1
ECON 104
Assignment 4
In this assignment you will estimate the effect of the MLDA on arrests. The figures and
tables you create will be used when you write your second paper.
The dataset “NHIS.csv” is a sample drawn from the National Health Interview Sample
Adult Files 1997-2007. It includes the following variables: HS diploma, Hispanic, White,
Black, uninsured, employed, married, working lw, going school, male, days 21 (days to 21st
birthday, with negative values for days prior to turning 21), drinks alcohol (indivual reports
that they drink alcohol), AGE yrs, perc days drink (percent of days on which they report
drinking). Each row on the dataset is a record for an individual.
The file “Arrest.csv” contains arrest rates per 10,000 person years by age overall and
broken down by cause. It includes the following variables days to 21, all (all arrests),
drunk risk (drunk possibly risk to self), dui (driving under the influence), liquor laws (violation of liquor law), combined oth (disorderly conduct or vagrancy), robbery, ot assault
(simple assault), and aggravated assault (aggravated assault).
Please include all Stata code.
1. Create a table of regression estimates of the level of each of the demographic variables
for people just under 21 and how they change at 21 (Hint: Use your preferred regression discontinuity specification from Questions 4 and 5 of Assignment 3, but with
demographic variables as the outcomes). This table documents that people just under
21 are very similar to those just over 21 in terms of their observable characteristics.
2. Make a figure with the age profile of overall arrests.
3. Make figures of the age profile for each type of arrest. When possible, include the age
profile for more than one type of arrest in a single figure (i.e., you can group outcomes
in figures as long as it does not get too crowded). In particular, you may want to
2
groups arrests directly associated with drinking (e.g., DUI) in one graph and those
indirectly related to drinking (e.g., robbery) in a separate graph.
4. Make a regression table with estimates of the increase in overall arrests and by each
type of arrest. Put one regression in each column.
5. Does the MLDA (being under 21) reduce arrest rates? Are the reductions for the type
of arrests that are likely to be affected by alcohol consumption?
6. In a table, report the instrumental variables effect of drinking on overall arrests and
each cause. Use a two sample instrumental variables approach (reduced form/first
stage).
7. Do you believe that the MLDA is a valid instrument for estimating the effect of being
a drinker on arrests? Consider the two conditions that are necessary for an instrument
to be valid.