Stochastic point processes are widely applied to model spatiotemporal earthquake occurrence. In particular, the epidemic type aftershock sequence (ETAS) model has been shown to successfully reproduce the short-term clustering of earthquakes. An important parameter of the model is the α-value describing the scaling of the aftershock productivity with magnitude of the triggering earthquake according to 10αM. Fitting of the space-dependent ETAS model to empirical data yields α-values that are typically much smaller than the scaling inverted from more simple stacking of aftershock sequences. We show by means of synthetic simulations that this is likely to result from assuming spatial isotropy of aftershock occurrence that in fact aligns along the mainshock rupture. We fit the space-dependent and space-independent ETAS models to simulations where each earthquake is a line source with an empirical magnitude-length relation. Although the space-time model describes past activity quite well, it overestimates the forecasted earthquake rate. On the other hand, the application of the space-independent ETAS model predicts future seismicity well and can therefore be applied for forecasting purposes. Our test for the observed aftershock sequence following the 1992 M 7.3 Landers earthquake supports these results.