Variance components Stefan Kiebel Wellcome Dept. of Imaging Neuroscience Institute of Neurology, UCL, London.
Download ReportTranscript Variance components Stefan Kiebel Wellcome Dept. of Imaging Neuroscience Institute of Neurology, UCL, London.
Variance components Stefan Kiebel Wellcome Dept. of Imaging Neuroscience Institute of Neurology, UCL, London Modelling in SPM functional data design matrix hypotheses smoothed normalised data pre-processing templates general linear model variance components parameter estimation SPMs adjusted P-values y X general linear model p 1 1 1 y N = X N N: number of observations p: number of regressors p + error normally distributed N model specified by 1. design matrix X 2. assumptions about Summary Sphericity/non-sphericity Restricted Maximum Likelihood (ReML) Estimation in SPM2 Summary Sphericity/non-sphericity Restricted Maximum Likelihood (ReML) Estimation in SPM2 ‚sphericity‘ C Cov( ) E ( ) T y X Scans ‚sphericity‘ means: Cov( ) I 2 i.e. Var ( ) i 2 Scans 2 1 ‚non-sphericity‘ 4 0 Cov( ) 0 1 non-sphericity means that the error covariance doesn‘t look like this*: Cov( ) I 2 1 0 Cov( ) 0 1 *: or can be brought through a linear transform to this form 2 1 Cov( ) 1 2 Example: serial correlations t a t 1 t with t ~ N (0, 2 ) autoregressive process of order 1 (AR(1)) N Cov( ) autocovariancefunction N Summary Sphericity/non-sphericity Restricted Maximum Likelihood (ReML) Estimation in SPM2 Restricted Maximum Likelihood y X Cov( ) ? observed Q1 T y y j j voxel j ReML estimated Q2 ˆ1Q1 ˆ2Q2 t-statistic (OLS estimator) c ˆ t T ˆ ˆ St d (c ) y X T ̂ X y 2V Cov( ) 2 T T ˆ Stˆd (c ) ˆ c X VX c T c = +1 0 0 0 0 0 0 0 0 0 0 X ˆ 2 y Xˆ 2 V tr ( RV ) R I XX approximate degrees of freedom following Satterthwaite ReMLestimate Variance components y X V Cov( ) 1Q1 2Q2 K QK The variance parameters are estimated by ReML. Variance components Q model the error Q1 Q1 Q2 1 and Q1 I 2 model for sphericity model for inhomogeneous variances (2 groups) Example I Stimuli: Auditory Presentation (SOA = 4 secs) of (i) words and (ii) words spoken backwards e.g. “Book” and “Koob” Subjects: Scanning: (i) 12 control subjects (ii) 11 blind subjects fMRI, 250 scans per subject, block design U. Noppeney et al. Population differences 1st level: Controls Blinds 2nd level: V cT [1 1] X Summary Sphericity/non-sphericity Restricted Maximum Likelihood (ReML) Estimation in SPM2 Estimating variances y X N 1 N p p1 EM-algorithm N 1 C | y ( X T C1 X ) 1 | y C | y X C y T maximise 1 E-step L ln p(y|λ) dL d d 2L J 2 d J 1 g g C k Qk k Assume, at voxel j: M-step jk j k K. Friston et al. 2002, Neuroimage voxelwise model specification Time parameter estimation hypothesis statistic Intensity Time series in one voxel SPM Spatial ‚Pooling‘ • • Assumptions in SPM2: global correlation matrix V local variance estimated observed ReML T y y j j Cˆ ˆ1Q1 ˆ2Q2 voxel j local in voxel j: ˆ 2 j T j j Cˆ j ˆ 2j V r r tr ( R) rj RV 1/ 2 y j , R I V 1/ 2 X (V 1/ 2 X ) Cˆ n V , ˆ trace(C ) global where V is N N Matrix Estimation in SPM2 y j X j j Cˆ Coˆv( ) ReML( T y y j j , X , Q) voxel j ReML (pooled estimate) ̂ j ,OLS X y j Ordinary least-squares •optional in SPM2 •one pass through data •statistic using (approximated) effective degrees of freedom ˆ j ,ML ( X TV 1 X ) 1 X TV 1 y j ‚quasi‘-Maximum Likelihood •2 passes (first pass for selection of voxels) •more precise estimate of V t-statistic (ML-estimate) y X W V 1/ 2 c ˆ 2 t V Cov( ) T ˆ Stˆd (c ) T ˆ 2 T T Stˆd (c ) ˆ c (WX ) (WX ) c T ̂ (WX ) Wy c = +1 0 0 0 0 0 0 0 0 0 0 X V ˆ 2 Wy WXˆ 2 tr( R) R I WX (WX ) ReMLestimate Example II Stimuli: Auditory Presentation (SOA = 4 secs) of words Motion Sound Visual Action “jump” “click” “pink” “turn” Subjects: (i) 12 control subjects Scanning: fMRI, 250 scans per subject, block design Question: What regions are affected by the semantic content of the words? U. Noppeney et al. Repeated measures Anova 1st level: Motion Sound ? Visual ? ? = = 2nd level: Action = X 1 1 0 0 cT 0 1 1 0 0 0 1 1 V X