Biodiversity indicators are now frequently used, e.g. inside multiple sustainable forest management frameworks or ecocertification schemes. Yet, these indicators are more based on consensus, on concepts or general ideas than on more detailed analyses associated to data. Does a given indicator reflect the variation of floristic or avifauna biodiversity? In which direction, with which magnitude? I will here present three statistical hints which I think are important to take into account in such an endeavour: (i) to answer such questions, one of the tools needed are parametric, usually multivariate, statistical models. Also, the Bayesian framework appears at present as practically more adequate than the frequentist one; (ii) second, I will develop the perspective that probabilistic quality of the models is important, especially with messy data such as biodiversity data. I will especially present new Bayesian goodness of fit tools as well as new count data probability distributions; (iii) finally, I will insist that estimation to study the magnitude of the effects is more interesting in this perspective than trying to reject point null hypotheses. Of course, this is only part of the story. I will finally discuss other points such as how to incorporate potential mechanisms in such studies.