Fast times during spring breaks: are traffic fatalities another consequence?
French, Michael T. ; Gumus, Gulcin
I. INTRODUCTION
Between the end of February and the beginning of April, college
students from all over the United States travel to warmer climates to
enjoy a week off from classes. This "spring break" phenomenon
dates back to the late 1930s when Florida, especially the city of Fort
Lauderdale, became a travel hotspot among college students (Bohn 2009).
Since then, spring break (SB) has progressively turned into a college
tradition, particularly for students in northern schools who take
advantage of the appealing climate in
Florida. Currently, SB travel encompasses virtually all college
students to become a huge component of American college culture. It is
estimated that every year millions of students travel for SB, spending
billions of dollars on transportation, lodging, food, and entertainment
(Ribeiro 2011; Scott-Halsell and Saiprasert 2011).
Since the 1930s, besides Florida, cities in California, Nevada,
Texas, Mexico, and the Caribbean have become magnets for spring breakers
(Bohn 2009). Popular destinations receive significant economic benefits
from increased tourism revenue and employment due to this travel
phenomenon. However, these destinations also experience adverse impacts,
such as increased traffic crashes, public intoxication including driving
under the influence (DUI), overcrowding, vandalism, littering,
hospitalizations, and noise pollution (Laurie 2008). As a result, local
communities and public officials struggle to weigh the stimulus to the
local economy against the unwanted harms and risky behaviors associated
with the SB environment.
One of the potentially dangerous and avoidable consequences of SB
is fatal and non-fatal traffic crashes, especially those caused by drunk
driving. However, this adverse outcome has received almost no attention
in the literature. To address this gap, we examine the impact of SB
season on fatal passenger vehicle crashes. Specifically, we investigate
whether traffic fatalities significantly increase during the weeks when
college students arrive at SB hotspots in the United States. We use
daily county-level longitudinal data on passenger vehicle fatalities
from the 1982-2011 Fatality Analysis Reporting System (FARS). In
addition to the aggregate analysis, we conduct separate analyses by age
groups and by alcohol involvement in the crash. Using the extensive
crash characteristics available in FARS, we also examine disaggregated
fatality rates corresponding to in-state versus out-of-state driver
involvement. Our findings indicate that passenger vehicle fatalities are
significantly overrepresented during the SB season.
II. BACKGROUND
College students, being mostly younger adults, are not only
relatively inexperienced drivers, but are also challenged by driving in
an unfamiliar environment if they travel far from home for their SB
vacation. Moreover, drinking and driving is quite prevalent among this
population. American College Flealth Association's (2013) National
College Flealth Assessment reveals that, when students were asked about
the past 30 days, 23.9% of them reported driving after having any
alcohol and 2.8% reported driving after consuming five or more drinks.
As more colleges and universities establish campus policies to ban
alcohol or restrict access, this can fuel the demand for heavy episodic
drinking during SB. Some studies even report students selecting their
vacation destinations in hopes of maximizing their ability to drink,
such as underage American students vacationing in Mexico where the
minimum legal drinking age is lower (Apostolopoulos, Sonmez, and Yu
2002).
Murphy et al. (2012, 339) report that, compared to adult drinkers,
"college students tend to drink episodically, in relatively large
social groups, outside of the context of meals, and often in large
quantities over short periods." SB season may exacerbate these
behaviors due to heightened impulsivity and greater peer pressure during
such holidays. Many college students use SB as an opportunity to
"let loose" and "party hard." Several studies
provide empirical evidence that students on SB vacation partake in
elevated levels of risky activities, especially when alcohol is
involved. For example, Apostolopoulos, Sonmez, and Yu (2002) report that
49% of male and 38% of female college students report having sex as a
direct outcome of excessive drinking on SB vacation. Moreover, they find
that one third of the students they surveyed reported that they had sex
with someone they first met on SB. These casual sexual encounters are
particularly risky given that 75% of students reported never or rarely
using a condom on SB (Apostolopoulos, Sonmez, and Yu 2002). Sonmez et
al. (2006) confirm these findings and discover that students are
significantly more likely to participate in riskier behaviors in the SB
environment than when they stay at home.
More recently, Patrick and Lee (2012) find that students engage in
more risky behaviors when they are vacationing with friends on SB. They
report that students are more likely to have sex, consume more alcohol,
achieve higher levels of blood alcohol concentration (BAC), and report
more episodes of intoxication when they are on SB. Lee, Maggs, and
Rankin (2006) show that students engage in relatively higher amounts of
alcohol consumption during SB if they travel with friends and if they
are members of fraternities or sororities. Similarly, Grekin, Sher, and
Krull (2007) find that college students traveling together with friends
tend to consume more alcohol whereas students who stayed home or
traveled with their parents during SB were less likely to engage in
excessive alcohol use.
Findings from the existing studies have significant public health
and social welfare implications, and they also highlight the need for
possible educational campaigns to protect students and others from
serious consequences occurring during the SB season. One important and
yet overlooked negative externality is traffic crashes in locations that
receive a large amount of college students during the SB season. It has
been established that motor vehicle deaths tend to increase during good
weather and periods of increased recreational travel (e.g., Farmer and
Williams 2005). Studies have also shown that fatalities and injuries in
traffic crashes are overrepresented during holidays due to increased
risky driving behaviors on the roads (e.g., Anowar, Yasmin, and Tay
2013). However, to the best of our knowledge, there is no rigorous study
examining the effects of SB season on traffic safety. Balkin and Ord
(2001. 11) report that Florida tends to have more fatal crashes in March
and they speculate that this pattern could be due to "the increase
of traffic from college students traveling to Florida on spring
break." However, they do not formally investigate this issue and
instead focus on how speed limit increases affect fatal interstate
crashes.
III. DATA AND EMPIRICAL METHODS
The passenger vehicle fatality data used in this study come from
the Fatality Analysis Reporting System (FARS). FARS is a publicly
available data source on fatal vehicular crashes and is maintained by
National Flighway Traffic Safety Administration (NHTSA). Passenger
vehicles include cars, vans, SUVs, and pickup trucks; motorcycles,
scooters, large trucks, buses, off-road vehicles, etc. are excluded.
Fatalities refer to both occupants (i.e., drivers and passengers) as
well as non-occupants (e.g., pedestrians, bicyclists). We obtained daily
fatality data from the 1982-2011 FARS, which were then aggregated to
construct weekly fatality totals, where a week is a period of seven
successive days from Sunday through Saturday. As part of robustness
checks, we adopt various other definitions of a week as discussed below.
The days of the year are not exactly divisible by seven, and so in some
years days are left over before the first full week and/or after the
last full week. Given that these "leftover days" are not
comparable to full weeks, they are excluded from our analyses.
