How Many Imputations are Really Needed? Some Practical Clarifications of Multiple Imputation Theory

By J. Graham, Allison E. Olchowski, T. Gilreath. Prevention Science, 2007

2,746 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/2a030bdf04e7e0e51dc06b487095f506d0088fee

How researchers see this paper

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

In Statistics and Probability

Citations in the corpus

Multiple imputation using chained equations for missing data in TIMSS: a case study (2013, neutral): “The largest fraction of missing information (FMI), also called the rate of missing information (Graham et al., 2007; Schafer, 2001; Schafer & Olsen, 1998), reports the largest of all the FMI on coefficient estimates due to missingness.”
How Many Imputations Do You Need? A Two-stage Calculation Using a Quadratic Rule (2016, background): “…then you may need more imputations—as many as M > 200 in some situations—though fewer imputations may still suffice in many settings (Bodner, 2008; Graham, Olchowski, & Gilreath, 2007; White, Royston, & Wood, 2011), It might seem conservative to set M = 200 (say) as a blanket policy—and that…”
Quest for significance and violent extremism : the case of domestic radicalization (2017, background): “According to Graham, Olchowski, and Gilreath (2007), the optimal number of created datasets depends on the fraction of missing information.”
Modeling dependent effect sizes with three-level meta-analyses: a structural equation modeling approach. (2014, background): “However, Graham, Olchowski, and Gilreath (2007) showed that many more imputations are required.”
Rebutting Existing Misconceptions About Multiple Imputation as a Method for Handling Missing Data (2019, background): “Normally this number should be larger (Graham, Olchowski, & Gilreath, 2007), but to keep the output small it was set to five here.”
Multiple Imputation by Chained Equations: What is it and how does it work? (2011, background): “Initial research indicated that 5–10 imputed datasets was sufficient; however, recent research suggests that, depending upon the amount of missing information in the data, increasing that to as many as 40 imputed datasets can improve power (Graham et al., 2007).”
Child engagement in the transition to school: Contributions of self-regulation, teacher–child relationships and classroom climate (2015, positive): “Based on recommendations by Graham, Olchowski, and Gilreath (2007), we created 20 imputed data sets.”
Using plausible values in secondary analysis in large-scale assessments (2017, background): “Graham et al. (2007) noted that this might be too few if we are interested in other estimates and recommended the use of 20 imputations for 10–30% missing information when examining loss of power in hypothesis testing.”
New Confidence Intervals and Bias Comparisons Show That Maximum Likelihood Can Beat Multiple Imputation in Small Samples (2013, neutral): “If the fraction of missing information is large, standard errors estimated by MI can require 200-300 imputed datasets to approach the efficiency of standard errors estimated by ML (Bodner, 2008; Graham et al., 2007).”