Title | Homework 4 - Econ104 |
---|---|
Author | Matthias Boehm |
Course | Is There Truth in Numbers: The Role of Statistics in Economics |
Institution | University of California, Santa Cruz |
Pages | 7 |
File Size | 381.9 KB |
File Type | |
Total Downloads | 31 |
Total Views | 114 |
homework...
Matthias Boehm Carlos Dobkin Econ 104 02/28/2018 1. Not included are the variables drinks_alcohol and percentage_days_drink because these are not demographic variables. These are our outcome variables for the MLDA effect because we suspect there to be a jump in alcohol consumption. This regression only displays the balance of observable demographic variables. We also excluded the variable YRS_age since it is self explanatory that there is a change in age when comparing the two age groups. Running the regression of the remaining 11 variables leaves us with the following table:
(1)
(2)
(3)
(4)
VARIABLE S
HS_Diplo ma
hispani c
white
black
post
0.011
-0.010
0.016
-0.014
-0.018
0.008
(0.011)
(0.012)
(0.014)
(0.010)
(0.013)
(0.014)
0.023***
0.000
-0.007
0.009
(0.007)
(0.008)
(0.009)
(0.007)
(0.008)
(0.009)
-0.020**
-0.000
0.006
-0.005
-0.017
(0.010)
(0.011)
(0.013)
(0.009)
(0.012)
agec
agec_post
Constant
(5)
(6)
(7)
uninsur employe married ed d
(9)
working_l going_scho w ol
(10)
(11)
male
days_21
0.010
0.007
0.015
-0.000
(0.010)
(0.014)
(0.011)
(0.014)
(0.000)
0.026*** 0.059*** 0.051***
0.058***
-0.058***
(0.006)
(0.009)
(0.007)
-0.026**
0.002
-0.024**
0.019*
(0.012)
(0.009)
(0.012)
(0.010)
0.821***
0.241** 0.554** 0.157** 0.318*** 0.642*** 0.152*** * * *
0.642***
0.166***
(0.008)
(0.009)
(0.010)
(0.008)
(0.010)
(0.007)
(0.010)
(0.010)
0.030***
(8)
(0.007)
0.023**
365.000** *
(0.009) 0.029**
(0.000) 0.000*
(0.013) 0.428** *
(0.000) -0.000
(0.010)
The only difference we find is in the marital status. This makes sense because the older a person turns, the more likely it is for the person to get married. All other differences are not statistically significant. We can infer that there is a balance between the age groups of under and over 21 year olds. 2.
(0.000)
3.
4. Overall deaths: (1) VARIABLES
All causes of death
agec
0.827 (0.792)
post
6.876*** (1.332)
agec_post
-3.012** (1.157)
birthday
4.863** (2.402)
Constant
93.618***
(0.901)
Observations
48
R-squared
0.697
Subcategorized into causes: (1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
VARIABLES
alcohol
homicide
suicide
mva
drugs
internal
external
External other
agec
0.202**
0.795***
0.029
-2.568***
0.333***
1.618***
-0.791
0.488**
(0.079)
(0.211)
(0.279)
(0.456)
(0.118)
(0.370)
(0.774)
(0.225)
0.282**
-0.052
1.993***
4.149***
0.235
0.129
6.748***
0.333
(0.133)
(0.355)
(0.469)
(0.768)
(0.199)
(0.622)
(1.303)
(0.378)
-0.227*
-1.027***
-0.570
-0.873
0.031
0.162
-3.173***
-0.565*
(0.116)
(0.308)
(0.407)
(0.667)
(0.173)
(0.540)
(1.131)
(0.328)
1.004***
0.970
-1.230
2.380*
-0.118
1.628
3.235
0.671
(0.240)
(0.640)
(0.845)
(1.384)
(0.360)
(1.122)
(2.349)
(0.682)
1.428***
17.425** *
11.662***
29.929** *
3.899***
20.107***
73.511***
9.697***
(0.090)
(0.240)
(0.317)
(0.519)
(0.135)
(0.421)
(0.881)
(0.256)
Observatio ns
48
48
48
48
48
48
48
48
R-squared
0.673
0.407
0.512
0.740
0.720
0.809
0.522
0.359
post
agec_post
birthday
Constant
5. When looking at the first table of nr. 4, the value of the post variable suggests that the MLDA does reduce the death rate. Overall, the death rate of people who are between 21 and 23 years is 6.876% points higher than those who are between 19 and 21 of age. This
estimate is statistically significant at the 99% level. It is worth mentioning that the coefficient of the variable homicide shows a slightly negative value. This would suggest that the rate of deaths due to homicides actually decreases with people turning 21. However, the value is not statistically significant and should not be considered for a concrete conclusion. The coefficients of three specific causes that are depicted in the second table also show a positive value but are not statistically significant. These variables include Drugs, Internal Factors and Other External Factors. We can not conclude that these variables have an effect on the overall death rate due to the MLDA. 6. IV estimate of the effect of drinking on overall deaths = 79.9 Gives us Local Average Treatment Effect. IV estimate of the effect of drinking on death by Suicide = 23.152 IV estimate of the effect of drinking on death by Drugs = 2.734 IV estimate of the effect of drinking on death by Homicide = -.609 IV estimate of the effect of drinking on death by External Other = 3.866 IV estimate of the effect of drinking on death by MVA = 48.198 IV estimate of the effect of drinking on death by Alcohol Intoxication = 3.272 IV estimate of the effect of drinking on death by Internal Factors = 1.495 IV estimate of the effect of drinking on death by External Factors = 78.381 7. Standard Error of estimate of effect on overall deaths = 20.3 T-stat = 3.93 → statistically significant SE of the IV estimate for the effect of drinking on death by Suicide = 6.647 SE of the IV estimate for the effect of drinking on death by Drugs = 2.359 SE of the IV estimate for the effect of drinking on death by Homicide = 4.123 SE of the IV estimate for the effect of drinking on death by External Other = 4.439 SE of the IV estimate for the effect of drinking on death by MVA = 11.934 SE of the IV estimate for the effect of drinking on death by Alcohol Intoxication = 1.638 SE of the IV estimate for the effect of drinking on death by Internal Factors = 7.234 SE of the IV estimate for the effect of drinking on death by External Factors = 19.885
8. Y_i = Death rate Z_i = Turning 21 and being exposed to the policy change
D_i = Drinking alcohol Assumption 1: Z_i needs to be as good as randomly assigned This assumption holds because we are comparing the same population, only at different points in time. As we can see in number 1, the observable demographic characteristics show that the groups are very similar. The only possible objection could be the marital status that is slightly elevated in the treatment group. I expect the extent of this factor to be minimal. Assumption 2: Z_i must affect D_i This assumption holds as well because we have seen in the last homework assignment that there is a clear-cut increase in the alcohol consumption rate once people turn 21. We were able to compute that people who turn 21 years old are 8.05% points more likely to indicate that they drink alcohol. This is clearly due to the MLDA policy. Assumption 3: Z only affects Y through D If this assumption is also interpreted as: “Only Z affects Y through D” (‘only’ before Z), then the assumption is violated because a proportion of the population is already drinking before the MLDA and have an effect on the death rate. The assumption then requires the only difference between the treatment and control group to be the fraction treated. However, there is evidently a portion in the control group who drank alcohol before they were legally allowed to. People who were already drinking before age 21 are possibly not prone to changing their alcohol consumption behavior at all. This is also a reason why our IV estimate could be biased....