We obtained fatality data for 21 different areas: 14 popular SB
counties and the states wherein these counties are located. These seven
states are Arizona, California, Florida, Nevada, South Carolina, Texas,
and Virginia. Commonly visited SB locations for college students also
include Mexico and the Caribbean, but we focus only on domestic
locations for which comparable fatality data are available. The 14
counties, together with the corresponding SB destination cities, are
listed in Appendix A. Using these county- and state-level data, we
construct fatality figures for the remaining combined non-SB counties in
each of the seven SB states. This allows us to carry out a comparison
between the SB counties and the "rest of the state."
To identify the most intensive weeks of SB season, we collected
data on SB schedules for virtually all colleges and universities in the
United States together with the corresponding enrollment data (see
Appendix B). Unfortunately, information is not available on how many
students travel during SB or their exact travel destinations. However,
the enrollment data coupled with SB schedules clearly show that the most
intensive weeks of travel are the 4 weeks of March, which we
consecutively denote by weeks 2-5. In addition, the SB shoulder weeks
include the last week of February (week 1) and the first week of April
(week 6). As part of a falsification test, we also define adjacent
weeks: week 0 is the third week of February (i.e., the week before the
SB season starts) and week 7 is the second week of April (i.e., the week
following the end of the SB season).
As part of our analysis, we utilize extensive crash characteristics
available in FARS to investigate whether the SB season has differential
impacts for various fatality measures. Total fatalities refer to the
number of passenger vehicle drivers and passengers killed in traffic
crashes. The FARS database enables us to decompose total fatalities by
license state. If the driver is licensed in the state where the crash
occurred, we count the fatalities as in-state. In contrast, out-of-state
fatalities occur in crashes that involve a driver who is licensed in a
state other than that where the crash occurred. Next, we divide total
fatalities into two groups according to the driver's BAC or the
driver with the highest BAC if multiple drivers are involved:
non-alcohol-impaired (BAC is under ,08g/dL) versus alcohol-impaired (BAC
is .08 g/dL or higher) fatalities. These BAC figures are either reported
by law enforcement or imputed based on characteristics of the crash and
driver (Adams, Blackburn, and Cotti 2012; Subramanian 2002). Finally, we
break down fatalities by the driver's age group. We distinguish
between fatalities in crashes that involve young drives (i.e., at least
one driver is 25 years old or younger) versus those that did not involve
any young drivers (i.e., all drivers are more than 25 years old).
A common approach in the traffic safety literature is to estimate
fixed-effects linear regression models using fatality rates defined
either as fatalities divided by population, vehicle registrations, or
vehicle miles traveled. Given that we analyze weekly data and frequently
observe zero fatalities in some counties, we estimate a conditional
fixed-effects negative binomial model to account for overdispersion in
fatality counts (Hausman, Hall, and Griliches 1984). The estimation is
by maximum likelihood and is conditional on the total weekly count of
fatalities in each county. Specifically, we estimate the following
equation:
(1) [F.sub.iwt] = S[B.sub.W][beta] + [[delta].sub.i] +
[[lambda].sub.t] + [gamma][T.sub.w] + [[delta].sub.i] x [T.sub.w] +
[[epsilon].sub.iwt]
where [F.sub.iwt] denotes the fatality count in area i, week w, and
year t. Depending upon the specification, SB is an indicator variable
for either the entire 6-week SB season or a vector of dummy variables
indicating each of the six SB weeks. Area-specific fixed-effects,
denoted by [[delta].sub.i], absorb time-invariant differences in
passenger vehicle fatalities across counties and states. (1)
[[lambda].sub.t] reflects the annual secular nationwide trends in
traffic safety. In addition, we account for weekly linear time trends
(within a given year), [gamma][T.sub.w], as well as area-specific weekly
linear time trends, [[delta].sub.i], x [T.sub.w]. As a result, the
estimated impact of SB on passenger vehicle safety is now identified by
weekly within county/state deviations, net of the weekly area-specific
time trends.
The coefficients of interest are contained in the [beta] vector.
The direction, magnitude, and significance of the estimated coefficients
indicate whether some or all of the SB weeks have an intuitively
predictable, statistically important, and practically meaningful effect
on passenger vehicle safety. Given that we include area-specific time
trends, which should account for much of the weekly variation in
fatalities, the estimated effects can be interpreted as lower bounds of
the true effect of SB. We also estimate the same specifications for the
combined non-SB counties in each of the seven SB states in our sample.
This provides a "difference-in-differences" type of comparison
between the 6-week SB season and the remaining 46 weeks across popular
SB destinations versus the rest of the counties in those same seven
states. Finally, we estimate specifications with supplemental indicators
for weeks that are adjacent to the SB season (i.e., week 0 and week 7)
as a falsification test. If the SB effects are strictly contained within
the SB season, then we should find non-significant coefficient estimates
for week 0 and week 7.
IV. RESULTS
Table 1 provides descriptive statistics for the full sample, which
includes weekly data for the 30 years from 1982 to 2011. We separate the
full sample into two groups: SB counties and the combined non-SB
counties in SB states. The top panel in this table presents summary
statistics for the SB counties. The average number of weekly fatalities
in SB counties is 2.175 with a standard deviation (2.557) that is higher
than the mean. The bottom panel reports descriptive statistics for the
rest of the counties in those states that include at least one SB
hotspot. The average fatality count in these seven SB states is 32.990
per week, which is obviously much higher compared to the county-level
average. When we break down the fatality figures by driver
characteristics, we see that fatalities in crashes involving drivers
with in-state licenses, with no alcohol impairment (BAC < .08), and
no young drivers are far more common, and this pattern holds for both
the top and bottom panels of Table 1.
<4113152H_TB002>
In Table 2, we present a comparison of average weekly fatalities
between SB weeks and the remaining weeks of the year. The top panel
again corresponds to the SB counties where average weekly fatalities
during SB weeks are higher compared to non-SB weeks. This is true not
only for total fatalities, but also for all the other categories we
consider. Except for fatalities of drivers with in-state licenses, all
the differences are statistically significant. The bottom panel reports
the average weekly fatalities in the combined non-SB counties for SB
states. In stark contrast to the top panel, average weekly fatalities
are higher during non-SB weeks compared to SB weeks with one exception
(drivers with out-of-state licenses). Some of these differences,
however, are not statistically significant. Although these comparisons
are a rather simple way of gauging the impact of the SB season on
traffic fatalities, they provide an initial glimpse before we proceed
with our regression analysis.
A. Main Estimates
Tables 3-6 present estimation results for the conditional
fixed-effects negative binomial models with the weekly count of
passenger vehicle fatalities as the dependent variable. For the key
explanatory variables, we report the estimated incidence rate ratios
(IRRs) and the corresponding standard errors (in parentheses). (2) IRRs
are the exponentiated coefficients and indicate the difference in
fatalities predicted by the model when the variable of interest is
increased by one unit above its mean value while all other variables are
kept constant at their means (see Table 1 for the summary statistics). A
value greater than one reflects a positive relationship between fatality
counts and the particular regressor, while a value less than one
suggests the opposite. Statistical significance is based on a test of
the null hypothesis that there is no relationship between fatalities and
the control variable (i.e., IRR is equal to one).
