In a recent work, Dai [1] searched for a variability in Newton’s constant G using the IGETS based gravitational acceleration measurements. However, this analysis, obtained from χ 2 minimization, did not incorporate the errors in the gravitational acceleration measurements. We carry out a similar search with one major improvement, wherein we incorporate these aforementioned errors. To model any possible variation in the gravitational acceleration, we fit the data to four models: a constant value, two sinusoidal models, and finally, a linear model for the variation of gravitational acceleration. We find that none of the four models provides a good fit to the data, showing that there is no evidence for a periodicity or a linear temporal variation in the acceleration measurements. We then redid these analyses after accounting for an unknown intrinsic scatter. After this, we find that although a constant model is still favored over the sinusoidal models, the linear variation for G is marginally preferred over a constant value, using information theory-based methods.

We implement a test of the variability of the per-cycle annual modulation amplitude in the different phases of the DAMA/LIBRA experiment using Bayesian model comparison. Using frequentist methods, a previous study [1] had demonstrated that the DAMA amplitudes spanning over the DAMA/NaI and the first phase of the DAMA/LIBRA phases, show a mild preference for time-dependence in multiple energy bins. With that motivation, we first show using Bayesian techniques that the aforementioned data analyzed in [1] show a moderate preference for exponentially varying amplitudes in the 2-5 and 2-6 keV energy intervals. We then carry out a similar analysis on the latest modulation amplitudes released by the DAMA collaboration from the first two phases of the upgraded DAMA/LIBRA experiment. We also analyze the single-hit residual rates released by the DAMA collaboration to further look for any possible time-dependency. However, we do not find any evidence for variability of either of the two datasets by using Bayesian model selection. All our analysis codes and datasets have been made publicly available.