economics paper 10 pages(due asap)

1

 Changed wording in several areas to more accurately describe analyses process  Changed regressions to have robust standard errors  Added additional death subcategories table without birthday variable for comparison  Removed Contractions  Removed unneeded titles from some graphs  Corrected conclusion

2

Brandon Lang

Econ 104

Prof. Dobkin

3/6/18

The Minimum Legal Drinking Age and Young Adult Mortality

Abstract

This paper attempts to show how the presence of a minimum legal drinking age (MLDA) affects the

likelihood of a young adult drinking and the mortality rates of young adults. Through the combined use of

NHIS data with mortality data, a link between drinking and mortality rates is able to be drawn, and it can

be observed as to whether or not the MLDA has an effect on these rates. If there are differences found in

the likelihood of drinking and mortality rates between the group that is below the MLDA and the group

above the MLDA while randomizing the groups so as to attempt to eliminate bias, the effect of the MLDA

on whether or not a person drinks and mortality rates can be found. In analyzing the effect of turning 21

on the drinking rate and mortality rates, a regression discontinuity design is used.

Introduction

For many years, a concern of citizens and policy makers has been at what age should the U.S. MLDA be

set at. For most of the country historically, there was no legal minimum, however this changed after the

end of prohibition, with many states setting the minimum age at which people could consume alcohol at

21 years old. This was changed in 1971 with the adoption of the 26th Amendment, which prohibited states

and the federal government from using age as a reason for denying the right to vote to citizens who were

at least 18 years old, and many states reduced their MLDA to the same. This did not last, however, as the

Drinking Age Act of 1984 required states to start introducing laws to raise the legal purchasing age of

3

alcohol by citizens to 21. A major point of concern for determining at what age people should be given

the ability to consume alcohol legally is its effects on public health. We want to if the presence of the

MLDA reduces the proportion of the population that drinks and by how much does it affect the mortality

rate of young adults

To address these questions, this paper analyzes data from the National Health Interview Sample and

mortality data for 19 to 23 year olds. These data set are used in conjunction with each other to find the

probability of people drinking and how mortality rates are affected by the presence of a MLDA and effects

of becoming 21 years old. Two groups are created for this, with one being a treatment group consisting

of subjects younger than the MLDA and a control group consisting of subjects that are older who are not

affected by the MLDA. Data on the subjects’ characteristics from the NHIS is used, with the sample means

of both groups being found to not be statistically different. With the groups being shown to be adequately

similar, we can look at what differences the MLDA causes between the groups, if it does at all. If it is found

that the trends differ notably between the groups and that the MLDA is the cause of this difference, it can

be estimated as to how many more people are kept from drinking or dying due to the MLDA’s existence.

Through analyses using regression discontinuity design, it is found that there is a very notable increase in

the likelihood of alcohol consumption and mortality rates for subjects after the MLDA treatment becomes

active on them. In looking closely at the causes of death, we see that several of the causes of death for

the subjects which increase in rate are related to the consumption of alcohol. This analysis supports the

hypothesis that the MLDA does in fact decrease the probability of an individual to drink and decreases the

mortality rate for subjects who’s ages are below the MLDA, restricting their ability, even though not

entirely, to consuming alcohol.

Data

4

This paper uses data from the National Health Interview Sample Adult Files 1997-2007 (NHIS), compiled

by the Centers for Disease Control (CDC) and from the mortality data of 19 to 23 year olds using official

death certificates. The NHIS selects from an average sample of 36,600 households from the civilian

population and using questionnaires they call the Family Core, Sample Adult Core, and Sample Child Core,

with data from the Family Core being used. To obtain data for the Family Core, family members who were

at their household at the time of visit were questioned about sociodemographic characteristics, the health

of the family, and any injuries of family members. The NHIS sample contains data for over 61,000 people,

showing whether or not 18 to 29 year olds consume alcohol and the rates at which those who do. It

displays variables such as if they don’t have insurance, if they have a high school diploma, they’re Hispanic,

white, black, have employment, are married, working a low wage job, are attending school, they’re male,

the number of days it’s been since they turned 21, the percentage of days out of the year that they drink

alcohol, their age, and whether or not they drink alcohol. The mortality data shows the causes of death

for specific age profiles, these include death by alcohol consumption, homicide, suicide, motor vehicle

accidents, drug use, and other external causes, with all causes being categorized as either internal causes

of death or external causes of death.