<4113152H_TB003>
First, we consider the SB counties in columns 1-3 of Table 3.
Column 1 includes a single dummy variable for the entire SB season
(i.e., weeks 1-6). The estimated IRR in this column indicates that the
weekly number of fatalities in SB counties is 9.1% higher during the SB
season compared to the remaining weeks of the year (p < .01). This
translates into about .2 additional traffic fatalities per week during
the SB season in a given county, which yields more than 16 additional
fatalities in our sample (i.e., all 14 counties) during the entire
6-week SB season in any given year (2.175 X .091 =. 198 fatalities per
week and SB county, and .198 X 14 SB counties = 2.771, and 2.771 x 6
weeks = 16.626). In column 2, we include separate dummy variables for
each of the 6 weeks of the SB season. Estimated effects for these
week-specific indicators vary between 6.5% and 13.2%. With the exception
of week 1, all of the estimates are statistically significant. In column
3 of Table 3, we augment the model in column 2 with dummy variables for
the two adjacent weeks (i.e., the week before the SB season starts and
the week after the SB season ends). If the fatality effects are confined
to the SB period, then the IRRs for these adjacent weeks should not be
significantly different from one. As expected, the IRRs for the adjacent
weeks (i.e., weeks 0 and 7 in Table 3) are close to one and
non-significant. In addition, the inclusion of these two indicators
leaves the estimated impact of the SB on fatalities virtually unchanged.
In the last three columns of Table 3, we consider the other
counties in SB states. The estimated IRR in the specification with a
single dummy variable for the entire SB season (column 4) is .990 and
non-significant. This suggests that SB season does not have a
significant impact on traffic fatalities in non-SB counties located in
SB states. When we include separate dummy variables for each SB week
(column 5), almost none of the estimated IRRs are significantly
different from one. The only exception is week 1, but the estimated IRR
is .963, suggesting a 3.7% decrease in fatalities during this week (p
< .05). The last column of Table 3 indicates that in non-SB counties,
the week right before the SB season starts (i.e., week 0) is associated
with 4.8% fewer fatalities (p < .05) while the other adjacent week
(i.e., week 7) is associated with a 3.6% increase in fatalities (p <
.10). While we are unable to posit definitive explanations for these
adjacent-week findings, we speculate that it might be due to travel
patterns associated with other (i.e., non-SB) vacationers during these
weeks. Nevertheless, the absence of a significant increase in fatalities
in non-SB counties during the SB season offers counterfactual evidence
in support of a true SB effect in SB hotspots.
Comparing columns 1-3 with columns 4-6 in Table 3 reveals that the
SB season is associated with higher fatalities in SB counties and this
pattern is clearly different from what is observed in the other counties
in SB states. (3) It is important to note, however, that spring breakers
often travel through other counties in a SB state and sometimes choose
other destinations in those states even if they are not necessarily SB
hotspots (e.g., counties on the west coast of Florida). As a result,
comparing SB hotspots and the "rest of the state" is not
always a clear distinction, so we view the estimated effects of SB on
traffic fatalities as lower bound estimates.
In Table 4, we decompose total passenger vehicle fatalities
according to the driver's license status (i.e., whether the driver
is licensed in or out of the state where the crash occurred). If college
students are challenged by driving in an unfamiliar environment as they
travel far from home during their vacation, then we may see a greater
effect on fatalities among out-of-state drivers during the SB season
compared to in-state drivers. We consider fatalities of in-state drivers
in columns 1 and 2 and out-of-state drivers in columns 3 and 4. Even
though both in-state and out-of-state fatalities are positively
associated with the SB season in SB counties, the impact is much more
pronounced for fatalities in crashes that involved out-of-state drivers.
Only week 5 for SB counties is significantly related to in-state
fatalities (p < .05) and none of the SB weeks are significantly
associated with in-state fatalities for non-SB counties. On the other
hand, as presented in column 3, the most popular weeks of the SB season
(i.e., weeks 2-5) are significantly associated with a greater number of
out-of-state fatalities in SB counties (p < .05). Moreover, the
estimated IRRs are relatively large, ranging from 19.7% to 24.6%. Two of
the estimated IRRs (for weeks 3 and 4) are statistically significant for
out-of-state fatalities among non-SB counties (column 4), but the impact
is more modest as both estimates are less than 10%. Thus, as expected,
an influx of college students traveling to SB counties is contributing
to a greater number of traffic fatalities overall, but especially among
out-of-state drivers.
Table 5 reports the results for fatalities broken down by the age
of the driver and once again we compare the SB counties with the non-SB
counties located in the SB states. The first two columns correspond to
crashes that involved one or more young drivers (age 25 or younger) and
the last two columns report estimates for crashes that did not involve
any young drivers. In SB counties, four of the six SB weeks are
associated with significantly higher fatality counts for crashes that
involved a young driver(s) and the corresponding IRRs are quite high
(11.3-20.4%). Only one of the IRRs is significantly different from one
for other counties in SB states (week 1), but the direction is negative
instead of positive. As for fatalities in crashes that involved no young
drivers, the first two SB weeks have marginally significant positive
effects in SB counties. The last two columns reveal that a similar
phenomenon does not appear to exist in non-SB counties in SB states.
The last of our disaggregated analyses addresses the issue of
alcohol involvement in traffic crashes by decomposing fatalities into
two groups: non-alcohol-impaired (BAC < .08) and alcohol-impaired
(BAC [greater than or equal to] .08). If spring breakers are drinking
heavily during their vacation and then driving after partying, we would
expect to see significantly more alcohol-impaired fatalities in SB
counties during SB weeks. As shown in Table 6, both types of fatalities
are typically higher during the SB season in SB counties whereas they
are generally lower during this period in non-SB counties. However, lack
of statistical significance does not allow us to draw any firm
conclusions about alcohol involvement. (4) One possible explanation is
that spring breakers may be using safer forms of transportation (e.g.,
taxi, designated sober drivers) after consuming alcohol. Alternatively,
they may be consuming alcohol in resorts or high-density beach
communities where they stay, hence eliminating the need for
transportation altogether.
B. Robustness
To examine the stability of our core findings, we conduct several
robustness checks. First, we implement a falsification test to determine
whether the SB season is atypical compared to other high-traffic
periods. Research shows that traffic is heaviest during the weeks
surrounding holiday periods. Independence Day (4th July), Labor Day
(first Monday of September), Christmas Day (25th December), and New
Year's Day (1st January) are the top four deadliest holidays in the
United States in terms of passenger vehicle fatalities (Farmer and
Williams 2005). Thus, these weeks should exhibit significantly higher
fatality counts compared to non-holiday periods, regardless of whether
the analysis applies to SB counties or non-SB counties in SB states. In
Table 7, we report the results of this falsification test. The fact that
traffic fatalities spike during most of these holiday periods for both
SB and non-SB counties offers face validity to our SB estimates. A
comparison between columns 1 and 2 does not reveal any systematic
differences in the direction or statistical significance of the
estimated IRRs between the SB counties and non-SB counties during these
holiday periods. The same holds for the comparison between columns 3 and
4. If SB counties are inherently more dangerous throughout the year,
then we should certainly see significant differences during these
holidays compared to the rest of the state. The absence of any
meaningful differences between SB and nonSB counties during these
holidays suggests that our SB findings are indeed unique and robust.
