In the first part we shall consider the statistical analysis of large scale structure in galaxy surveys. We demonstrate a method for jointly constraining cosmology and photometric redshift distributions using cross correlations between photometric and spectroscopic redshift bins. This allows one to reduce the bias in the inferred cosmological parameters which may be propagated from errors in the redshift distributions. We demonstrate this using parameters for a DES-like survey, using galaxy number count C(l)s and CMB-TT information. We continue in this vein to apply these methods to the search for modified gravity using the Euclid survey. We forecast constraints on Horndeski theories using α-function parameterisation, using combined probes of galaxy number counts cross correlated with weak lensing shear, and independently adding CMB-TT information. We see that, as expected, the constraints on the α-parameters are not significantly degenerate with the other cosmological parameters; this is promising as it means that their detection would be less prone to misconstruction. In the second part we consider the universe on considerably smaller scales, and concern ourselves with the local group (LG). We first explore the use of artificial neural networks for estimating the mass of the LG. Using the Timing Argument as a bench mark, we find that the ANN can make use of novel physical information (in our case, the eigenvalues and eigenvectors of the velocity shear tensor) to improve the scatter of the estimates considerably. We then proceed to explore the analytic Timing Argument mass further, exploring its dependency on dark energy and modified gravity models. Beginning with Λ, we proceed to perfect fluid models, quintessence fields, and scalar-tensor theories of gravity in the weak field limit.