Given that we are trying to analyze the effect of the MLDA on a group which is subject to it while

comparing them against a control group that isn’t with the two groups being similar in every way except

for whether they are under the MLDA, American subjects who are under the MLDA cannot be compared

to subjects in other countries who are not under the affect of an MLDA because there are too many

different factors that can be affected by the difference in geography and culture to provide unbiased data.

Because of this limitation, we need to be able to compare subjects who are just under the MLDA to

subjects that are just older than the MLDA and can drink legally.

Methods

5

Due to the sheer volume of subjects being observed, attempting to look at the trends that may be present

between the likelihood of individuals drinking and ages becomes very difficult as figures displaying this

are far too cluttered. To provide clear, readable, and ultimately meaningful results, the regressions had

to be designed in such a way that they analyzed the average ages instead of individual ages, reducing the

clutter of the produced figures substantially. Binwidth is used to determine the distribution of the age

profile of whether or not people drink alcohol. After trial and error, the bin width that I decided to use is

50 days because it has 21 as the midpoint, so the bin spreads from 21 but don’t cut across 21, separating

people by the number of months until or after their 21st birthday. I chose this bin width because it does

not average across the threshold so that people who are almost 21 are not in the same bin with people

that just turned 21 and the graph’s results are very clearly readable. Bandwidth is used to determine what

range of age should be used to spread out the points. After I determined that the range pf ages that should

be used is between 19 and 23, I chose to use a bandwidth of 2 years. I chose this bandwidth because

people younger than 19 and older than 23 are too far away from the MLDA so they likely aren’t similar

enough to be comparable, such as people under 19 still attending high school and people over 23 likely

being safer more experienced drinkers. The results also don’t average across the threshold, creating a

clear, centered jump with easily readable results. I then chose to use a range (Y) of likelihood of alcohol

consumption of 0.4-0.7. I chose this range because it presents the results very clearly and centers the very

noticeable jump at 21. The results of this are shown in Figure 1.

Finally, to help make the estimates as accurate as they can, quadratic regressions are introduced. In order

to increase the estimate’s accuracy as much as possible, new information has to be brought in to reduce

bias. To do this, new variables are created in Stata:

-agec: The difference of a subject’s age from the MLDA

-post: Set to 1 if a subject is older than the MLDA and 0 if younger; also called over21

-agec_post: Interaction variable for agec and post; also called over21_post

6

-age_50days: Width used to determine the bin subjects are placed in

-agec_sq: Quadratic form of agec

-agec_sq_post: Interaction variable for agec_sq and post; also called agec_sq_over21

-agec_50days_sq: Quadratic form for agec_50days

-agec_50days_cu: allows for third degree regression

To find the drinking rate age profile, the first stage relationship between the MLDA and whether or not

people drink alcohol must be found; this asks subjects “do you drink yes or no; how many drinks per week

do you have on average”. This is the proportion of people that began drinking at 21. To do this, a scatter

plot showing a regression discontinuity design will be created using the regressions:

𝐷𝑟𝑖𝑛𝑘𝑠_𝐴𝑙𝑐𝑜ℎ𝑜𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒_50𝑑𝑎𝑦 + 𝐵 𝑎𝑔𝑒_50𝑑𝑎𝑦_𝑐𝑢 + 𝜀 if age >= 19 & age <= 21

𝐷𝑟𝑖𝑛𝑘𝑠_𝐴𝑙𝑐𝑜ℎ𝑜𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒_50𝑑𝑎𝑦 + 𝐵 𝑎𝑔𝑒_50𝑑𝑎𝑦_𝑐𝑢 + 𝜀 if age >= 21 & age <= 23

We then need to find the effect that the MLDA has on the probability of subjects to consume alcohol.