The core analyses for this study use total and disaggregated
fatality counts. To provide an alternative comparison between popular SB
destinations and the rest of the counties in the same states, we also
estimate fixed-effects linear regression models using
"adjusted" differences in passenger vehicle fatalities. This
alternative dependent variable is defined as
([F.sub.SB] - [F.sub.NSB])/[F.sub.S], where [F.sub.SB],
[F.sub.NSB], and [F.sub.S] denote weekly fatality counts for a SB county
in a given state, for the combined non-SB counties in the same state,
and for the entire state (i.e., [F.sub.S] = [F.sup.SB] + [F.sub.NSB]),
respectively. Although extremely rare, this ratio is undefined (and thus
treated as a missing observation) when zero fatalities occurred in the
entire state for a particular week. After creating this adjusted
dependent variable, we estimate linear fixed-effects models with the
same set of regressors as in Equation (1) above. The corresponding
results are presented in Tables 8 and 9.
In Table 8, we focus on the adjusted differences in total
fatalities. The estimate for the dummy variable associated with the
6-week SB season (column 1) is positive and highly statistically
significant (p < .01). This alternative specification indicates that,
once again, during the SB season, fatalities in SB counties are
significantly higher compared to fatalities in non-SB counties. Even
though the estimates in Table 8 do not offer a quantitative
interpretation as intuitive as the IRRs from the count models presented
earlier, we believe this adjusted measure has the advantage of allowing
us to directly gauge the differential effects of SB on traffic
fatalities in SB counties versus the rest of the states (i.e., focusing
on a single set of regression results rather than making side-by-side
comparisons as in Tables 3-7). When including dummy variables for each
of the six SB weeks (column 2), four of the coefficient estimates are
positive and statistically significant (p < .05). Finally, as
reported in column 3, the effects associated with the adjacent weeks are
nonsignificant and the estimates for the SB weeks are unchanged.
Table 9 provides a similar analysis to that in Table 8 for the
decomposed fatality measures. While the SB season is associated with
positive and statistically significant effects in SB counties for all
the decomposed fatality measures, the effects are even more pronounced
for the out-of-state and younger driver fatalities. Regardless of
alcohol involvement, fatalities are higher in SB counties during the SB
season (columns 5 and 6). However, empirical support for higher
alcohol-impaired fatalities during SB is rather weak. Thus, vis-a-vis
our core set of analyses, these findings provide additional empirical
support for the differential effects of SB in SB counties. We also
replicate the analysis in Table 7 using the adjusted differences in
total passenger vehicle fatalities and these estimation results yield no
statistically significant coefficients associated with the weeks
surrounding Independence Day, Labor Day, Christmas Day, and New
Year's Day. Again, this confirms that the patterns we uncover
pertaining to the SB season are not detected for other holiday periods
during which these travel hotspots may be receiving a significant number
of visitors.
Our final robustness check involved alternative definitions for the
start and end of a SB week. The core analysis is based on a definition
for a week that starts on Sunday and ends the following Saturday. Using
the daily fatality data from FARS, we experimented with two alternative
definitions of a week: one from Saturday through Friday and the other
from Monday through Sunday. We then re-estimated all models using these
alternative definitions. Regardless of how we define a week, the
estimation results are largely unchanged. These estimates are not
presented in tables, but the full set of estimation results from these
robustness checks can be obtained from the authors upon request.
V. DISCUSSION AND FUTURE RESEARCH
This study considers traffic fatalities as a potential adverse
consequence of SB. Specifically, we examine whether and to what extent
SB season has an impact on fatal passenger vehicle crashes in 14 popular
SB destinations in the United States. Our findings indicate that
passenger vehicle fatalities are significantly higher in SB counties
when comparing SB season to other weeks of the year and also when
comparing SB counties to non-SB counties in the same states. This SB
effect is more pronounced for out-of-state drivers (relative to in-state
drivers) and younger drivers ([less than or equal to] 25 years). Despite
strong concerns about drunk driving during the SB season, our results
reveal few differences between alcohol-impaired and non-alcohol-impaired
fatalities. We also carry out various falsification and robustness
checks that confirm these core results.
The analysis presented here has a few shortcomings that should be
noted. First, given data limitations, we are unable to adjust passenger
vehicle fatalities by vehicle miles traveled because such data are not
available by county and week. Second, we are unable to control for the
exact number of tourists (including spring breakers) in each county and
week, so the models do not adjust for population density or tourism
exposure. Third, we collect information on SB schedules for all colleges
and universities across the United States to identify the most popular
SB weeks, which formed the primary independent variables. These
schedules coupled with student enrollments may be the best available
information for determining the participation and intensity associated
with SB travel, but it obviously does not fully capture the actual
amount of SB visitors in each of the destinations. Fourth, stay-at-home
spring breakers and those who travel abroad may also be contributing to
adverse consequences in those locations, but we lack corresponding data
to include them in our analysis. Lastly, while our analysis reveals an
increased traffic death toll associated with SB season, it does not
resolve the possible mechanisms. Increased passenger vehicle fatalities
during SB could be due to a variety of reasons including more motor
vehicles on the roads; a greater number of passengers per vehicle; a
higher prevalence of impaired driving,
inexperienced drivers, and speeding; reduced use of seat belts; and
other factors. Perhaps future research could shed some light on these
alternative explanations in order to assist policy makers as they
consider educational programs and more aggressive police enforcement
strategies targeting risky driving behaviors during SB season.
It should be noted that we are unable to distinguish between the
traffic effects associated with an influx of tourists (both spring
breakers and others) from those related to a higher prevalence of risky
behavior and/or peer effects among spring breakers. However, existing
literature presents compelling evidence that students are significantly
more likely to participate in risky behaviors when they are in the SB
environment with friends than when they stay at home (e.g., Grekin et
al. 2007; Patrick and Lee 2012; Sonmez et al. 2006). This suggests that
SB seasons can indeed lead to increased traffic fatality rates overall,
rather than just shifting students to certain destinations where more
crashes occur due to increased population and congestion. More
importantly, even if SB simply shifts students to where traffic crashes
occur, instead of leading to more crashes, the strength and stability of
our findings suggest a role for policy measures to address the fatal
consequences associated with SB. Given the relatively small number of SB
hotspots in the country where at-risk students are grouped together for
only a few weeks each year, enforcement of traffic and alcohol policies
at these destinations can be enhanced and intensified in an effort to
save lives. As an example, for those counties that continue to host
spring breakers, perhaps local officials can improve traffic safety
during the SB season by providing free taxi and/or public transportation
vouchers to students with a valid college ID.
