Freshwater scientists routinely analyze data with linear (e.g., regressions and ANOVAs) or non-linear models, including hierarchical models with random effects and non-normal error structures. A common challenge with these models is how to make an inference on the hypothesis of interest: difference in treatments, slope of a regression, etc. This is typically achieved with a significance test using p-values or confidence intervals. However, significance tests convert rich ecological and quantitative information into a simple dichotomy of "significant" or "non-significant". Such focus on dichotomies often comes at the expense of more ecologically meaningful quantities like effect sizes and uncertainty. We argue that Bayesian data analysis naturally reorients the focus of analyses to effect sizes and to quantifying uncertainty. It does this by placing emphasis on model structure, likelihoods, and parameter estimates, rather than dichotomous significance tests. However, while modern software has made it easier to conduct Bayesian analysis (and frequentist analysis for that matter), there remain deep philosophical differences in these two approaches that are difficult to reconcile just by changing some code. We explore those differences and discuss how they have changed our approach to statistics using examples of studies from aquatic subsidy research, food webs, and ecotoxicology. We also include examples of positive peer-review that contradict commonly held assumptions that Bayesian analyses are harder to publish because reviewers are unfamiliar with it.