The study, by Dr. James P. Kossin of the Cooperative Institute for Meteorological Satellite Studies at the University of Wisconsin, first looked at the dates of the first storm and last storm for each season. The results showed that "the time of occurrence of the first and last annual formation events exhibit decreasing and increasing trends, respectively, suggesting that the length of the hurricane season is increasing." The chart to the right above shows a graph of the dates and trend lines. The red lines are the trends for the entire period 1851 to 2007. The blue lines cover only the period of aircraft reconnaissance data, and the green lines are for the satellite observation era.
Looking at just the first and last dates doesn't say anything about shifts in the distribution within the season, however, so Kossin also used a statistical analysis technique called "quantile regression." The quantiles represent subsets within the distribution. For example, 20% of the storm formation dates occur prior to the date at the 0.2 quantile. The quantiles, shown in the charts below (click to enlarge), indicate a negative trend (earlier formation) for the beginning of the season and a positive trend (later formation) for the latter part of the season.
Kossin's overall conclusion is:
A consistent signal emerged that suggests the season has become longer as the earliest formation dates of the season have become earlier and the latest dates have become later. While the sign of these trends remain consistent when different periods are considered, the uncertainty is sometimes high. . . . The analyses presented here show how storm formation dates have been changing within the historical hurricane record, but cannot be used to directly implicate cause for these changes or to accurately predict future changes.