After becoming an iconic SB destination during the last quarter of
the 20th century, the city of Fort Lauderdale adopted stricter public
drinking laws, banned sleeping in cars as well as parking overnight at
the beach, and publicly announced that college students are no longer
welcome (Briggs 2008). For many other SB destinations across the
country, however, such restrictions may not be feasible given the
substantial economic benefits they receive from spring breakers. A full
economic evaluation associated with hosting spring breakers is outside
the scope of the present study, but this may be a fruitful area for
future research. Nevertheless, our findings contribute new information
to the spirited policy debates on SB restrictions as host communities
investigate whether the economic benefits generated by SB visitors
outweigh various adverse consequences they create, including traffic
fatalities. Moreover, college students might reflect on these findings
as they plan future SB itineraries. Perhaps college administrators and
parents will also consider these results as they try to improve student
safety during SB travels.
ABBREVIATIONS
BAC: Blood Alcohol Concentration
DUI: Driving Under the Influence
FARS: Fatality Analysis Reporting System
IRR: Incidence Rate Ratio
SB: Spring Break
doi: 10.1111/ecin.12157
APPENDIX A
The list of popular SB destinations was compiled
based on existing literature reviews conducted by Lau-
rie (2008) and Ribeiro (2011) as well as other historical
reviews including Hobson and Josiam (1996) and Mcleod
(1998). We also gathered information from various SB
travel websites such as the MTV Spring Break Timeline:
http://www.time.eom/time/nation/article/0,8599,1888317,00.
html (accessed June 16, 2013).
City County State
Lake Havasu City Mohave County Arizona
Palm Springs Riverside County California
San Diego San Diego County California
Panama City Beach Bay County Florida
Fort Lauderdale Broward County Florida
Fort Myers Beach Lee County Florida
South Miami Beach Miami-Dade County Florida
and Miami
Key West Monroe County Florida
Daytona Beach Volusia County Florida
Las Vegas Clark County Nevada
Myrtle Beach Horry County South Carolina
South Padre Island Cameron County Texas
Austin Travis County Texas
Virginia Beach Virginia Beach City Virginia
APPENDIX B
SB weeks together with the corresponding student
population figures are as follows:
Student
Weeks Population Percent (%)
Week 1 (last week of February) 181,357 3
Week 2 (first week of March) 707,904 11
Week 3 (second week of March) 2,658,760 40
Week 4 (third week of March) 1,845,916 28
Week 5 (last week of March) 946,270 14
Week 6 (first week of April) 237,113 4
Total 6,577,320 100
This information was obtained from: http://www.
tripsmarter.com/travelinfo/panama-city-beach/spring-break-college-university-dates (accessed June 16, 2013).
REFERENCES
Adams, S., M. L. Blackburn, and C. D. Cotti. "Minimum Wages
and Alcohol-Related Traffic Fatalities among Teens." Review of
Economics and Statistics, 94(3), 2012, 828-40.
Allison, P. D., and R. P. Waterman. "Fixed-Effects Negative
Binomial Regression Models." Sociological Methodology, 32(1), 2002,
247-65.
American College Health Association (AHCA). National College Health
Assessment II: Reference Group Data Report Fall 2012. Hanover, MD:
American College Health Association, 2013.
Anowar, S., S. Yasmin, and R. Tay. "Comparison of Crashes
during Public Holidays and Regular Weekends." Accident Analysis and
Prevention, 51, 2013, 93-97.
Apostolopoulos, Y., S. Sonmez, and C. H. Yu. "HIV-Risk
Behaviours of American Spring Break Vacationers: A Case of Situational
Disinhibition?" International Journal of STD & AIDS, 13(11),
2002, 733-43.
Balkin, S., and J. K. Ord. "Assessing the Impact of
Speed-Limit Increases on Fatal Interstate Crashes." Journal of
Transportation and Statistics, 4(1), 2001, 1-26.
Bohn, L. "A Brief History of Spring Break," Time
Magazine, March 30, 2009. Accessed June 16, 2013. http://www.
time.com/time/nation/article/0,8599,1888317,00.html.
Briggs, B. "Florida Towns Send Spring Break Crowd Abroad: Fort
Lauderdale Shows How to Transform Image by Turning Away Business,"
NBC News, March 17, 2008. Accessed June 16, 2013.
http://www.nbcnews.com/
id/23565992/ns/business-us_business/t/florida-towns-send-spring-break-crowd-abroad.
Farmer, C. M., and A. F. Williams. "Temporal Factors in Motor
Vehicle Crash Deaths." Injury Prevention, 11(1), 2005, 18-23.
Grekin, E. R., K. J. Sher, and J. L. Krull. "College Spring
Break and Alcohol Use: Effects of Spring Break Activity." Journal
of Studies on Alcohol and Drugs, 68(5), 2007, 681-8.
Hausman, J. A.. B. H. Hall, and Z. Griliches. "Econometric
Models for Count Data with an Application to the Patents-R&D
Relationship." Econometrica, 52(4), 1984, 909-38.
Hobson, J. S. P., and B. Josiam. "Spring Break Student Travel:
A Longitudinal Study." Journal of Vacation Marketing, 2(2), 1996,
137-50.
Laurie, J. "Spring Break: The Economic, Socio-Cultural and
Public Governance Impacts of College Students on Spring Break Host
Locations." Paper 876, University of New Orleans Theses and
Dissertations, 2008.
Lee, C. M., J. L. Maggs, and L. A. Rankin. "Spring Break Trips
as a Risk Factor for Heavy Alcohol Use among First-Year College
Students." Journal of Studies on Alcohol, 67(6), 2006, 911-6.
Mcleod, R. "A Brief History of Spring Break." Rolling
Stone, 783, 1998, 56-61.
Murphy, J. G., N. P. Barnett, and C. J. Correia. "Future
Directions in College Student Alcohol Abuse Research and
Prevention," in College Student Alcohol Abuse: A Guide to
Assessment, Intervention, and Prevention, edited by C. J. Correia, J. G.
Murphy, and N. P. Barnett. Hoboken, NJ: John Wiley & Sons, 2012,
339-47.
Patrick, M. E., and C. M. Lee. "Daily Variations in Spring
Break Alcohol and Sexual Behaviors Based on Intentions, Perceived Norms,
and Daily Trip Context." Journal of Studies on Alcohol and Drugs,
73(4), 2012, 591-6.
Ribeiro, N. F. C. C. D. "Culture, Consensus and Behavior among
North-American Spring Breakers." Unpublished dissertation,
Pennsylvania State University, 2011.
