Analyzed data set which contained list of possible factors that can explain the mortality rates. Ran regression models with various permutations and combinations of variables to arrive at a list of variables that can explain variances in mortality rates.
1. How do drunk driving
laws affect traffic
deaths?
Medha Tiwary (mxt164730)
Anvitha Ananth (axa169131)
BUAN 6312.003
2. Contents
Introduction..............................................................................................................................................................................3
Examining Models..................................................................................................................................................................4
Vehicle Fatality Rates ~ Variables directly related to drunk driving.................................................................5
Vehicle fatality rates across age groups ~ Variables that are directly related to drunk driving laws ..6
Vehicle fatality rates ~ Population distribution and variables directly related to drunk driving laws8
Vehicle fatality rates ~ Variables directly and indirectly related to drunk driving laws...........................9
Vehicle fatality rates and Socio-Economic Variables.............................................................................................11
Vehicle fatality rates across age groups and Socio-Economic variables........................................................12
Which is more significant - spircons, jaild, or unrate?..........................................................................................14
Age group with the highest vehicle fatality rate......................................................................................................15
Do different variables affect fatality rates in the 21-24 age group? ................................................................16
Conclusion..............................................................................................................................................................................17
3. Introduction
This project investigates the impact beer tax and other related policies have on the fatality rate in
forty-eight states in the US. This investigation is based on a data set spanning over six years and it
encompasses effect of other economic as well as cultural factors on the fatality rate. The data
provided is a panel dataset. Given that the data set is based on the forty-eight states, we chose to
incorporate regression models based only on fixed effects. Random effect model will not provide
any insight in this case since the data points are from a defined set of populations.
Data Analysis:
The data spans from 1982 to 1988 and uses data for forty-eight states. The segregation of fatality
rate is based on the following:
1) Vehicle Fatality Rate (mrall)
2) Night-time Vehicle Fatality Rate (mralln)
3) Alcohol-Involved Vehicle Fatality Rate (mraidall)
Since the aim is to determine the impact of drunk driving laws on traffic deaths, the above-
mentioned attributes are chosen as relevant dependent variables in the model.
Once the significance of the explanatory variables is established for the above mentioned
dependent variables, age group specific fatality rates will also be considered in the regression
model.
Given the problem at hand, the most significant explanatory variables are: beertax, jaild (mandatory
jail sentence), comserd (mandatory community service), minimum legal drinking age (mlda).
Beertax and mlda impact the alcohol consumption rate which although this doesn’t directly pertain
to a law, tax on case of beer is usually indicative of alcohol related taxes, hence it is included.
Prior to building a model it is important to understand how the explanatory variables interact with
each other. Figure 1 presents the correlation matrix whereby the correlation of the most relevant
explanatory variables is mentioned. It is natural to expect that the consumption rate (spircons) will
be strongly negatively correlated with the beer tax but it is evidently not so. Upon further analysis,
it was discovered that the beer taxes on an average have declined over the period of six years
(Figure 2). Since, the beer taxes have decreased over the years the consumption rate (spircons) is
weakly negatively correlated with beer tax rate.
4. Figure 1: Correlation matrix
Figure 2: Average beer tax trend over 6 years across 48 states
Examining Models
The provided data set is panel data about the drinking preferences and fatality rate across 48 states.
Random effects is appropriate in cases where the sample was selected by random sampling. In the
given dataset, we have data pertaining to the lower 48 states in the United States. Hence random
effects is not appropriate for the given dataset.
Between pooled least square estimators and fixed effects, fixed effects is more accurate as it also
considers the “within variation”. Hence, we conclude that fixed effects is the most appropriate
model for the given dataset.
5. Vehicle Fatality Rates ~ Variables directly related to drunk driving
Figure 3: Regression output for vehicle fatality rate (mrall)
Figure 4: Regression output for night-time vehicle fatality rate (mralln)
Figure 5: Regression output for alcohol-involved vehicle fatality rate
Inference from Figure 3-5:
The above regressions were carried with the perspective that the three dependent variables are
6. most explicable by the explanatory variables pertaining to laws and taxes.
Beertax was found to be slightly significant for mrall, but highly insignificant for mralln and
mraidall. This is contrary to our theoretical understanding of the model but given the decline in the
beer tax rate over time this isn’t a very surprising result.
The following are the inferences from the above regressions:
● Mlda was highly insignificant for mrall, but slightly more significant for mralln and mraidall.
● Minimum legal drinking age Mraidall which is the alcohol-involved vehicle fatality rate should
● Jaild has been significant for mrall, mralln and mraidall. The possibility of getting arrested
certainly discourages people from drinking and driving
● Comserd was insignificant in mrall, and slightly more significant in mralln and mraidall
● It can be concluded that the only explanatory variable that was significant across the three
dependent variables was – jaild
It is important that the explanatory variable is significant for vehicle fatality rates, night time
vehicle fatality rates and alcohol related fatality rate as the drunk driving laws are do not
differentiate between the time of the day, they are valid throughout the day. In the previous
regressions, we understood that jaild was the only significant variable.
