Singling out modified gravity parameters and datasets reveals a dichotomy between Planck and lensing
An important route to testing General Relativity (GR) at cosmological scales is usually done by constraining modified gravity (MG) parameters added to the Einstein perturbed equations. Most studies have analyzed so far constraints on pairs of MG parameters, but here, we explore constraints on one parameter at a time while fixing the other at its GR value. This allows us to analyze various models while benefiting from a stronger constraining power from the data. We also explore which specific datasets are in tension with GR. We find that models with ($μ=1$, $η$) and ($μ$, $η=1$) exhibit a 3.9-$σ$ and 3.8-$σ$ departure from GR when using Planck18+SNe+BAO, while ($μ$, $η$) shows a tension of 3.4-$σ$. We find no tension with GR for models with the MG parameter $Σ$ fixed to its GR value. Using a Bayesian model selection analysis, we find that some one-parameter MG models are moderately favored over $Λ$CDM when using all dataset combinations except Planck CMB Lensing and DES data. Namely, Planck18 shows a moderate tension with GR that only increases when adding any combination of RSD, SNe, or BAO. However, adding lensing diminishes or removes these tensions, which can be attributed to the ability of lensing in constraining the MG parameter $Σ$. The two overall groups of datasets are found to have a dichotomy when performing consistency tests with GR, which may be due to systematic effects, lack of constraining power, or modelling. These findings warrant further investigation using more precise data from ongoing and future surveys.