In the practice and research on reverberation chambers, the concept of unstirred field components have often been invoked as soon as empirical distributions fail to comply with goodness-of-fit tests applied to data collected at a frequency of operation far beyond the lowest usable frequency. The current explanation for this phenomenon is that under certain conditions, the field samples generated by means of a stirring procedure are not characterized by a zero average-value, but actually present a deterministic offset term that is linked to a line-of-sight contribution. In this paper, we prove that this practice is not sound, as it does not acknowledge the fact that even a very low but non-zero residual correlation between the samples is enough to put in jeopardy the validity of the limits imposed by most hypothesis-test statistics, and hence their ability in properly detecting any constant contribution. An alternative approach is here proposed, based on the analysis of the variability of the line-of-sight contribution estimate, capable of accounting for the residual correlation in a reverberation chamber. Experimental results are presented to support the validity of our approach, exposing the critical use of goodness-of-fit tests as currently applied.