Singin’ in the rain
The damage caused by hurricanes and tropical storms in the Gulf of Mexico regions is astronomical. But every few years we hear the same story all over again: entire neighbourhoods brought down by strong winds — or even flooded — and dozens of avoidable casualties. The question is, why don’t people learn their lesson? While being 100% hurricane-proof is not currently possible, there is definitely room for improvement.
To investigate human behaviour before and during hurricanes, we decided to dive into the big-data world of Google Trends, a web application offered by Google Inc. This tool allows you to see the “interest over time” for a specific search-term or a combination of terms. The output is a line chart in which the y-axis values represent search interest relative to the highest point on the chart for the given region and time. A value of 100 is the peak popularity for the term during the specified time period. A value of 50 means that the term is half as popular as at its peak. Likewise, a score of 0 means the term was less than 1% as popular as the peak. The search can be refined geographically and temporally.
We analysed data spanning from April 1st, 2004 to March 31st, 2007 and restricted the selection to four US states bordering the Gulf of Mexico: Louisiana, Mississippi, Alabama, and Florida. We selected a predominantly English-speaking region to simplify which search terms to focus on. During the three-year period, twelve major hurricanes hit the selected areas: Bonnie, Charley, Frances, Ivan, Jeanne, and Matthew in 2004; Dennis, Katrina, Rita, and Wilma in 2005; Alberto, and Ernesto in 2006. Such hurricane-dense period offers also the possibility to see if, after the event, a change in search behaviour occurs.
So what does “Singin’ in the rain” have to do with all of this? Well, pun intended. To compare the different sets of data among each other we had to find a reference value for rescaling. By reference value, we mean a term with fairly constant search interest over time, that can be used as a fixed point when comparing search trends among during and after different hurricanes. We need such a term to compensate for the fact that Google Trends only provide relative frequencies, not absolute numbers of searches. After investigating a few different potential reference values, we figured out that the term lyrics presented a generally steady flat pattern across regions and time periods, making it the perfect candidate for comparisons. With this reference value, here is what we found out about hurricanes:
The blue line represents searches for hurricane during the period from April 1st, 2005 to March 31st, 2006. The search trend peaks in conjunction with meaningful events (from left to right, hurricane Dennis, Katrina, Rita, and Wilma). Hurricane Wilma is listed in the top 10 costliest hurricanes in US history, together with Harvey, in 2017 and Katrina, in 2005. Nonetheless, the chart shows that, in Louisiana, which was hit by hurricane Katrina few weeks earlier, more people in the affected region were searching for song lyrics (red line) than for hurricane-related information.
A greater interest in songs than survival wasn’t limited to the people of Louisiana. We noted a similar pattern in Alabama, which borders with Florida, the state most impacted by Wilma.
The conclusion that people don’t use Google to prepare properly is depressing. But, looking on the bright side, our finding might actually be of use for anyone interested in analysing Google search trends. Before discovering lyrics, we tried a series of other search terms, such as dog, weather, beach, and surf. Each of these either showed an unsteady pattern, or were not scoring significantly compared to hurricane. For instance, at first we thought that dog was an acceptable candidate only to later discover that in March 2007 there was a cat and dog food recall in North America, Europe, and South Africa, caused by renal failures in pets, and that probably contributed to the search trend for dog spiking up (e.g. in Alabama, below). Weather and beach, on the other hand, presented a strongly unsteady pattern, most likely caused by the seasonality. Surf, while appearing relatively steady, shows a low relative score in comparison to the other search term hurricane, making it inappropriate to use as a reference term.
Looking at the entire period of interest we see that searches for the word lyrics outnumber searches for hurricane even during some of the most devastating hurricanes to hit the Gulf area. People are, quite literally, singing in the rain.
This post was written in collaboration with Melanie Mucke, Dept. of Physics and Astronomy, and Johanna Strömberg, Dept. of Mathematics, Uppsala University.