Before it is possible to analyze the effect that the MLDA has on the mortality rate, it is important to

establish that it actually has an effect on the drinking rate which potentially affects the mortality rate in

many cases. This is found through a table made with the regressions:

𝐷𝑟𝑖𝑛𝑘𝑠_𝐴𝑙𝑐𝑜ℎ𝑜𝑙 = 𝐵 + 𝐵 𝑜𝑣𝑒𝑟21 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑜𝑣𝑒𝑟21 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀 if age >= 21 & age <= 23

𝐷𝑟𝑖𝑛𝑘𝑠_𝐴𝑙𝑐𝑜ℎ𝑜𝑙 = 𝐵 + 𝐵 𝑜𝑣𝑒𝑟21 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑜𝑣𝑒𝑟21 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑜𝑣𝑒𝑟21 + 𝜀 if age >= 21 & age <= 23

𝐷𝑟𝑖𝑛𝑘𝑠_𝐴𝑙𝑐𝑜ℎ𝑜𝑙 = 𝐵 + 𝐵 𝑜𝑣𝑒𝑟21 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑜𝑣𝑒𝑟21 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑜𝑣𝑒𝑟21 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑜𝑣𝑒𝑟21 + 𝜀 if age >= 21 & age <= 23

We will then need to check our balances and see if the MLDA has any statistically significant effect on

any of the sample characteristics from the NHIS data. To do this, a regression table will be created. The

regressions used for this include:

7

𝑈𝑛𝑖𝑛𝑠𝑢𝑟𝑒𝑑 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

𝐻𝑠_𝑑𝑖𝑝𝑙𝑜𝑚𝑎 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

𝐻𝑖𝑠𝑝𝑎𝑛𝑖𝑐 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

𝑊ℎ𝑖𝑡𝑒 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

𝐵𝑙𝑎𝑐𝑘 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑑 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

𝑀𝑎𝑟𝑟𝑖𝑒𝑑 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

𝑊𝑜𝑟𝑘𝑖𝑛𝑔_𝑙𝑤 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

𝐺𝑜𝑖𝑛𝑔_𝑆𝑐ℎ𝑜𝑜𝑙 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

𝑀𝑎𝑙𝑒 = 𝐵 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝜀 if age >= 19 & age <= 23

Now we can look at how all causes of death affect different age groups. This is done by creating a scatter

plot diagram in which each individual point represents a group of subjects that are of similar ages in so

that they are the same number of months away from turning or having turned 21, making a much more

readable chart. The regressions used to give the results for this are:

𝐴𝑙𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝜀 if age >= 19 & age < 21

𝐴𝑙𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝜀 if age >= 21 & age < 23

Then once we are able to see what the trends are for mortality rates and age groups, it is necessary to

look deeper by analyzing specifically how fatalities by alcohol and motor vehicle accident are affected.

To do this, we produce a scatter plot and use these regressions:

𝑀𝑉𝐴 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝜀 if age >= 19 & age < 21

𝑀𝑉𝐴 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝜀 if age >= 21 & age < 23

𝐴𝑙𝑐𝑜ℎ𝑜𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝜀 if age >= 19 & age < 21

𝐴𝑙𝑐𝑜ℎ𝑜𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝜀 if age >= 21 & age < 23

8

Finally, we can find the effects of age and the MLDA on the different mortality rate categories and

groups. To do this we use multiple regressions:

𝐴𝑙𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑝𝑜𝑠𝑡 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀

𝐼𝑛𝑡𝑒𝑟𝑛𝑎𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑝𝑜𝑠𝑡 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀

𝐸𝑥𝑡𝑒𝑟𝑛𝑎𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑝𝑜𝑠𝑡 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀

𝐻𝑜𝑚𝑖𝑐𝑖𝑑𝑒 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑝𝑜𝑠𝑡 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀

𝑆𝑢𝑖𝑐𝑖𝑑𝑒 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑝𝑜𝑠𝑡 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀

𝐷𝑟𝑢𝑔𝑠 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑝𝑜𝑠𝑡 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀

𝑀𝑉𝐴 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑝𝑜𝑠𝑡 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀

𝐴𝑙𝑐𝑜ℎ𝑜𝑙 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑝𝑜𝑠𝑡 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀

𝐸𝑥𝑡𝑒𝑟𝑛𝑎𝑙𝑜𝑡ℎ𝑒𝑟 = 𝐵 + 𝐵 𝑎𝑔𝑒𝑐 + 𝐵 𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢 + 𝐵 𝑎𝑔𝑒𝑐_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑠𝑞_𝑝𝑜𝑠𝑡 + 𝐵 𝑎𝑔𝑒𝑐_𝑐𝑢_𝑝𝑜𝑠𝑡 + 𝐵 𝑏𝑖𝑟𝑡ℎ𝑑𝑎𝑦 + 𝜀

We can then attempt to find how the MLDA affects the mortality rate specifically through drinking as a an

instrumental variable (IV). To do this, the reduced form must be found for the mortality categories. The

results of the reduced form and the first stage are then divided by each other to produce the instrumental

variable (IV) estimate. To confirm the finding of the IV estimate that is found through this method, an IV

regression is used.