Scott-Halsell, S. A., and W. Saiprasert. "Spring Break:
Pulling in the Student Market." Journal of Tourism Insights, 2(2),
2011, Article 3.
Sonmez, S., Y. Apostolopoulos, C. H. Yu, S. Yang, A. Mattila, and
L. C. Yu. "Binge Drinking and Casual Sex on Spring-Break."
Annals of Tourism Research, 33(4), 2006, 895-917.
Subramanian, R. "NHTSA Technical Report: Transitioning to
Multiple Imputation--A New Method to Estimate Missing Blood Alcohol
Concentration (BAC) in FARS." DOT HS 809 403. Washington, DC:
Department of Transportation, 2002.
(1.) As demonstrated by Allison and Waterman (2002), the
conditional fixed-effects negative binomial model is not a
"true" fixed-effects method in the sense that it does not
control for all non-time-varying covariates.
(2.) For brevity, we suppress the estimation results for
county/state fixed-effects, year fixed-effects, linear weekly time
trends, and area-specific time trends. These results can be obtained
from the authors upon request.
(3.) We also estimated conditional fixed-effects Poisson models
that allow for clustering at the area level. Although this exercise
increased the standard errors somewhat, it did not meaningfully alter
any of our conclusions.
(4.) We also constructed fatality groups as non-alcohol-involved
(BAC = 0) and alcohol-involved (BAC > 0). Again, this alternative
grouping did not meaningfully alter the conclusions presented here.
MICHAEL T. FRENCH and GULCIN GUMUS *
* We gratefully acknowledge Tonja Lindsey at the National Center
for Statistics and Analysis Office of Traffic Records and Analysis,
National Highway Traffic Safety Administration, for her assistance with
the FARS data. We are indebted to two anonymous referees, Jeremy Hagler,
Brittany M. Harder, Emmanouil Mentzakis, Linnea A. Polgreen, Lisa M.
Powell, Daniel I. Rees, Nicholas Sando, and participants at the 9th
World Congress of the International Health Economics Association, the
2013 Addiction Health Services Research Conference, the 6th IZA Annual
Meeting on the Economics of Risky Behaviors, and the 5th Biennial
Conference of the American Society of Health Economists for their
constructive comments on earlier versions of the paper. We also thank
Audry M. Klossner for excellent research assistance and Carmen Martinez
for administrative support.
French: Professor, Department of Sociology, Department of
Economics, and Department of Public Health Sciences, University of
Miami, 5202 University Drive, Merrick Building, Room 12IF, P.O. Box
248162, Coral Gables, FL 33124-2030. Phone 1-305-284-6039, Fax
1-305-2845310, E-mail mfrench@miami.edu
Gumus: Assistant Professor, Department of Management Programs,
Florida Atlantic University, 111 Glades Road, Boca Raton, FL 33431.
Phone 1-561-297-2115, Fax 1-561-297-2675, E-mail ggumus@fau.edu); IZA,
Bonn, Germany.
TABLE 1
Descriptive Statistics. 1982-2011
Mean Standard
Deviation
Spring break counties (N = 21,840)
Total passenger vehicle fatalities 2.175 2.557
Fatalities in crashes involving:
Drivers with in-state licenses 1.734 2.206
Drivers with out-of-state licenses .440 .868
Drivers with BAC < .08 (a) 1.443 1.887
Drivers with BAC [greater than or .748 1.157
equal to] .08 (a)
Young drivers (age [less than or .883 1.348
equal to] 25)
No young drivers (age > 25) 1.292 1.708
Combined non-spring-break counties for spring break states (N = 10.920)
Total passenger vehicle fatalities 32.990 25.921
Fatalities in crashes involving:
Drivers with in-state licenses 27.370 22.927
Drivers with out-of-state licenses 5.620 4.233
Drivers with BAC < .08a 20.774 16.348
Drivers with BAC [greater than or 12.068 10.921
equal to] .08 (a)
Young drivers (age [less than or 13.350 11.687
equal to] 25)
No young drivers (age > 25) 19.640 15.295
Min Max
Spring break counties (N = 21,840)
Total passenger vehicle fatalities .000 21.000
Fatalities in crashes involving:
Drivers with in-state licenses .000 16.000
Drivers with out-of-state licenses .000 10.000
Drivers with BAC < .08 (a) .000 15.000
Drivers with BAC [greater than or .000 12.000
equal to] .08 (a)
Young drivers (age [less than or .000 14.000
equal to] 25)
No young drivers (age > 25) .000 14.000
Combined non-spring-break counties for spring break states (N = 10.920)
Total passenger vehicle fatalities .000 139.000
Fatalities in crashes involving:
Drivers with in-state licenses .000 128.000
Drivers with out-of-state licenses .000 32.000
Drivers with BAC < .08a .000 84.000
Drivers with BAC [greater than or .000 60.000
equal to] .08 (a)
Young drivers (age [less than or .000 70.000
equal to] 25)
No young drivers (age > 25) .000 75.000
Notes: Observations are weekly for each spring break (SB) county and
for the combined non-SB counties in states that have at least one SB
county. As a result, there are 21,840 observations for SB counties (52
weeks x 30 years X 14 counties) and 10,920 observations for SB states
(52 weeks x 30 years x 7 states).
(a) BAC is the blood alcohol concentration of the driver.
Source: Authors' own calculations using FARS data.
TABLE 2
Average Weekly Fatalities for Spring Break (SB) Season Versus Rest of
the Year, 1982-2011
SB Counties
SB Rest of
Season the Year
(1) (2)
Total passenger vehicle fatalities 2.313 2.157
Fatalities in crashes involving:
Drivers with in-state licenses 1.784 1.728
Drivers with out-of-state licenses .530 .429
Drivers with BAC < .08 (a) 1.538 1.430
Drivers with BAC [greater than or .785 .744
equal to] .08 (a)
Young drivers (age [less than or .961 .873
equal to] 25)
No young drivers (age >25) 1.353 1.284
At = 21,840 2,520 19,320
SB Counties
Difference p Value
(1)-(2) (1)-(2)
Total passenger vehicle fatalities .157 .004
Fatalities in crashes involving:
Drivers with in-state licenses .056 .233
Drivers with out-of-state licenses .101 .000
Drivers with BAC < .08 (a) .108 .007
Drivers with BAC [greater than or .042 .089
equal to] .08 (a)
Young drivers (age [less than or .088 .002
equal to] 25)
No young drivers (age >25) .069 .057
At = 21,840
Combined Non-SB Counties
for SB States
SB Season Rest of
(1) the Year
(2)
Total passenger vehicle fatalities
Fatalities in crashes involving: 31.690 33.160
Drivers with in-state licenses 26.025 27.545
Drivers with out-of-state licenses 5.665 5.614
Drivers with BAC < .08 (a) 19.865 20.893
Drivers with BAC [greater than or 11.710 12.114
equal to] .08 (a)
Young drivers (age [less than or 12.875 13.412
equal to] 25)
No young drivers (age > 25) 18.814 19.747
N = 10,920 1,260 9,660
Combined Non-SB Counties
for SB States
Difference p Value
(1)-(2) (1)-(2)
Total passenger vehicle fatalities
Fatalities in crashes involving: -1.470 .058
Drivers with in-state licenses -1.521 .027
Drivers with out-of-state licenses .051 .690
Drivers with BAC < .08 (a) -1.027 .036
Drivers with BAC [greater than or -.405 .216
equal to] .08 (a)
Young drivers (age [less than or -.537 .125
equal to] 25)
No young drivers (age > 25) -.933 .042
N = 10,920
Notes: Observations are weekly for each spring break (SB) county and
for the combined non-SB counties in states that have at least one SB
county.
