Resonances appearing above the ionization threshold in coupled-state calculations of positron-atom scattering are discussed. Calculations in the six-state approximation Ps(1s, 2s, 2p)+H(1s, 2s, 2p), which show such resonance structure are compared with a more extensive 18-state approximation Ps(1s, 2s, 3s, 4s, 2p, 3p, 4p, 3d, 4d)+(H(1s, 2s, 3s, 4s, 2p, 3p, 4p, 3d, 4d) in which channels other than 1s, 2s and 2p are represented by pseudostates. The results strongly indicate that the above ionization threshold resonances observed in the six-state approximation, and in other small basis set calculations, are not real. It is suggested that they are a consequence of the neglect, or inadequate representation in other approximations, of ionization channels. In the six-state approximation the positronium component of the system wavefunction attempts to represent the missing ionization channels but in so doing produces unreal resonances above the ionization threshold. More generally, it is suggested that, in coupled-state calculations of positron-atom scattering, the atom part of the system wavefunction will try to compensate for defects in the positronium component and vice versa. When the defects are serious, for example, the omission of important ionization channels, unusual spurious behaviour is to be expected.