I used robust standard errors for all of my regressions.

Results

9

Listed first are some of the figures that resulted from the trial and error process to determine the right

parameters. Using the age range of 19-23 for subjects and grouping them by the number of months they

have until or after their 21st birthday provides clear and easily understandable results.

Figure 1: Scatter Plot Tests

It appears that a bin width of 50 days allows for a good balance of minimizing clutter for ease of reading

and showing enough results so that trends in the data can be identified. Using these parameters, we can

move on to look at the effect of regressing the probability of subjects consuming alcohol with age.

10

Figure 2: Drinking Age First Stage Relationship

This provides a first stage relationship between the MLDA and whether or not people drink alcohol. It

shows that a large jump in subjects’ probability of drinking occurs at the age of 21 when they are no longer

below the MLDA.

11

Table 1: Changes in Drinking with Birthday Variable

(1) (2) (3) VARIABLES drinks_alcohol drinks_alcohol drinks_alcohol over21 0.088 0.111 0.115

(0.083) (0.083) (0.085) agec 0.044*** -0.024 -0.051

(0.009) (0.037) (0.093) o.agec_over21 - - -

o.birthday - - -

agec_sq -0.034* -0.068

(0.018) (0.108) agec_cu -0.011

(0.035) o.agec_sq_over21 - -

o.agec_cu_over21 -

Constant 0.559*** 0.536*** 0.532***

(0.010) (0.016) (0.021)

Observations 8,933 8,933 8,933 R-squared 0.003 0.003 0.003

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

This tables shows us that when subjects are aged above the MLDA, the probability that they will consume

alcohol increases by 11.5 percentage points. This supports the hypothesis that the MLDA has a negative

effect on the probability of people below the age of 21 consuming alcohol. With these results providing

assurance that the MLDA does in fact have a negative effect on the likelihood of minors drinking, the

effect of the MLDA on mortality rates can be observed.

12

Table 2: Sample Characteristics Test

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLES uninsured hs_diploma hispanic white black employed married working_lw going_school male post -0.018 0.011 -0.010 0.016 -0.014 0.008 -0.030*** 0.010 0.007 0.015

(0.013) (0.011) (0.012) (0.014) (0.010) (0.014) (0.010) (0.014) (0.011) (0.014) agec 0.026*** 0.023*** 0.000 -0.007 0.009 0.059*** 0.051*** 0.058*** -0.058*** -0.023**

(0.008) (0.007) (0.008) (0.009) (0.007) (0.009) (0.006) (0.009) (0.007) (0.009) agec_post -0.017 -0.020** -0.000 0.006 -0.005 -0.026** 0.002 -0.024** 0.019* 0.029**

(0.012) (0.010) (0.011) (0.013) (0.009) (0.012) (0.009) (0.012) (0.010) (0.013) Constant 0.318*** 0.821*** 0.241*** 0.554*** 0.157*** 0.642*** 0.152*** 0.642*** 0.166*** 0.428***

(0.010) (0.008) (0.009) (0.010) (0.007) (0.010) (0.007) (0.010) (0.008) (0.010)

Observations 18,824 18,824 18,824 18,824 18,824 18,824 18,824 18,824 18,824 18,824 R-squared 0.001 0.003 0.000 0.000 0.000 0.014 0.018 0.014 0.019 0.000

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

This table shows us that the post variable—indicating if a subject is over the age of 21 and is in the

treatment group—is not statistically significant for any of the subject characteristics with the only

exception being the marriage variable. Given that the average American adult marries at about 28 years

of age, the effect of being older than 21 being statistically significant with regards to marriage is consistent

as post-21-year-olds are closer to the average marriage age than pre-21-year-olds. This should not be a

cause for bias in this analyses, so the characteristics are shown to be similar enough.

