The Effect of Small Sample Size on Two-Level Model Estimates: A Review and Illustration

By Daniel M. McNeish, Laura M. Stapleton. Educational Psychology Review, 2014

671 citations recorded by the source corpus.

Subfields: Economics and Econometrics, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/2de8700c96cabee4481f4c1222162086e25a3198

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Economics and Econometrics

In Statistics and Probability

Citations in the corpus

On Using Bayesian Methods to Address Small Sample Problems (2016, background): “As is widely known, with small samples, the variance components and fixed effect standard error estimates are highly downwardly biased with ML (Browne & Draper, 2006; McNeish & Stapleton, 2014).”
The Effects of Providing and Receiving Peer Feedback on Writing Performance and Learning of Secondary School Students (2020, background): “Also note that since there were only two schools, nesting students within schools in a hierarchical linear model also conveys no advantages for modeling the data (McNeish & Stapleton, 2016).”
Multilevel Mediation With Small Samples: A Cautionary Note on the Multilevel Structural Equation Modeling Framework (2017, background): “…multilevel data tend to be greatly affected when the data either have few clusters (e.g., Bell, Morgan, Schoeneberger, Kromrey, & Ferron, 2014; Browne & Draper, 2006; Hox et al., 2012; Maas & Hox, 2005; McNeish & Stapleton, 2016) or if there are few observations per cluster (e.g., Clarke, 2008).”
Scaffolding Student Understanding in Small-Group Work: Students’ Uptake of Teacher Support in Subsequent Small-Group Interaction (2018, methodology): “Therefore, we accounted for the nesting within teachers by including teacher dummy variables (cf. McNeish & Stapleton, 2016).”
Multilevel and Single-Level Models for Measured and Latent Variables When Data Are Clustered (2016, methodology): “Within contexts with a small number of clusters or a small sample size in general, design-based estimation may be preferable given the instability in multilevel model estimates that can be found with samples containing few clusters (McNeish & Stapleton, 2016; Schochet, 2015).”
Modeling Clustered Data with Very Few Clusters (2016, methodology): “…in Stata 14 released in April 2015), and several studies have explored the properties of the Kenward-Roger correction (e.g., Bell et al., 2014; Ferron et al., 2009; Kowalchuk, Keselman, Algina, & Wolfinger, 2004; McNeish & Stapleton, 2014; Spilke, Piepho, & Hu, 2005; Vallejo & Livacic-Rojas, 2005).”
Careless responding in questionnaire measures: Detection, impact, and remedies (2020, background): “…for clustering (i.e., TYPE=COMPLEX in Mplus or vce(cluster) in Stata) under such circumstances may have produced downwardly biased standard errors (Maas & Hox, 2005; McNeish & Stapleton, 2016), which in turn could have resulted in parameter tests in which the H 0 would have been rejected too often.”
Small Sample Size Solutions (2020, methodology): “However, if the 222 Joop Hox and Daniel McNeish interest is only in the fixed effects of the lowest-level variables, these approaches are simple and effective with as few as four clusters (McNeish & Stapleton, 2016b).”