Very Long Baseline Interferometry (VLBI) is the only space geodetic technique capable to realise the Celestial Reference Frame and tie it with the Terrestrial Reference Frame. It is also the only technique, which measures all the Earth Orientation Parameters (EOP) on a regular basis, thus the role of VLBI in determination of the universal time, nutation and polar motion and station coordinates is invaluable. Although geodetic VLBI has been providing observations for more than 30 years, there are no clear guidelines how to deal with the stations or baselines having significantly bigger post-fit residuals than the other ones. In our work we compare the common weighting strategy, using squared formal errors, with strategies involving exclusion or down-weighting of stations or baselines. For that purpose we apply the Vienna VLBI Software VieVS with necessary additional procedures. In our analysis we focus on statistical indicators that might be the criterion of excluding or down-weighting the inferior stations or baselines, as well as on the influence of adopted strategy on the EOP and station coordinates estimation. Our analysis shows that in about 99% of 24-hour VLBI sessions there is no need to exclude any data as the down-weighting procedure is sufficiently efficient. Although results presented here do not clearly indicate the best algorithm, they show strengths and weaknesses of the applied methods and point some limitations of automatic analysis of VLBI data. Moreover, it is also shown that the influence of the adopted weighting strategy is not always clearly reflected in the results of analysis.