13

Figure 3

This graph shows the mortality rate by all causes per 100,000 people for subjects within the age range of

19 and 23, with each individual point representing a group of subjects that are of similar ages in so that

they are the same number of months away from turning or having turned 21. This chart indicates that the

number of fatalities faced by subjects rises very notably once the age of 21 is reached, with the trend in

mortality diminishing as subjects become older. This means that there must be something happening at

and after the age of 21 to cause this jump.

14

Figure 4

This chart breaks down the mortality categories to isolate the observations for fatalities related to motor

vehicle accidents (MVA) and alcohol. It shows that there is a sharp increase in both mortality rates, with

both trends also diminishing as the age of the groups increases. Given that death by alcohol consumption

and MVA are directly or at least typically related to drinking and that the jump is occurring at the age of

21, this indicates that the MLDA certainly is having some kid of effect on the treated group.

15

Table 3: Effects of drinking on subcategories of deaths

(1) (2) (3) (4) (5) (6) (7) (8) (9) VARIABLES all internal external homicide suicide drugs mva alcohol externalother agec -5.634 -3.706 -1.929 1.158 1.336 0.895 -5.500 0.679 -0.020

(8.154) (3.482) (7.120) (2.127) (1.745) (1.267) (4.247) (0.592) (1.835) post 8.720*** 1.962 6.757*** -0.878 2.834** -0.084 3.905** 0.470** 0.746

(2.904) (1.421) (2.360) (0.667) (1.336) (0.345) (1.485) (0.230) (0.488) agec_sq -6.922 -6.650 -0.272 0.060 1.572 0.669 -3.435 1.005 0.468

(9.933) (4.032) (9.303) (2.532) (2.218) (1.610) (5.445) (0.631) (2.602) agec_cu -2.055 -2.227 0.172 -0.070 0.513 0.216 -1.098 0.430** 0.409

(3.290) (1.345) (3.207) (0.835) (0.747) (0.561) (1.862) (0.197) (0.924) agec_post 2.378 1.983 0.395 2.232 -6.254 0.950 4.557 -1.779 -0.078

(12.430) (5.727) (12.097) (3.206) (5.129) (1.746) (6.997) (1.192) (2.738) agec_sq_post 6.995 9.836 -2.841 -4.543 2.792 -2.751 1.666 0.407 -0.796

(13.796) (6.333) (13.892) (3.776) (5.410) (2.142) (8.034) (1.428) (3.388) agec_cu_post 2.305 1.405 0.900 1.580 -1.753 0.540 1.405 -0.924* -0.221

(4.376) (2.034) (4.442) (1.225) (1.670) (0.732) (2.573) (0.474) (1.143) o.agec_post - - - - - - - -

birthday 4.403** 0.832 3.571** 1.508*** -2.133* 0.042 3.068*** 0.865*** 0.583*

(2.159) (1.029) (1.652) (0.374) (1.121) (0.214) (1.060) (0.136) (0.342) Constant 92.279*** 19.208*** 73.071*** 17.571*** 11.896*** 4.001*** 29.385*** 1.421*** 9.373***

(1.495) (0.778) (1.247) (0.483) (0.315) (0.229) (0.795) (0.151) (0.258)

Observations 48 48 48 48 48 48 48 48 48 R-squared 0.703 0.823 0.526 0.447 0.534 0.740 0.747 0.727 0.408

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

These results indicate that we can expect alcohol to increase the mortality rate from alcohol, motor

vehicle accidents, and suicide. Ease of access to alcohol increasing deaths from alcohol consumption when

not being under the age of the MLDA is pretty direct, but for MVA and suicide, the correlation isn’t quite

as direct but can still be made. With nearly one third of MVA’s being a result of alcohol-impaired driving

16

crashes, it makes sense that the mortality rate from MVA’s should increase once people have access to

alcohol. In the case of suicide, alcohol is considered to be a depressant which can limit a person’s ability

to think rationally, and when a person is considering suicide this can affect their decision making process

in the worst way.

