HST583 Fall 2006 1. Consider a design with: 87 time points, TR = 2.0 sec, 3 conditions. a. How many rows and columns would the design matrix have for a design with using a fixed impulse response (FIR) model with a time window of 26 sec for each condition? b. How many rows and columns would it have if a gamma models were used for each condition instead? 2. What are the three main sources of noise in fMRI? 3. What is the difference between the residual noise variance and the variance of the regression coefficients? How are each of these affected by unmodeled temporal correlation? 4. Respiration can create fMRI noise due to small changes in the B0 field. a. How would you incorporate this effect in the design matrix? b. What effect will this have on the estimate of the task-related regression coefficients and their variance? 5. Consider a design in which subjects are presented with four types of face stimuli in which the emotion and gender are varied: (1) happy-male, (2) happy-female, (3) neutral-male, (4) neutral-female. The design matrix is set up with a regressor for each of the conditions using a gamma model. Give the contrast matrices to test: a. The difference between the response to male and female faces regardless of emotional content. b. The difference between the response to happy and neutral faces regardless of gender. c. What would the two contrast matrices be if a nuisance regressor were added as a fifth regressor? 6. Consider a pure-noise fMRI data set for an individual with volume size 64x64x30 in which one third of the voxels are brain. a. How many brain voxels would you expect to be significant at the .01 level? b. What threshold would you use to assure that any voxel declared active is significant at the .01 level? 7. What is the difference between a random and fixed effects model in the analysis of group data.