Table 3:

Comparisons of properties among different HRF modeling approaches.

Individual level
	Canonical HRF	Sampled HRF	Smooth HRF
shape assumption:	(a priori) fixed	adaptive, data-driven
basis:	Gamma variates	deconvolution with piecewise linear splines (e.g., TENT, FIR)
dimension:	scalar (0D)	vector (1D) of equally spaced samples
trial-level effects:	possible	difficult
Population level
	Canonical HRF	Sampled HRF	Smooth HRF
dimension:	scalar (0D)	vector (1D) of equally spaced samples	smooth curve (1D)
interpretability:	high: signed peak magnitude	moderate: detailed shape features, but relies on visual examination
robustness:	low: inflexible and vulnerable to under-fitting	moderate: adaptive but vulnerable to over-fitting	high: adaptive and regularized