Donald Hedeker

Overview

Donald Hedeker is Professor in the Department of Public Health Sciences at the University of Chicago (formerly at the Biometric Laboratory, University of Illinois at Chicago). His research focuses on multilevel and mixed-effects regression models for binary, ordinal, and nominal responses from clustered and longitudinal studies, with applications in psychiatric epidemiology, public health, and prevention science.

Key Contributions / Features

Hedeker & Gibbons (1994): Extended the two-level random effects probit model to ordinal responses; threshold model with $C-1$ cutpoints for $C$ -category outcomes; random intercept and/or trend; full-information MMLE via Gauss-Hermite quadrature. Biometrics 50: 933–944.
Gibbons & Hedeker (1994): Application of random-effects probit regression models to repeated binary outcomes — random intercept + random trend; medical malpractice and clinical trial applications. Journal of Consulting and Clinical Psychology 62: 285–296.
Gibbons-Hedeker-Charles-Frisch (1994): Random effects probit model for predicting medical malpractice claims — two-level REPM with physician-level random intercept; MMLE via Gauss-Hermite quadrature; applied to 30,000 physician-years. Journal of the American Statistical Association 89: 760–767.
Gibbons & Hedeker (1997): Three-level random effects probit and logistic regression — conditional independence trick reduces integration from $(n_i r+1)$ to $r+1$ dimensions; Cholesky orthogonalization for quadrature; TVSFP application with ICC classroom = 6.6%, ICC student = 25.0%; CC×TV interaction significant only in the three-level model.
Hedeker-Gibbons (1997, Psych. Methods): Pattern-mixture model for nonignorable dropout — enters dropout pattern as between-subjects covariate in mixed-effects model; nested LR test of pattern effects diagnoses MNAR; no separate dropout mechanism required.
Hedeker-Mermelstein (1998, MBR): Partial proportional odds model for ordinal outcomes — $K-1$ threshold-varying coefficients for a subset of covariates; LR χ² = 36.14 (df = 4, p < .001) for proportionality violation in NORC social attitudes data.
Gibbons & Hedeker (2000): Applications of mixed-effects models in biostatistics (Sankhyā 62: 70–103) — unified tutorial covering continuous MRM (NIMH TDCRP: ICC_therapist = 0.02, ICC_patient = 0.27; experience × treatment p < .004), binary probit (NIMH Schizophrenia: $\hat{\sigma}_{\beta_0}=0.860$ , $\hat{\sigma}_{\beta_1}=0.630$ ), multivariate probit (coal miners: 5 symptoms, 2 factors), and ordinal partial proportional odds (NORC); Gaussian factorization of $\boldsymbol{\Sigma}_\beta$ for numerical stability.
Hedeker (2003): Mixed-effects multinomial logistic regression for nominal clustered/longitudinal responses — flexible $(C-1)\times C$ contrast matrix D (reference-cell and Helmert special cases); MML via Gauss-Hermite quadrature ( $Q^r$ points); category-specific ICC $r_c = \hat{\sigma}_c^2/(\hat{\sigma}_c^2 + \pi^2/3)$ ; MHRP housing-status application (n=361, ICC ≈ 0.39–0.44; Section 8 certificates increase independent vs. community housing); MIXNO software (Journal of Statistical Software 4(5): 1–92).
Hedeker (2007) (Handbook of Multilevel Analysis, ch. 6, Springer): Unified MML/GHQ treatment of multilevel ordinal and nominal models — proportional odds, partial proportional odds (LR $\chi^2_7=52.14$ , p<.001, MHRP), complementary log-log (discrete-time survival), nominal with general contrasts; IRT bridge (Rasch = homogeneous- $a$ multilevel logistic; 2PL = heterogeneous- $a$ ; Samejima and Bock nominal IRT as special cases); heterogeneous-variance formulation for polychoric/tetrachoric correlations in twin designs; MHRP extended results (ordinal ICC=0.39; nominal $r_1=0.19$ , $r_2=0.62$ ; Helmert contrasts). See Multilevel Ordinal Regression.

Donald Hedeker

Overview

Key Contributions / Features

Related