The results of an investigation into whether or not eclipse shadow bands have an atmospheric origin are presented. Using high altitude balloon and ground-based photodiode arrays during the 21 August 2017 total solar eclipse, data revealing the light patterns before and after totality were collected. These data were then analyzed using spectrograms. Both at the altitude of the balloon and on the ground, a sustained ~ 4.5 Hz signal was detected a few minutes before and after totality. This signal was coherent over a scale greater than 10 cm and detected in four separate balloon photodiodes and 16 ground photodiodes. At higher frequencies, up to at least 30 Hz, brief chaotic signals that were disorganized as a function of time were detected on the ground, but not at the altitude of the balloon and appeared mostly uncorrelated over a length scale of 10 cm. Some of our ground arrays utilized red and blue filters, but neither the sustained 4.5 Hz signal nor the higher frequency signals showed a strong dependence on filter color. On the ground we made a video of the shadow bands on a scaled white screen. We judged that the bands were roughly parallel to the orientation of the bright thin crescent Sun before and after totality and inferred that their propagation velocity was about v ~ 59 cm /s. Shadow band signals other than the sustained signal at ~ 4.5 Hz are consistent with atmospheric scintillation theory. These results are surprising. Based on accounts in the literature we expected to confirm the atmospheric scintillation theory of eclipse shadow bands, but instead we detected a sustained ~ 4.5 Hz signal at both high altitude and on the ground. This signal cannot be due to atmospheric scintillation and we ran a check to make sure this signal is not an artifact of our electronics. We recommend that additional searches for eclipse shadow bands be made at high altitude in the future.