(a) BAC is the blood alcohol concentration of the driver.
Source: Authors' own calculations using FARS data.
TABLE 3
Total Passenger Vehicle Fatalities, Conditional Fixed-Effects Negative
Binomial Models
SB Counties
(1) (2) (3)
SB season (weeks 1 -6) 1.091 ***
(.022)
Week I 1.065 1.063
(.047) (.047)
Week 2 1.098 ** 1.097 **
(.047) (.047)
Week 3 1.089 ** 1.088 *
(.047) (.047)
Week 4 1.083 * 1.082 *
(.047) (.047)
Week 5 1.132 *** 1.131 ***
(.048) (.048)
Week 6 1.077* 1.076 *
(.046) (.046)
Week 0 .950
(.044)
Week 7 1.039
(.046)
Log-likelihood -36,576.80 -36.576.17 -36,575.10
N 21,840 21,840 21,840
Combined Non-SB Counties for SB States
(4) (5) (6)
SB season (weeks 1 -6) .990
(.008)
Week I .963 ** .962 **
(.018) (.018)
Week 2 .993 .992
(.018) (.019)
Week 3 1.003 1.002
(.018) (.019)
Week 4 .983 .981
(.018) (.018)
Week 5 .981 .980
(.018) (.018)
Week 6 1.015 1.014
(.018) (.018)
Week 0 .952 **
(.018)
Week 7 1.036 *
(.019)
Log-likelihood -35,590.61 -35,587.96 -35.582.45
N 10,920 10,920 10,920
Notes: Observations are weekly for each spring break (SB) county and
for the combined non-SB counties in states that have at least one SB
county. As a result, there are 21,840 observations for SB counties (52
weeks x 30 years x 14 counties) and 10,920 observations for SB states
(52 weeks x 30 years x 7 states). Each model includes county-state and
year fixed effects, a linear time trend, and area-specific time
trends. For each explanatory variable, we report the estimated
incidence rate ratios (IRRs) and the standard errors in parentheses.
Statistical significance is based on the test of the hypothesis that
IRR= 1.
*. **, *** Significant at the 10%, 5%, and 1% level, respectively.
Source: Authors' own calculations using FARS data.
TABLE 4
Passenger Vehicle Fatalities by License Status, Conditional
Fixed-Effects Negative Binomial Models
In-State
SB Counties (1) Combined Non-SB
Counties for SB
States (2)
Week 1 1.078
(.051) (.019)
Week 2 1.053 .987
(.050) (.019)
Week 3 1.026 .984
(.050) (.019)
Week 4 1.042 .968
(.050) (.019)
Week 5 1.102 ** .971
(.051) (.019)
Week 6 1.056 1.003
(.050) (.019)
Log-likelihood -32,624.87 -33.466
N 21840 10,920
Out-of-State
SB Counties (3) Combined Non-SB
Counties for SB
States (4)
Week 1 1.080 .933 *
(.092) (.037)
Week 2 1.230 *** 1.023
(.099) (.039)
Week 3 1.246 *** 1.082 **
(.099) (.040)
Week 4 1.230 1.071 *
(.097) (.040)
Week 5 1.197 ** 1.023
(.095) (.039)
Week 6 1.139 1.057
(.093) (.039)
Log-likelihood -17,567.84 -25,544.91
N 21,840 10,920
Notes: Observations are weekly for each spring break (SB) county
and for the combined non-SB counties in states that have at least
one SB county. As a result, there are 21,840 observations for SB
counties (52 weeks x 30 years x 14 counties) and 10,920
observations for SB states (52 weeks x 30 years x 7 states). Each
model includes county/state and year fixed effects, a linear time
trend, and area-specific time trends. For each explanatory
variable, we report the incidence rate ratios (IRRs) and the
standard errors in parentheses. Statistical significance is based
on the test of the hypothesis that IRR = 1.
*, **, *** Significant at the 10%, 5%, and 1% level, respectively.
Source: Authors' own calculations using FARS data.
TABLE 5
Passenger Vehicle Fatalities by Driver's Age, Conditional Fixed-
Effects Negative Binomial Models
Young Drivers (Age
[less than or equal to] 25)
SB Counties (1) Combined Non-SB
Counties for SB
States (2)
Week 1 1.061 .943 **
(.068) (.026)
Week 2 1.093 .998
(.069) (.027)
Week 3 1.162 ** .997
(.071) (.027)
Week 4 1.113 * 1.008
(.069) (.027)
Week 5 1.204 *** 1.019
(.072) (.027)
Week 6 1.184 *** 1.023
(.071) (.027)
Log-likelihood -24,847.53 -29,254.77
N 21,840 10,920
No Young Drivers (Age > 25)
SB Counties (3) Combined Non-SB
Counties for SB
States (4)
Week 1 1.092 * .977
(.057) (.021)
Week 2 1.091 * .994
(.056) (.021)
Week 3 1.015 1.009
(.054) (.021)
Week 4 1.049 .969
(.055) (.021)
Week 5 1.062 .956 **
(.055) (.021)
Week 6 .996 1.010
(.053) (.021)
Log-likelihood -29,235.83 -31,865.03
N 21,840 10,920
Notes: Observations are weekly for each spring break (SB) county and
for the combined non-SB counties in states that have at least one SB
county. As a result, there are 21,840 observations for SB counties (52
weeks x 30 years x 14 counties) and 10,920 observations for SB states
(52 weeks X 30 years x 7 states). Each model includes county/state and
year fixed effects, a linear time trend, and area-specific time
trends. For each explanatory variable, we report the estimated
incidence rate ratios (IRRs) and the standard errors in parentheses.
Statistical significance is based on the test of the hypothesis that
IRR = 1.
*, **, *** Significant at the 10%, 5%, and 1% level, respectively.
Source: Authors' own calculations using FARS data.
