I'm trying to do what I thought was a basic problem but doesn't seem to be working out properly. I'm looking at disproving a claim that tides cause earthquakes, where the main mechanism that people claim is that it happens during a new or full moon, and especially during a perigee moon (when it's closest to Earth).

People like to give ±1 week windows which is basically half a lunar cycle, so it seems basic that the probability an earthquake would happen, given random chance, would be 50% that it would happen within a week of, e.g., a full moon. Similarly, since perigee/apogee happens roughly on the same timescale, it seems as though it would be 50% chance that an earthquake would happen within 1 week of a perigee moon. Put the two together and you have a 25% chance that an earthquake would happen within both a perigee and full moon, correct?

(Important note, updated: Actual period between new/full moons averages 29.52 days, actual time between perigee moons averages 27.56 days. However, apogee and perigee are NOT normally distributed. Apogee happens a mode of 27.78 days (mean is 27.55±0.27), while perigee is much more asymmmetric, having a mode of 28.4 days at the peak of approximately a Lorentzian. Half the max of the Lorentzian is ±0.08 days. But, the mean is 27.56 with a standard deviation of 1.12; the range is 24.6-28.6 days. I'm thinking this could throw off the modeling?)

Assuming that's correct, I seem to be running into issues when trying to figure out the probability that an earthquake would happen by chance within ±X days of BOTH a perigee and full moon. I thought the equation would be simply (2*X/(# days in lunar month, 29.5))*(2*X/(# days between perigee, 27.5)).

However, when I do a monte carlo simulation with 500,000 randomly chosen dates within the time period of 1933 to 2012 (just happens to be when I have earthquake data), the fractions do not line up. For example, the simulation shows that 14.4% should be within 5 days of both a full and perigee moon, but my above math says it's only 12.3%.

I have checked the results of my simulation against days from perigee, apogee, new moon, and full moon times. As expected from a random distribution with a large $N$, the number of times that the simulated earthquake is a given time period away from maximum perigee/apogee/new/full moon is even. Except for perigee, where I see a fall-off for $>|±12.5|$ days from when it's closest to perigee. I'm thinking this has to do with the non-Gaussian distribution of perigee times? And could that account for the 2% difference at the 5-day example?

Is the best way to approach this, because these perigee times are a bit crazy, to simply go with the Monte Carlo results?

P.S. This has been updated to better reflect a correction I made in my data. When I initially posted this, I had some incorrect full/new moon dates in my table that were throwing some results off.

  • $\begingroup$ I'm confused. Why are you estimating the proportion of days that are at different phases of the moon? You have actual data - that is, you know, for each day, where the moon was in its cycle. $\endgroup$
    – Peter Flom
    Aug 28 '12 at 20:59
  • 1
    $\begingroup$ Given what you've got, I would not simulate data at all. I would look into time series analysis. This would let you account for the fact that earthquakes in a given time period are not independent of earthquakes in the previous time period. It would also make it possible to look at moonphase (probably a variation on seasonality). $\endgroup$
    – Peter Flom
    Aug 28 '12 at 21:15
  • 1
    $\begingroup$ When I do the calculation described in the question (by sampling daily for $80$ years) I get $12.3$% for the answer. In R: days <- 1:(80*365.25); k <- length(days[days %% 27.56 < 2*5 & days %% 29.52 < 2*5]); k/length(days). Varying the starting phases changes the result by less than $0.03$%. The $15.2$% must be a computational error. $\endgroup$
    – whuber
    Aug 28 '12 at 22:52
  • 1
    $\begingroup$ And yes, I know that there isn't a correlation, but I'm actually doing this for a podcast I run, "Exposing PseudoAstronomy," and this is the next episode topic. So I wanted to do the analysis myself and then compare that with what the handful of pseudoscientists claim. I find going through the exercise oneself is often enlightening, and I'll be in a better position to inform my listeners. $\endgroup$ Aug 29 '12 at 2:23
  • 2
    $\begingroup$ If your code doesn't produce the same answer as whuber, then you probably have a bug in your code, and there isn't much we can do. $\endgroup$ Aug 29 '12 at 6:13

There is an oddity that $29.52/27.56$ is unusually close to $15/14.$ So, even though the proportion of times within $5$ days of both will eventually be about $\frac{2X}{29.52} \frac {2X}{27.56}$, it might take longer to converge than you expect. Instead of trying to calculate by a formula, you can simply ask a computer to count how many minutes are within $X$ of the perigee and full moon over that whole range. However, I don't think this effect is large enough to explain the difference between $15.2\%$ and $12.3\%$.

A Monte Carlo simulation normally would mean that you choose times randomly, not evenly spaced. One possibility is that you chose a gap which is too close to a simple rational times one of the periods. $79$ years/$43000$ is $1.49$ days, which might be too close to $29.52/20 = 1.48$. If you used exactly $29.52/20$ then instead of $10/29.52 = 33.88\%$ samples within $5$ days you would get $7/20 = 35\%.$ Again, this doesn't seem to be large enough to explain the discrepancy, but why add significant errors of this type?

  • $\begingroup$ Thank you for the reply. I've updated my question - I had an error in my tables for new/full moon times where some were crossed, resulting in non-uniform results when I corrected the Monte Carlo to large, random $N$ times. $\endgroup$ Aug 29 '12 at 1:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.