Last Update 20 Jan 2005

This page is a working document and is not to be quoted or referenced

New Year Variance, Analysis 2004

The following are plots of the New Year variance event (the cumulation over 5 years) using different selections of timezones. The separate bars show the composite Z-score across the N largest zones, by population, where N is the number of zones. For example, the set of zones labled "24" uses the 24 most populous timezones.

In addition to the original 10 minute period surrounding midnight, we also examine broader periods. This gives some insight into the structure of the event. It appears that the effect is greater when more time is included, especially after midnight.

Most of the sets exclude 2 zones: Nepal and Chatham Islands. These zones are shifted by 15 minutes which makes the periods overlap if we look beyond -5 +5 minutes. Excluding them does not make a big difference.

C30 and C20 are the complements of the 30 and 20 zone sets; these are sets with least populations.

It's interesting to see that the complement zones are indeed weak.

The lower right plot shows actual z-scores for the individual years for 2 timezone sets: the 35 zone set and the 23 zone set. We can see from this display that the general indication of stronger effect in populated zones is not driven by an outlier. It is interesting to note that the individual z's are (almost) all positive, which is in itself somewhat unlikely, with a pval of 0.03, on 5 events.

This study depends on the tabulation of information like the list of time zones used in the sets, with their populations. The short answer: sets 37 to 14 all have 95% of world pop. The complementary sets C30 and C20 are comprised of zones containing less than 5% of the world's pop.

More needs to be done for this analysis, including a similar examination of the Meanshift, but it already provides interesting insight into what appears to be substantial structure, which seems to have some face (ecological) validity.

New Year Variance, Analysis
2004

Seven New Years, Two Analyses

Another look at the data for the New Year transition shows that after correcting a calculation error, the evidence for a mean shift (squared Stouffer Z measure) is negligible. However the device variance measure shows sufficient consistency in the shape-based analysis over seven years to be significant. This link also discusses the error and its correction.

GCP Home