Do some basic descriptive statistics. Some will be used for calculating sampling errors.
Calculate the average normalized article count per rounded degree.
A periodogram is useful for seeing periodicities in the data.
Compare to a random simulation to see some idea of unusual behavior.
There is a peak at 2 and 3. Let’s get more particular about pulling out one Fourier Transform peak at 30. It corresponds to a frequency of ~12, i.e. this may be same basis as for the astrology signs. Also, there is an even stronger signal at 93, corresponding to a frequency of ~4. This may be the basis for squares.
A spectrogram visually confirms this frequency and period. A random data spectrogram follows for comparison.
We can see more particularly the spikes directly in the Fourier transform of the data.
Going back to the data values, let’s fit the results. They look like they fit well to a normal curve.
The match to the normal curve is seen in a probability plot.
Let’s look closer at the sampling errors of the average normalized article counts. At most they are less than 1 percent of their corresponding values.
A plot of the values with error bars
Let’s look at the peaks and the base of the peaks.
The base as a moving average with a window of 7 degrees may show 12 peaks, consistent with the Fourier transform peak above at 30 and perhaps substantiating the idea of (unequal) houses.
Peaks of peaks may land in house 1, 4, 7, 10; i.e. the square houses a.k.a. kendras.
Idealized Theoretical Result
Play with sliders.
More documentation will be added in the coming days ... .
Reasons why we can’t go down to one hundredths of a degree: I suspect that precision largely does not reflect the multi-day process of the news publication cycle, also sampling errors would be too high. Both are good reasons to avoid it.
Phi Golden Angles (Landscheidt’s grand crosses, etc.)
Monte Carlo Simulation to get a sense of the p-value of such successful intersections