A Primer on Maximum Likelihood Algorithms Available for Use With Missing Data

By Craig K. Enders. 2001

1,066 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Environmental Engineering, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/5f597d88bef107531c72fb04b0ef7f2f4ed2ea6e

How researchers see this paper

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

In Artificial Intelligence

In Environmental Engineering

In Statistics and Probability

Citations in the corpus

A Latent Growth Curve Analysis of Prosocial Behavior Among Rural Adolescents (2007, methodology): “Recent studies (e.g., Collins, Schafer, & Kam, 2001; Enders, 2001; Schafer & Graham, 2002) have shown that this method has desirable sta tistical properties (e.g., low estimator bias) even with mild departures from multivariate normality and data that is assumed to be missing at random.”
MaLaDaPTIve scheMas anD DePressIon: TesTs of sTress generaTIon anD DIaThesIs-sTress MoDeLs (2011, methodology): “Maximum likelihood estimation was used to model missing data (e.g., Enders, 2001; Schafer & Graham, 2002).”
Development of Life Satisfaction in Old Age: Another View on the "Paradox'' (2006, methodology): “The development of FIML, as well as other maximum likelihood algorithms for use with missing data (Enders, 2001a), was inspired by the typology for the probabilities of missingness presented by Rubin (1976), who showed that maximum likelihood based missing data procedures could yield unbiased…”
Missing data techniques for multilevel data: implications of model misspecification (2011, background): “The maximum point on the function is the ML, and the set of parameter estimates corresponding to that point represents the estimates of population parameters that are most likely to have produced the observed data (for details about MLE, see [4,5,15,16,37]).”
Why do people prefer gratitude journaling over gratitude letters? The influence of individual differences in motivation and personality on web-based interventions (2015, methodology): “61, and were imputed using Expectation–Maximization algorithm (Enders, 2001) in SPSS 21.”
Exploring gains in reading and mathematics achievement among regular and exceptional students using (2013, methodology): “Under MAR, ML-based procedures (full information ML, the expectation-maximization [EM] algorithm) can effectively handle missing data in the respect of providing unbiased estimates (Enders, 2001; Graham, 2003; Shin, Davison, & Long, 2009).”
The Role of Moral Disengagement and Self-Efficacy in Cyberbullying (2015, methodology): “The EM procedure has been shown to be superior to listwise deletion, pairwise deletion, and means substitution (Allison, 2002; Enders, 2001; Schafer & Graham, 2002).”
Construct Validity of the Multidimensional Structure of Bullying and Victimization: An Application of Exploratory Structural Equation Modeling. (2011, methodology): “We relied on the full-information MLR estimator (see Enders, 2001; Graham, 2009) to account for missing data.”
Effects of Missing Data Methods in Structural Equation Modeling With Nonnormal Longitudinal Data (2009, methodology): “…methods based on LD or PD cannot in general be recommended due to loss in efficiency as well as estimator bias and distortion of variable interrelations (Arbuckle, 1996; T. E. Duncan, Duncan, & Li, 1998; Enders, 2001b; Muthén, Kaplan, & Hollis, 1987; Schafer & Graham, 2002; Yuan & Bentler, 2000).”
How Do Missing Data Bias Estimates of Within-Group Agreement? Sensitivity of SD WG, CVWG, rWG(J), rWG(J) * , and ICC to Systematic Nonresponse (2009, methodology): “We recommend that, in practice, item nonresponse be addressed through meanitem imputation (Roth, Switzer, & Switzer, 1999), and scale nonresponse be addressed through ML or MI techniques (Enders, 2001; Schafer, 1997).”
(Why) Should We Use SEM? Pros and Cons of Structural Equation Modeling (2003, methodology): “The SEM framework has been enriched with the newest procedures to handle missing data (Enders, 2001; Enders & Bandalos, 2001) and was extended to moderator models.”
Analysis of Longitudinal Studies With Missing Data Using Covariance Structure Modeling With Full-Information Maximum Likelihood (2005, background): “…qualities of this modeling approach when its assumptions are violated, and particularly that of data missing at random, some recent investigations point to a considerable degree of robustness against up to mild deviations from them (e.g., Collins et al., 2001; Enders, 2001; Schafer & Graham, 2002).”
Comparison of Methods of Handling Missing Time-Invariant Covariates in Latent Growth Models Under the Assumption of Missing Completely at Random (2007, methodology): “The estimated mean and covariance matrix are then used as input in SEM (e.g., Enders, 2001).”
The Performance of the Full Information Maximum Likelihood Estimator in Multiple Regression Models with Missing Data (2001, background): “A brief description of the FIML estimator follows; the interested reader is encouraged to consult Enders (2001) for a more detailed overview of ML missing data estimators.”
Maximum Likelihood Multiple Imputation: Faster Imputations and Consistent Standard Errors Without Posterior Draws (2012, positive): “To get ML estimates, software can use full information maximum likelihood or the EM algorithm [10].”
Two thirds of the age-based changes in fluid and crystallized intelligence, perceptual speed, and memory in adulthood are shared (2012, positive): “Introducing such informative covariates in a regression model (hence, by extension also in a multilevel model) using FIML estimation greatly reduces the bias in parameter estimates, because all estimated effects are conditioned on the information that is associated to the reasons of dropout (Enders,”