In order to obtain an accurate and quantitative positron emission tomography (PET) image, emission data need to be corrected for random coincidences, photon attenuation and Compton scattering of photons in the tissue, and detector efficiency response or normalization. The accuracy of these corrections strongly affects the quality of the PET image. There is evidence that time-of-flight (TOF) PET reconstruction is less sensitive than non-TOF reconstruction to inconsistencies between emission data and corrections. The purpose of this study is to analyze and discuss such experimental evidence. In this work, inconsistent correction data (inconsistent normalization, absence of scatter correction and mismatched attenuation correction) are introduced in experimental phantom data. Both TOF and non-TOF reconstructed images are analyzed to examine the effect of flawed data. The behavior of TOF reconstruction in respiratory artifacts, a very common example of inconsistency in the data, is studied in patient images. TOF reconstruction is less sensitive to mismatched attenuation correction, erroneous normalization and poorly estimated scatter correction. Such robustness depends strongly on the time resolution of the TOF PET scanner. In particular, the robustness of TOF in the presence of attenuation correction inconsistencies is discussed, using a simulation of a simple model of respiratory artifacts. We expect new generations of PET scanners, with improved time resolution, to be less and less sensitive to poor quality normalization, scatter and attenuation corrections. This not only reduces artifacts in the PET image, but also opens the way to less stringent requirements for the quality of the CT image (reducing either the equipment cost or the dose to the patient), and for the normalization protocols (simplifying or shortening the normalization procedures). Moreover, TOF reconstruction can be beneficial in multimodalities such as PET/MR, where a direct attenuation measurement is not available and attenuation correction can only be approximated.