It will be seen that the number of hallucinations recorded as occurring between nine and ten years ago is less than one-third of the number recorded for the last twelve months. Nay, if the analysis is carried still further, it is found that within the last year the number of hallucinations remembered decreases month by month as we recede further from the present. The inference is irresistible, that the great majority even of interesting hallucinations do not sufficiently impress the memory to be preserved for a few years. After a careful analysis of the figures the Committee are of opinion that the number of visual hallucinations actually experienced by their informants since the age of ten would be approximately secured by multiplying the recorded number by _four_.[105] But if hallucinations in general are not remembered enough, coincidental hallucinations, at least those which coincide with the death of the person seen,[106] would appear to be remembered too well, as will appear from the following figures. There are 13 such cases recorded during the last ten years. Now if we assume that this figure accurately represents the number of such coincidences that have occurred in the experience of our informants during the last ten years, then, since the average age of our informants in this particular case is 46, we should expect to find for the whole period since the age of ten years 47 such coincidences reported; that is on the assumption that no death-coincidence is ever forgotten, and that the liability to such hallucination is practically uniform during the entire period. We do actually find 65 cases; from which it should, the Committee think, be inferred, not only that few or no death-coincidences are forgotten,--a result which is probably not surprising,--but also that a certain number of cases which are not death-coincidences have by the lapse of time grown to appear so.[107] Nor is it difficult to conjecture the particular form of error. It is probable that in most of the 18--more or less--spurious death-coincidences, there was an actual phantasm and an actual death, but that the two events did not stand in close relation to each other. We have already (see Chap. VI.) seen reason to suspect a constant tendency to magnify the closeness of a coincidence of this kind. Seen from a distance the two events--like a binary star-system--are apt to coalesce into one; and a new spectral analysis is required to dissociate them.