Vehicle fatality rates across age groups ~ Variables that are directly
related to drunk driving laws
Of all the law related variables, Jaild has emerged as the only significant variable.
Further, we chose to investigate whether the variables that are directly related to drunk driving
laws are more influential across an age group.
Prior to running regression models, it is expected that mlda is more likely to be significant across
the youngest age group (15-17) and will have a negative coefficient given that the minimum age for
drinking across most states is 21.
7. Figure 6: Regression output for vehicle fatality rate for age group 15-1
Figure 7: Regression output for vehicle fatality rate for age group 18-20-year-old
Figure 8: Regression output for vehicle fatality rate for age group 21-24-year-old
Inference from Figures 6-8:
● Beertax is significant in the 15-17 age group, but is insignificant in the other two groups. This is
an obvious outcome for this age category since drivers in this age group are not allowed to drink
legally. The same is explained by the coefficient and significance of mlda in this category which
is otherwise insignificant in the age groups of 18-20 and 21-24
● Irrespective of the age group Jaild continues to be significant
● Comserd, variable for community service is significant in the 15-17 age group, but is
insignificant thereafter. This result is different from the initial expectation of that community
service will the significant across all age groups
Based on the two sets of regression outputs it can be concluded that jaild is the only variable that is
strongly significant both across age groups as well as various fatality rates.
8. Vehicle fatality rates ~ Population distribution and variables directly
related to drunk driving laws
Given the significance of the jail term which is a law directly affecting the fatality rate the next step
would be to include the population variables to see if the population distribution has an impact on
the fatality rates, and to also see if the interaction of population distribution with the variables
directly related to drunk driving laws has an impact.
Figure 9: Regression output for vehicle fatality rate against population
Figure 10: Regression output for night-time vehicle fatality rate against population
9. Figure 11: Regression output for alcohol-involved vehicle fatality rate against population
Inference from figures 9-11:
● Contrary to what it was assumed, adding the population distribution variables does not give us
results that vary from the previous regressions
● None of the population distribution variables: pop1517, pop1820 and pop2124 are significant in
this model. Not only were the p-values high but also the coefficients for certain population
groups were insignificant
● Jaild was insignificant for the mrall and mralln regressions, it was highly significant for mraidall.
Thus, it is reasonable to conclude that jaild continues to be the only variable that is directly
related to the drunk driving laws
Vehicle fatality rates ~ Variables directly and indirectly related to
drunk driving laws
To be able to thoroughly understand the impact of the laws on the fatality rate in the population it is
important that other indirect variables are considered in the regression model. Thus, far directly
impacting variables have been considered in the models. Moving ahead the idea is to include
variables such as spircons, dry, yngdrv.
Since jaild has been the only variable significant in enough number of regression models, including
it in the following model adds more meaning to the model.
Hence, the following regression will include variables that are both directly and indirectly related to
drunk driving laws:
10. Figure 12: Regression output for vehicle fatality rate against economic factors
Figure 13: Regression output for night-time vehicle fatality rate against economic factors
Figure 14: Regression output for alcohol-involved vehicle fatality rate against economic
factors
11. Inference from figure 12-14:
● Yngdrv is significant for vehicle fatality rate, but is highly insignificant for the night time and
alcohol related fatality rates
● Fatality rate data includes values of all the age groups which is perhaps why Yngdrv is
significant only for vehicle fatality rate
● Spircons is significant for vehicle fatality rates, night time vehicle fatality rates and alcohol
related fatality rates. This result can be explained by the obvious fact that higher level of spirit
consumption will lead to higher fatality rate in the population
● By this point, it can be concluded that jaild and spircons are significant explanatory variables
with regards to vehicle fatality rates
Vehicle fatality rates and Socio-Economic Variables
Laws and taxes cannot be examined exclusively without factoring the interaction with socio-
economic variables. These variables are not directly related to drunk driving laws but including
these variables in the model will help explain the impact of jaild better in terms of fatality rate.
Figure 15: Regression output for vehicle fatality rate against economic factors
Figure 16: Regression output for night-time vehicle fatality rate against economic factors
12. Figure 17: Regression output for alcohol related vehicle fatality rate against economic
factors
Inference from Figure 15-17:
● Upon examining the result of the regressions, it can be concluded that there is no one specific
socio economic variable that is significant across the three regressions
● Perhaps, the fatality rates are getting affected by the interaction of several socio-economic
factors or that the socio-economic factors are more significant in an age group
Vehicle fatality rates across age groups and Socio-Economic variables
● Therefore, another aspect to consider is that there is a possibility that the socio-economic
variables could be significant for an age group
● Hence, the dependent variable to be used in the subsequent set of models will be vehicle fatality
rates across different age groups. The explanatory variables in each of the three models will be
the socio-economic variables discussed in the previous models.
