We examine differences of empirical site characteristics among S waves, P waves, coda, and microtremors using records at 20 sites in and around the Sendai basin, Japan, and interpret the differences theoretically. At soft soil sites the horizontal-to-vertical spectral ratios (HVRs) for early P coda become different from HVRs for a P wave with increasing time and eventually converge on HVRs for microtremors. The HVRs for an S coda become similar to HVRs for microtremors with increasing time in the frequency range lower than 3 Hz at soft soil sites. By contrast, at a rock site and two hard soil sites, HVRs for S coda agree well with HVRs for an S wave. The soil-to-rock spectral ratios for horizontal (HHRs) and vertical (VVRs) components for early S coda are larger than those for an S wave at soft soil sites. When we use the deep sedimentary structures above the bedrock before Tertiary age, theoretical HVR for the fundamental mode of Rayleigh waves is consistent with observed HVR for microtremors and theoretical HVR for an obliquely incident SV wave is consistent with observed HVR for an S wave. Theoretical S-wave site amplification factor explains well observed HHR for S wave but does not coincide with HVR for microtremors. In general the frequencies of maximum peaks of HVRs for microtremors do not coincide with those of HVRs and HHRs for S wave. However, if we select HVRs with peak frequencies lower than 1 Hz and peak amplitudes larger than three, the peak frequencies of HVRs for microtremors roughly coincide with those of HVRs and HHRs for S wave. Even under these constraints, their amplitudes do not coincide with each other. Concerning coda, we conclude that the Rayleigh wave contamination in coda is significant in the frequency range lower than 3 Hz at soft soil sites.