In a previous post I used a dataset taken from my Facebook friends to debunk the claim that more babies are born under a new moon. I have now a chance to increase the test statistics significantly, not because in the meanwhile I made many more friends, but because people collaborating to the FiveThirtyEight blog made available two datasets from the US . See their post “Some people are too superstitious to have a baby on Friday the 13th”.

These very large datasets span the periods 1994-2014 and are provided in a very conveniente CSV format which can be easily manipulated with the python cvs reader. I then re-used the analysis program based on CERN ROOT I used for the previous post.

To summarize the method, I tested four different models:

- The “null hypothesis”, i.e. the distribution is flat and can be described by a constant function;
- The “main hypothesis”, i.e. that the distribution peaks during the first days of the lunar cycle. In this case, the gaussian is scaled by a signal strength parameter
**µ**, and the standard deviation is set to 5 (less than a week);
- The “modified main hypothesis”, i.e. the distribution peaks in the first days, but can account for a very long tail (the standard deviation
**σ** is a free parameter to be fitted);
- A periodic model, described by a cosine function

As you can imagine, the best result is obtained for the null hypothesis. Interestingly, the other two main models can be fitted successfully only if the signal strength is allowed to be very small, essentially zero f.a.p.p. Finally, the periodic model is still allowed by the fit, but the amplitude is strongly constrained and the period is around one lunar cycle.

But the question remains: is there any scientific basis for the claim? A possible explanation hinges on the McClintock effect, also known as menstrual synchrony, combined with the known observation that on average the menstrual cycle has approximatively the same length of the lunar month. Allegedly, women who begin living together in close proximity experience their menstrual cycle onsets becoming closer together in time. However, even in this case, the scientific evidence for this hypothesis is very shaky to say the least. Similarly, the relationship between the two cycles is now believed to be a coincidence.

If you want to play around with the same dataset using iPython, here’s some code provided by Marcello.

### Click here to share this post:

### Like this:

Like Loading...

*Related*

Ciao Riccardo, why do you have those large error bars ? Aren’t the data Poisson in the bins ? I would imagine uncertainty bars extending to +-40 counts in each bin.

Besides, the constant fit is too good – the data with those huge error bars are within much less than one sigma away from the constant.

Cheers,

T.