Disentangling magnification in combined shear-clustering analyses

We investigate the sensitivity to the effects of lensing magnification on large-scale structure analyses combining photometric cosmic shear and galaxy clustering data (i.e. the now commonly called "3$\times$2-point" analysis). Using a Fisher matrix bias formalism, we disentangle the contribution to the bias on cosmological parameters caused by ignoring the effects of magnification in a theory fit from individual elements in the data vector, for Stage-III and Stage-IV surveys, assuming well-known redshift distributions, sample selection based on magnitude, and magnification strengths inferred from CFHTLenS and DES Y1 data. We show that the removal of elements of the data vectors that are dominated by magnification does not guarantee a reduction in the cosmological bias due to the magnification signal, but can instead increase the sensitivity to magnification. We find that the most sensitive elements of the data vector come from the shear-clustering cross-correlations, particularly between the highest-redshift shear bin and any lower-redshift lens sample, and that the parameters $\Omega_{M}$, $S_8=\sigma_8\sqrt{\Omega_M/0.3}$ and $w_0$ show the most significant biases for both survey models. Our forecasts predict that current analyses are not significantly biased by magnification, but this bias will become highly significant with the continued increase of statistical power in the near future. We therefore conclude that future surveys should measure and model the magnification as part of their flagship "3$\times$2-point" analysis.