Table 4: Effects of drinking on subcategories of deaths without birthday

(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES all internal external homicide suicide drugs mva alcohol agec -5.634 -3.706 -1.929 1.158 1.336 0.895 -5.500 0.679

(8.052) (3.438) (7.030) (2.101) (1.723) (1.251) (4.194) (0.585) post 11.263*** 2.443** 8.820*** -0.007 1.602 -0.060 5.678*** 0.970***

(2.529) (1.029) (2.067) (0.739) (1.082) (0.262) (1.565) (0.353) agec_sq -6.922 -6.650 -0.272 0.060 1.572 0.669 -3.435 1.005

(9.808) (3.982) (9.186) (2.500) (2.190) (1.590) (5.376) (0.623) agec_cu -2.055 -2.227 0.172 -0.070 0.513 0.216 -1.098 0.430**

(3.249) (1.328) (3.167) (0.824) (0.738) (0.554) (1.839) (0.194) agec_post -6.790 0.250 -7.040 -0.908 -1.814 0.863 -1.832 -3.580**

(11.262) (4.559) (10.692) (3.267) (4.394) (1.493) (6.952) (1.453) agec_sq_post 16.047 11.547** 4.500 -1.443 -1.593 -2.665 7.974 2.185

(12.869) (5.380) (12.645) (3.787) (4.894) (1.920) (7.963) (1.630) agec_cu_post -0.331 0.907 -1.238 0.678 -0.476 0.515 -0.431 -1.441***

(4.158) (1.800) (4.127) (1.227) (1.575) (0.673) (2.566) (0.529) o.agec_post - - - - - - -

Constant 92.279*** 19.208*** 73.071*** 17.571*** 11.896*** 4.001*** 29.385*** 1.421***

(1.476) (0.769) (1.231) (0.477) (0.312) (0.226) (0.785) (0.149)

Observations 48 48 48 48 48 48 48 48 R-squared 0.689 0.821 0.511 0.402 0.491 0.740 0.729 0.668

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Looking at Table 3 and Table 4 we see that being over 21 is a statistically significant factor for mortality

due to all, external, suicide, MVA, alcohol and external other causes. When we adjust for the inclusion

subjects’ birthdays the coefficients on all significant categories decrease, excluding suicides. This means

that the information supplied by Table 3 is an over estimate that doesn’t consider the fact that when

people turn 21 they are likely to overly take advantage of their newly legal abilities were illegal for them

until their 21st birthday.

17

The IV regression resulted in an estimate of 79.875512 with a standard error of 19.246303

Conclusion

The presence of the MLDA does reduce the proportion of people who drink, having an effect of

approximately 11.5 percentage points on the drinking rate. The MLDA also appears to reduce the death

rate. Categories such as internal cause of death and drugs are outside of the effects of the MLDA as the

MLDA does not stop 19-year olds from catching diseases which makes them sick or obtaining drugs

illegally off the streets. Categories such as Alcohol, MVA, and Suicide are or can be related to alcohol

consumption, so the enforcement of a MLDA helps to prevent the deaths of young adults under the age

of 21 by restricting access to alcohol. As soon as it’s legal for an individual to drink, we see immediate

spikes in the overall death rate.

Given the IV estimate that was found of approximately 80, this means that once the MLDA is no longer in

effect on subjects, the consumption of alcohol causes the death rate to increase by nearly 80 per 100,000.

This supports the claim that the MLDA does reduce the mortality rate of individuals who are aged under

the MLDA. However, when determining if a variable can be considered instrumental, there are three

assumptions that must be met. Those assumptions are the exclusion restriction, relevance condition, and

the exogeneity assumption. The MLDA affects our outcome on deaths through restricting access to alcohol

for people under the age of 21, so it has a direct effect on the outcome due to it restricting one group

from accessing alcohol, failing to meet the exclusion restriction. The MLDA meets the relevance condition

given that, despite the fact it legally restricts people from obtaining alcohol, it cannot stop them from

consuming alcohol through illegal means. Finally, the MLDA does meet the exogeneity assumption given

that the groups are shown to be randomized, even if the MLDA only affects people under the age of 21

18

while having no effect on those older, meaning that people in one age group can’t be randomly assigned

to be in the other. This leads to the conclusion that MLDA is an inadequate instrument.