Abstract
Our current understanding is that an environment - mainly consisting of gas or stars - is required to bring massive black hole binaries (MBHBs) with total redshifted mass $M_z\sim [10^{4},10^7]~{\rm M}_{\odot }$ to the LISA band from parsec separation. Even in the gravitational wave (GW) dominated final inspiral, realistic environments can non-negligibly speed up or slow down the binary evolution, or leave residual, measurable eccentricity in the LISA band. Despite this fact, most of the literature does not consider environmental effects or orbital eccentricity in modelling GWs from near-equal mass MBHBs. Considering either a circular MBHB embedded in a circumbinary disc or a vacuum eccentric binary, we explore if ignoring either secular gas effects (migration and accretion) or eccentric corrections to the GW waveform can mimic a failure of general relativity (GR). We use inspiral-only aligned-spin 3.5 post-Newtonian (PN) waveforms, a complete LISA response model, and Bayesian inference to perform a parameterized test of GR. For a 4-yr LISA observation of an MBHB with $M_z=10^{5}~{\rm M}_{\odot }$, primary-to-secondary mass ratio $q=8$, and component BHs' dimensionless spins $\chi _{1,2}=0.9$ at redshift $z=1$, even a moderate gas-disc imprint (Eddington ratio ${\it f}_{\rm Edd}\sim 0.1$) or low initial eccentricity ($e_0\sim 10^{-2.5}$) causes a false violation of GR in several PN orders. However, correctly modelling either effect can mitigate systematics while avoiding significant biases in vacuum circular systems. The adoption of LISA makes it urgent to consider gas imprints and eccentricity in waveform models to ensure accurate inference for MBHBs.