Figure 2.
Schematic of data models. (A) Stimulus matrix X (FIR model). The matrix dimensions are 1,020 time points × 252 parameters. The matrix is the concatenation of the stimulus convolution matrix for each of the 12 event types. The stimulus convolution matrix for a given event type consists of shifted versions of a binary sequence, where ones indicate event occurrences. There are 21 shifts, one shift for each time point in the HDR estimate. The inset (upper‐left) depicts an enlarged view of the parameters for the first two event types. (B) Nuisance matrix S (polynomial version). The matrix dimensions are 1,020 time points × 5 parameters. The matrix consists of Legendre polynomials of degrees 0 through 4. The inset (upper‐left) depicts the polynomials in a line format. (C) Convolution of stimulus matrix X 2 and time kernel k (time‐event separable model). The matrix dimensions are 1,020 time points × 12 parameters. Stimulus matrix X 2 (1,020 × 12) consists of one parameter for each of the 12 event types. The parameter for a given event type is a binary sequence, where ones indicate event occurrences. Time kernel k (21 × 1) is a voxel‐specific response timecourse estimated from the data. The inset (upper‐left) depicts an enlarged view of the parameters for the first two event types. The inset (left) depicts the time kernel in a line format.