Correlations from Daily Activity

If states of mind and emotion within great numbers of people affect the GCP network, we may conjecture that the rhythms of the day might have impose some structure on the data. The thought is that an indirect effect might be generated by way of the changes in consciousness related to daily activities. The effect would be subtle but might be detected in the form of a correlation between the two independent measures of event-linked effects. The netvar and covar measures do show a similar pattern of response to the events, on average, and we can hypothesize that this tendency to track might also be found in responses to daily activity variations. The following is a description of Peter Bancel's test of this notion.

I look at the netvar/covar correlation for the average day. I calculate the average netvar and covar for each minute of the day over all days in the database. The assumption is that there is something regular about people's acitvity during a 24 hr period and this should be a weak signal in the netvar and covar. If the 2 stats are picking this up, then the "average" days should have a small correlation between these stats. This is a correlation on two 1440 element vectors (working with minute data).

It is not significant.

However, in net/cov correlations on the event data we learned there are net/cov correlations on the timescale of 1-2 hours. We see this but blocking or doing moving averages on that timescale. The event results give correlation z-scores around 2.

If I do this for the average daily net and cov, I DO find a positive correlation of about Z = 1.5. Coincidence?

I can check by doing the following:

If the correlation is tied to the day cycle it will wash out if I average "days" which are incommensurate with the real 24-hr day. Easiest and most efficient way to do this is to take "day" lengths very close to 24-hrs. This gives maximal smearing of any signal locked to the 24-hr cycle. So I redo the calc with "days" of lenth 1430 to 1450 minutes. Only the "day" with 1440 minutes locks onto the 24-hr cycle, thus testing for a correltaion signal. Here's a quick plot which I hope is self-explanatory:

The peak is bang on the 1440 minute "day". Stats are weak, but this is really intriguing and may bolster other long-term stuff we have. Note that a sidereal day is at about -4 and appears to show no effect.

Correlations from
Daily Activity

The peak is bang on the 1440 minute "day". Stats are weak, but this is really intriguing and may bolster other long-term stuff we have. Note that a sidereal day is at about -4 and appears to show no effect.


GCP Home