAS macro that we developed to compute power and sample sizes for correlated binary outcomes.

show abstract

A simulation study of sample size for multilevel logistic regression models

Cited by 328 publications

References 17 publications

Child Overweight and Obesity in Shanghai, China: Contextualizing Chinese Socioeconomic and Gender Differences

Child Overweight and Obesity in Shanghai, China: Contextualizing Chinese Socioeconomic and Gender Differences

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

Sample size and power calculations based on generalized linear mixed models with correlated binary outcomes

Contact Info

Product

Resources

About