Does the lunar cycle affect human behavior? This seems to be a question that refuses to die, no matter how hard it is to confirm any actual effect. It’s now a cultural idea, deeply embedded and not going anywhere. A recent study, however, seems to show a correlation between suicide and the week of the full moon in a pre-Covid cohort of subjects from Marion County. What are we to make of this finding?

Specifically they show that:

We analyzed pre-COVID suicides from the Marion County Coroner’s Office (n = 776), and show that deaths by suicide are significantly increased during the week of the full moon (p = 0.037), with older individuals (age ≥ 55) showing a stronger effect (p = 0.019). We also examined in our dataset which hour of the day (3–4 pm, p = 0.035), and which month of the year (September, p = 0.09) show the most deaths by suicide.

They also found suicides were not significantly associated with the full moon for subjects <30. They did not give the p-value for those between 30 and 55, and I suspect, given the numbers, that group was either not significant or barely significant. The first question we should ask when confronted with data like this is – is this likely to be a real effect, or just a statistical false positive? We can think about prior plausibility, the statistical power of the study, and the degree of significance. But really there is one primary way this question is sorted out – replication.

In fact, studies like this are best done in at least two phases, an initial exploratory phases and a follow up internal replication to confirm the results of the first data set. But confirmation can also be done through subsequent replication, by the same group or others. A real effect should replicate, while a false positive should not. The authors note that they did a literature search and found the results “mixed”. In fact, this study is a replication (not exact replication, but still a replication) of earlier studies asking the same question. Did they replicate the results of previous studies? Let’s take a look.

A study from 1977 looking at suicides in Cuyahoga County, Ohio, for 1972–1975 found, “…an increase is observed in this sample with respect to new moon phase but not for full moon phase.” That is essentially the opposite of what the current study showed, a decrease in suicides during the full moon.

A 1992 review of 28 studies found:

“Most studies indicated no relation between lunar phase and the measures of suicide. The positive findings conflicted, have not been replicated, or were confounded with variables such as season, weekday, weather, or holidays. It is concluded that there is insufficient evidence for assuming a relationship between the synodic lunar cycle and completed or attempted suicide.”

A 2021 study from Northern Finland found an association with the full moon, but only in premenopausal women.

A 2005 study from Middle Franconia found:

“No significant relationship was detected between the full, absent, and moon’s interphases and suicide incidence. Nevertheless, there was a weak association between the absent moon and choice of a non-violent suicide method in men aged less than the median of 40.2 yrs.”

A 2008 study from Australia involving 65,000 suicide cases found:

“Observed proportions of both male and female suicide occurrence did not deviate from expected proportions during the new, crescent, full, and decrescent moon quarters or from those expected for 3-day windows centered around new and full moon, relative to the interphase. Subgroup analysis (by sex and year), additionally conducted for demonstration purposes, yielded results conspicuously resembling those of related studies with positive findings; namely, sporadically emerging significant findings that were entirely absent in the overall analysis and directionally erratic, thus suggesting they were spurious (false positive).”

These were the studies that came up when I searched on “suicide” and “lunar cycle” in Pubmed. This is not a thorough systematic review, but it is a representative sampling. The Australian study, I think, nails it quite well. When you do an analysis on this question, with lots of variables, you are likely to get false positive results, but they will be erratic and in different directions. When you do a large enough study, these erratic findings tend to average out and the overall numbers become negative.

It’s not enough to find some statistical correlation. Especially when you find it only in a subgroup analysis, or only when comparing some variables and not others, such correlations may be false, just random noise in the data. There are a couple of technical questions to ask – did the researchers control for multiple comparisons? To partially control for false positives, you can adjust the statistics to account for the fact that you are making multiple comparisons. Each comparison is another roll of the dice. If, for example, you make 20 comparisons, on average one will reach a p-value of 0.05. But it may not be significant if you adjust for the fact that you made 20 comparisons.

Also, I have to wonder in papers such as this – are there unpublished comparisons? Did they look at the data before settling on which comparisons they would publish? That’s a form of p-hacking. Of course, you can’t know what was not published (unless the authors disclose unpublished data).

But again, one primary method we use to determine if an effect is real is to look for consistency in the literature – does it replicate? In this case I think the obvious answer is – no. The results are not only mixed in terms of being positive or negative, they are mixed in terms of the details of the correlation. In one study the correlation is for the new moon, another the full moon, one for young women, another for only those >55 years old. The results are all over the place. This is the classic pattern we see when the results are just quirky false positives.

So I am not convinced by this one small study showing results that conflict in detail with the rest of the published literature. This looks like just one more random false positive result. Until we find a pattern that consistently replicates, there is no reason to think the effect is real.