| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
LETTERS TO THE EDITOR |
Department of Psychology; University of California; San Diego, CA (Christenfeld)
We, sharing what seems widely experienced surprise at the magnitude of the effects we reported of initials on longevity (1), welcome efforts, such as those of Morrison and Smith (2), to explore further the phenomenon. Welcoming such efforts, however, does not perhaps absolutely require accepting the offered conclusions, and some of the points raised in that paper warrant further scrutiny.
Morrison and Smith suggest that they know of no persuasive biological theory for why initials might have a profound effect on longevity. We would suggest, instead, that one such theory is apparent in the data we originally presented. The causes of death with the largest effects, across men and women, positive and negative initials, are suicide and accidents, with HIV and the miscellaneous categories not so far behind. The biological link between a gun to the head or careening into a bridge abutment and premature mortality is perhaps not so very mysterious. Most causes of death have behavioral antecedents, which makes the impact on health of psychological states credible, and for our findings, concentrated in suicides, accidents, and HIV, that could hardly be more direct.
More central than our ability to generate intuitions about putative mechanisms, however, is the issue of the effect itself. Morrison and Smith suggest that our analyses are biased because we examine people who have died with various initials, positive, negative, and neutral, and then check how many years earlier they were born. They point out, quite rightly, and we noted this ourselves in the original paper, that trends in the popularity of initials over time can bias such a measure. Initials that have only recently become popular, because their owners, in order to be included in the death files, must have died young, will falsely be associated with reduced mortality. To control for such effects, we included a control group whose initials showed the same trend in popularity as the negative initials. We did not, as the comment on our paper suggests, simply replicate, for the control group, the ratio of popularity in the last year to the first, as such a procedure would be hopelessly noisy, but instead fit a regression line through all the years, and found a group with a matched trend. It is still, of course, possible that some second-order or higher characteristics of the trend differ, but the procedure does deal with basic concerns such as those raised in the comment.
Critically, the comment suggests that trying to factor out trends is not the correct approach, but that matching people on birth year will provide an unbiased look at the effect. They point out, again quite rightly, that, given that many people born in the years studied have not yet shuffled off their mortal coil, the average age at death will underestimate the actual longevity of the population. They also point out, and still we agree, that, were the two populations—those with initials of interest and those with initials of gibberish—identical, this measure, even though underestimating longevity, would not introduce a false difference between groups. We suggest, however, that finding no difference where none exists is only one important attribute for a statistical test—and not an especially high hurdle to clear—and that being able to find an effect where one does exist also represents a worthy goal for a statistical test. Their test, and for just such a reason did we originally rejected that approach, does not score that worthy goal.
Why Morrison and Smith's approach does not offer a fair test is complex, and we must ask for some minutes of attention, or at least indulgence, while we lay out illustrative examples. Let us suppose that we perform our analysis on those born in 1967, and examine age at death for people with bad versus neutral initials. Let us further suppose that those with bad initials are indeed prone to premature mortality. This will make such people much more likely to be included in our sample, as those outliving their fourth decade will be found perhaps on tennis courts or golf courses, but not in death registries. While bad initials will be overrepresented compared with the proportion of babies burdened with such monograms, there is no necessary effect on their average age at death. If their survival curve simply plunges toward extinction twice as fast over this period as does the curve of those with neutral initials, the average age at death would be identical. And, if we may tax your patience yet further, scenarios can readily be constructed where those with bad initials, even though prone to premature mortality, actually, when examined in this way, have a higher average age at death. To illustrate this, employing the same example, let us further posit—and in their commentary the authors so propose—that the effects of initials are not likely to be profound for infant mortality; those with bad and neutral initials will be equally likely to survive those dangerous years. The effect on mortality will become increasingly manifest with age, so that those with bad initials are increasingly included in the death records as they approach the age of 40, while those with neutral initials, mostly still healthy, do not yet make much of an appearance there. This is what you would expect especially if the effects are concentrated in suicides, accidents, and HIV, but will generally be true regardless. What one would then find is that, in the sample, the average age of death for the group burdened with negative initials trends toward the later years, when their premature mortality occurs, while the positive group will have a younger average age, as infant mortality will comprise a larger proportion of this population. In this case, finding a greater age at death for the negative group is an artifact of not including anyone who lives past 40, and so failing to give the neutral group credit for doing so at a greater rate.
One might think, and certainly hope, that the problem would be reduced when examining the earliest years of birth, because then the sample will include those who have managed to become old before succumbing, as all must ultimately do. However, with good mortality data starting only in 1960, the same problem exists, only now from the other end. (This problem would not exist with complete mortality data, but, although the authors include mortality data from 1905, they, rightly, point out that the early data is effectively irrelevant, as it almost always is missing the birth year.) People born in 1900 who, because their parents failed to protect them from bad initials, die young are not necessarily included in the sample, and so the average age at death of that group will not reflect their actual longevity, but instead just the distribution of death risk across the available interval. Just as with those born in 1967, there are possible, and even plausible, distributions that would not only obscure a real effect, but even reverse it.
We are not suggesting that this artifactual backwards effect will necessarily be present. For some birth years, even with an effect of initials, one would expect to detect nothing with such an analysis, with some birth years one might well find a negative effect, and with some the positive effect, all depending on the exact shape of the survival curve, and when the effects of initials become manifest. Lumped together, the safest bet is that the test will reveal no effect and, critically, one cannot distinguish with such a result, or lack of one, a failure to find a real effect from a disconfirmation of the theory.
We do not suggest that our original paper is the last word on the topic, and hope that this response is also not that. There are reasonable concerns about just how one selects the sets of positive and negative initials. The reported effect is large and surprising. So, too, though, are reports of the apparent effects of initials, single ones even, and names on preferences, occupations, and residences (3), and effects on longevity do have a lifetime to accumulate. Whether the original effect will continue to bear up to scrutiny or not, we cannot report any displeasure or disappointment yet that we have gone through life without initials such as D.I.E.
DOI:10.1097/PSY.0b013e318159b4a9
REFERENCES
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
