The Fourth of July weekend is a time for barbeques in the park, relaxing in the summer sun, and of course enjoying one of America’s most long-standing traditions: fireworks. In fact, Americans have been celebrating the Fourth with these controlled explosions since 1777 . For the vast majority of citizens, fireworks serve as a fun and exciting form of entertainment. There are, however, a sizable number of people injured by fireworks. In what follows, we explore the nuances of this phenomenon.
The average American is 100 times more likely to injure themselves with fireworks on July 4th than should be expected based on random chance alone.
The Consumer Product Safety Commission (CPSC) has collected data on product related injuries through the National Electronic Injury Surveillance System (NEISS) since 1991. We gathered data from the NEISS on fireworks relate injuries between 2000 and 2015 and specifically focused in on the days leading up to and following July 4th.
On average, an estimated 9,784 people visit the ER every year with injuries related to fireworks or flares – based on estimates from 2000 to 2015. A staggering number of these injuries occur on and around the 4th. Between 2000 and 2015, injuries on the 4th account for between 18.8% and 31.3% of the yearly estimates (mean = 25%).
However, anyone with a dog in the house knows that fireworks explosions are not confined to July 4th alone; people tend to celebrate with explosives for several days leading up to and following the holiday. When we take this into account, the numbers become even more surprising. Between 2010 and 2015, the Fourth of July holiday weekend, July 4th and 5th (3rd and 4th in 2011), accounts for between 40.2 and 47.4 percent of the total yearly injuries[a]!
Does a bad economy affect fireworks injuries on the 4th of July?
We have come to expect an increase in fireworks related injuries on and around the 4th of July each year. However, should we expect that the number of injuries we see each year should be the same? Or, might there be circumstances that can lead to greater or fewer injuries in a given year. Locally it would make sense that adverse weather conditions leading to firework show cancelations or postponements could reduce the number of injuries in that area. However, since regional estimates aren’t available from the NEISS, we are unable to examine the relationship between weather and fireworks injuries.
One question we might consider is whether or not economic decline or downturn might significantly impact injury rates. For instance, take consumer confidence. If Americans have low consumer confidence, i.e. the feeling of less expendable income, they are less likely to spend money on luxury or recreational goods such as fireworks. Finally, fewer fireworks being set-off leads to fewer fireworks related injuries.
In order to examine this potential relationship we looked at national yearly estimates of firework related injuries from 2000 to 2015 in relation to some economic variables. We found that severely adverse economic conditions are substantively and significantly related to fewer firework injuries.
First, we looked at the Index of Consumer Sentiment (ICS), developed by researchers at the University of Michigan, for the month of June as an independent variable. This indicator variable took on the value of one in 2001 and 2008, and equaled zero in all other years. We attempted to incorporate the uncertainty in the NEISS’ estimates by simulating 10,000 models. In each model we used a new set of national injury estimates, based on the data provided by the NEISS. The results look very similar to fitting a single model using the raw NEISS estimates. The simulation models’ pushed the estimated strengths of variable relationships a touch closer to zero and slightly increased the estimated uncertainty of those relationships.
Additionally, we fit a second model. In this model we included an indicator variable for the beginning of a recession. This indicator variable took on the value of one in 2001 and 2008, and equaled zero in all other years. We also ran 10,000 simulations for this model.
We attempted to account for some uncertainty in the NEISS’ estimates by simulating 10,000 models. In each model we used a new set of national injury estimates, based on the data provided by the NEISS. In the figure below we present the results of two models based on these simulation that control for two instances of economic recession in 2001 and 2008. For all of the 10,000 simulations of a respective model, we estimated the number of firework related injuries occurring each year[b].
Because we are specifically interested in the relationship between economic downturn and firework injuries, we have highlighted 2001 and 2008 in both plots. In the no-indicator specification (left) we can clearly see that the model is consistently overestimating the number of injuries in both 2001 and 2008, as the box is very close to the upper end of the NEISS estimate confidence interval.
However, when we add the recession indicator, the model is no longer exhibiting this pattern of consistent overestimation. The boxes are now nicely centered between their respective NEISS bands. This fairly drastic change suggests that there is something special about recessions, which isn’t captured by the ICS measure.
Based on the better fitting second model simulations, we do in fact find evidence that economic downturn is significantly related with fewer fireworks injuries over the 4th of July holiday. A 13 point decrease in the ICS is associated with 717 fewer fireworks injuries in a given year. Further, the beginnings of the 2001 and 2008 recessions are associated with 1,050 fewer injuries, although this result is not statistically significant.
 ” http://www.sca.isr.umich.edu/tables.html”
[a] For a variety of reasons, the NEISS can’t collect data from every hospital in the United States every day. So, in order to get around this problem, they sample hospitals. This allows them to estimate the number of injuries nationally without actually seeing data from every hospital. This method does come with a downside though: uncertainty. Those black bars in the figure above reflect that uncertainty and are called “confidence intervals”. Say that we randomly sampled from the population hospitals again 100 times and for each new sample we calculated a new “confidence interval”. Well, we would expect that 95 of those new confidence intervals would contain the actual number of injuries that occurred nationally.
[b] These predictions are shown with the blue boxplots. The red lines represent the confidence interval estimates from the NEISS – these are the same as the error bars in the earlier charts.