Figure 18: Regression output for age-group 15 to 17 against socio-economic variables
13. Figure 19: Regression output for age-group 18 to 20 against socio-economic variables
Figure 20: Regression output for age-group 21 to 24 against socio-economic variables
Inference from figures 18-20:
● Although none of the socio-economic variables were significant in determining the vehicle
fatality rates, it is evident that a few of them are significant in determining the fatality across
specific age groups
● In Figures 16-18, unrate was significant in determining the fatality rates across specific age
groups
● Reiterating from the previous regressions it is evident that jaild was the only variable that
was significantly related to drunk driving laws
● Spircons which is indirectly related to drunk driving laws was also significant
● Upon using age groups, it is evident that unrate, was the only socio-economic variable that
was significant. Although this was not significant in the vehicle fatality regressions, it has
been significant in the age specific fatality rate regressions.
14. Which is more significant - spircons, jaild, or unrate?
Hence, the next step would be to include the above-mentioned variables as explanatory variables
and regress those against vehicle fatality rates.
Figure 21: Regression output of fatality rate against consumption, jail term and
unemployment rate
Figure 22: Regression output of night-time fatality rate against consumption, jail term and
unemployment rate
Figure 23: Regression output of alcohol-involved fatality rate against consumption, jail term
and unemployment rate
15. Inference from figures 21-23:
● It is evident that jaild is not significant, and comserd and unrate are significant
● Thus, if we were to explain the variation in the vehicle fatality rate, night time vehicle fatality
rate and alcohol related fatality rate, we can conclude that comserd and unrate are significant
So far, the above regressions have given a very high level understanding of the factors that affect
vehicle fatality rates across states.
Age group with the highest vehicle fatality rate
To have a better understanding of the factors, we need to understand which age group contributes
the most to the vehicle fatality rates. Further, we can understand which variables are significant for
that age group. If those variables are different from the ones that influence overall vehicle fatality
rates, it will provide a distinct perspective on the significance of the model.
The vehicle fatality rates are regressed against the fatality rates for each age group.
Obviously, it is expected that all three explanatory variables will be very significant, but the focus
here is on the coefficient. The coefficient will help us understand which age group contributes more
to the overall vehicle fatality rates.
Figure 24: Regression output of vehicle fatality rate against all age groups to check the
consumption, jail term and unemployment rate
16. Figure 25: Regression output of night-time vehicle fatality rate against all age groups to
check the consumption, jail term and unemployment rate
As expected, all the explanatory variables were significant, but mra2124 had the highest coefficient,
hence we can conclude that it contributes the most towards the overall vehicle fatality rate.
This outcome can be explained by the fact this age group comprises of people who are legally
permitted to drink. If people are permitted to drink the likeliness of them being involved in a fatal
vehicle accident increases
Do different variables affect fatality rates in the 21-24 age group?
Further, to get a sounder perspective on which variables impact the fatality rates regression
involving all the variables against the mortality rates for age group 21-24 is performed. The idea is
that if there are huge deviations from our previous inferences, it may provide new insights.
Figure 26: Regression output for age group 21-24 against all other relevant variables
The last regression output shows that jaild, spircons, dry, comserd, unrate are the significant
variables. Previously, we saw that jaild, spircons and unrate were significant, this regression shows
something along the same lines. Hence, it confirms that our inference was valid. Next, we will take
the significant variables (jaild, spircons and unrate) and regress those against the vehicle fatality
rates for 21-24 to further understand which of those variables significantly explain the variance in
the vehicle fatality rates for the 21-24 age groups. We are not including dry and comserd as these
variables never appeared to be significant in the previous regressions, hence it is not prudent to
include those in the analysis.
17. As with previous regressions, jaild is not very significant, but spircons and unrate are significant.
This is in line with our previous findings. Hence, we can conclude that the variables that
determine the vehicle fatality rates are same across age groups, of the three variables, jaild
proves to be the least significant.
Conclusion
Of all the variables that were analyzed, jaild, spircons and unrate are the variables that influence
vehicle fatality rates. Amongst the three variables, jaild proves to be the least influential.
Hence, we can conclude that variables that are directly related to the drunk driving laws do not
have a significant impact on the variation of vehicle fatality rates.
On the other hand, it is interesting to note that variables that are indirectly related to drunk
driving laws (spircons) and socio economic variables (unrate) are more significant in explaining
the variation in vehicle fatality rates.
Thus, to have an impact on vehicle fatality rates, law makers and other interested parties should
be focusing on the per capita alcohol consumption and rates of unemployment prevalent in
areas where the vehicle fatality rates are high.