TABLE 6
Passenger Vehicle Fatalities by Driver's BAC, Conditional
Fixed-Effects Negative Binomial Model Results
Non-Alcohol-Impaired (BAC < .08)
SB Counties (1) Combined Non-SB
Counties for SB
States (2)
Week 1 1.105 ** .944 ***
(.055) (.020)
Week 2 1.061 1.010
(.053) (.020)
Week 3 1.091 * 1.009
(.054) (.020)
Week 4 1.079 .987
(.053) (.020)
Week 5 1.103 ** .978
(.054) (.020)
Week 6 1.070 1.017
(.053) (.020)
Log-likelihood -30,204.04 -31,721.28
N 21,840 10,920
Alcohol-Impaired (BAC [greater than
or equal to] .08)
SB Counties (3) Combined Non-SB
Counties for SB
States (4)
Week 1 .975 1.001
(.066) (.027)
Week 2 1.135 ** .970
(.072) (.027)
Week 3 1.048 .999
(.068) (.027)
Week 4 1.051 .987
(.069) (.027)
Week 5 1.130 * .988
(.071) (.027)
Week 6 1.042 1.017
(.068) (.027)
Log-likelihood -23,153.61 -28.448.21
N 21,840 10,920
Notes: Observations are weekly for each spring break (SB) county and
for the combined non-SB counties in states that have at least one SB
county. As a result, there are 21,840 observations for SB counties (52
weeks x 30 years X 14 counties) and 10,920 observations for SB states
(52 weeks X 30 years X 7 states). BAC is the blood alcohol
concentration of the driver. Each model includes county-state and year
fixed effects, a linear time trend, and area-specific time trends. For
each explanatory variable, we report the estimated incidence rate
ratios (IRRs) and the standard errors in parentheses. Statistical
significance is based on the test of the hypothesis that IRR = 1.
*, **, *** Significant at the 10%, 5%, and 1% level, respectively.
Source: Authors' own calculations using FARS data.
TABLE 7
Falsification Test Using Holiday Periods, Conditional Fixed-Effects
Negative Binomial Model Results
SB Counties (1) Combined Non-SB
Counties for SB
States (2)
Independence Day period 1.089 ** 1.094 ***
(.046) (.019)
Labor Day period 1.032 1.032 **
(.032) (.013)
Christmas Day period 1.011 .972
(.045) (.018)
New Year's Day period .955 .976
(.114) (.048)
SB season (weeks 1-6) -- --
Log-likelihood -36.583.69 -35,573.71
N 21,840 10,920
SB Counties (3) Combined Non-SB
Counties for SB
States (4)
Independence Day period 1.101 ** 1.093 **
(.047) (.019)
Labor Day period 1.036 1.032 **
(.033) (.013)
Christmas Day period 1.003 .973
(.045) (.018)
New Year's Day period .985 .974
(.117) (.048)
SB season (weeks 1-6) 1.094 *** .993
(.022) (.008)
Log-likelihood -36,573.76 -35.573.38
N 21,840 10,920
Notes: Holiday periods correspond to the week that contains each
holiday. Observations are weekly for each spring break (SB) county and
for the combined non-SB counties in states that have at least one SB
county. As a result, there are 21,840 observations for SB counties (52
weeks x 30 years x 14 counties) and 10,920 observations for SB states
(52 weeks x 30 years x 7 states). Each model includes county-state and
year fixed effects, a linear time trend, and area-specific time
trends. For each explanatory variable, we report the estimated
incidence rate ratios (IRRs) and the standard errors in parentheses.
Statistical significance is based on the test of the hypothesis that
IRR = 1.
*, **, *** Significant at the 10%, 5%, and 1% level, respectively.
Source: Authors' own calculations using FARS data.
TABLE 8
Adjusted Differences in Total Passenger Vehicle
Fatalities, Linear Fixed-Effects Regression
Models
(1) (2) (3)
SB season (weeks .015 *** -- -1-6)
Week 1 (.005) .030 ** .030 **
(.011) (.012)
Week 2 -- .017 ** .018 **
(.008) (.008)
Week 3 -- .011 ** .011 **
(.004) (.005)
Week 4 -- -.003 -.003
(.008) (.008)
Week 5 -- .022 ** .022 **
(.008) (.009)
Week 6 -- .014 .014
(.009) (.010)
Week 0 -- -- .001
(.007)
Week 7 -- -- .005
(.008)
Mean of the -.765 -.765 -.765
dependent variable
Standard deviation of .302 .302 .302
the dependent
variable
Log-likelihood 8,862.19 8,867.19 8,867.39
N 21,824 21,824 21,824
Notes: Dependent variable is defined as the total fatalities
in a given spring break (SB) county minus the total fatalities
for the combined non-SB counties in the corresponding state
that includes the specific SB county, and these differences are
then adjusted by the total fatalities in the entire state (including
both the SB and non-SB county figures). Observations
are weekly for each SB county. As a result, there are a total
of 21,840 observations (52 weeks x 30 years x 14 counties).
The missing observations are due to zeros in the denominator
of the adjusted differences measure, i.e., zero total fatalities
in the entire state in a given week. Each model includes
county/state and year fixed effects, a linear time trend, and
area-specific time trends.
*, **, *** Significant at the 10%, 5%, and 1% level,
respectively.
Source: Authors' own calculations using PARS data.
TABLE 9
Adjusted Differences in Passenger Vehicle Fatalities by
License Status, Driver's Age, and Driver's BAC, Linear
Fixed-Effects Regression Models
Young
Drivers Drivers Drivers
with In- with Out- (Age [less
State of-State than or
Licenses Licenses equal to]
(1) (2) 25) (3)
SB season (weeks 1-6) .013 ** .026 *** .021 ***
(.005) (.009) (.006)
Mean of the dependent variable -.765 -.769 -.770
Standard deviation of the .336 .352 .340
dependent variable
Log-likelihood 5,807.59 -3.849.96 1,163.49
N 21,757 21,407 21,524
Drivers
with BAC
No Young [greater
Drivers Drivers than or
(Age > 25) with BAC < equal to]
(4) .08 (5) .08 (6)
SB season (weeks 1-6) .013 * .013 *** .018 *
(.006) (.003) (.008)
Mean of the dependent variable -.772 -.764 -.775
Standard deviation of the .314 .320 .339
dependent variable
Log-likelihood 4,217.73 4,984.77 463.73
N 21,768 21,765 21,568
Notes: Dependent variable in each column is defined as the
fatalities in a given spring break (SB) county minus the
fatalities for the combined non-SB counties in the
corresponding state that includes the specific SB county,
and these differences are then adjusted by the fatalities in
the entire state (including both the SB and non-SB county
figures). Observations are weekly for each SB county. As a
result, there are a total of 21,840 observations (52 weeks X
30 years x 14 counties). The missing observations are due to
zeros in the denominator of the adjusted differences
measure, i.e., zero fatalities in the entire state for any
given measure. BAC is the blood alcohol concentration of the
driver. Each model includes county-state and year fixed
effects, a linear time trend, and area-specific time trends.
*, **, *** Significant at the 10%, 5%, and 1% level,
respectively.
Source: Authors' own calculations using FARS data.
