Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Answer 1

Task 1

Variable assigned_small_class = 1 if the student was assigned to small class

(1)

count mean
size of class 1758 15.64903
Observations 1758
assigned_small_class= 1

(1)

count mean
size of class 3659 23.74064
Observations 3659
assigned_small_class=0
The average number of students in a class for a student assigned to a small class is
approximately 16, while the average for a student who was not assigned to a small
class is approx. 24. We can see from the graph that those assigned to small class are
actually in smaller classes. The majority students assigned to small class have class
size around 15, with maximum of 19 and a minimum of 11. On the other hand, the
majority of students not assigned to small class have size around 21-25 have a
minimum of 16 and maximum of 29.

This has proved that the students assigned to smaller class end up in smaller class and
those not assigned to small class end up in the larger classes. The size of the class can
also determine the good performance and smaller classes are expected to perform
better. The t-test on the maths tests between those assigned in small small and those
who are not is significant with the ttest of -6.39; these differences are statistically
signification at least at the 1% level.

It is worth mentioning that using a binary variable, which is based on the


samples “small class” or “large class” will create a grey area. Further explained, if
a limit is set at 25 students and a student is assigned to a class with 24 students
the mean will be affected, if there is a correlation, as this would still count as a
“small class”.
One solution to this problem could be to create limits that are separated by a
larger number. For instance, if we define the limits for a small and large class to
be (small class ≤ 15 students) and (large class ≥ 25)

Task 2

Estimation of the effect of class size on math test scores


(1) OLS (2) 1st stage (3) 2nd stage

class_size -0.928*** -0.694***


(0.115) (0.133)

Male -0.395 -0.0579 -0.413


(1.004) (0.0582) (1.004)

white 14.07*** -1.272*** 14.41***


(1.479) (0.0851) (1.534)

free_lunch -17.57*** -0.413*** -17.46***


(1.152) (0.0666) (1.167)

days_absent -0.600*** 0.00224 -0.606***


(0.0838) (0.00485) (0.0856)

inner_city_school -1.283 -1.424*** -0.958


(1.749) (0.101) (1.704)

assigned_small_clas
-8.087***
s
(0.0621)

_cons 642.0*** 25.07*** 636.8***


(3.040) (0.0948) (3.475)
N 5417 5417 5417
Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001

Using IV estimation, a reduction in the maths test score by 0.69 points can be seen
with the addition of one more student. This effect is statistically significant at 1%.
The difference in OLS and IV results can be seen. The OLS estimates shows that the
effect of class size on math test score is -0.93 whereas the IV estimates shows -0.69.
There seems to be a smaller effect of class size on math test score when using IV
estimation compared to OLS estimation.

Task 3
In order for the IV-estimation to work we need the proposed instrument (small class)
to only affect class_size, and not the dependent variable math_test_score. There are
two different assumptions in order for IV estimation to hold:
1) Exogenous
We need the instrument to be exogenous to the structural equation. This means
two things:
- The instrument has no direct partial effect on the dependent variable. Only
indirect via the variable that is “instrumented”. This means that whether a
student is placed in a small class or not should not affect the math test
score directly.
- The instrument should not correlate with the error term u.
This is the most difficult assumption, as it is fundamentally untestable, due to
the fact that we can not observe u.
However we can use our logical sense and discuss whether it might seem to be
exogenous or not. There might be many explanations to how the variable
assigned_small_class might correlate with the error term. For instance we can
imagine that some geographical areas have more or less amount of smaller
classes. For example you could argue that in Oslo the classes are smaller (due
to for example more schools and more access to teachers). Which means that
in Oslo there will be a greater chance for the assigned_small_class to be 1.
However, we might observe that the students in Oslo score better(or worse for
that matter) on the math test, due to other explenations than small or big
classes, but maybe due to culture, norms or access to private teaching. In this
case, the instrument is correlated to “location” (which is a part of the error-
term), and location is affecting the math test score. If this is true, then the
instrument is not exogenous. But these are only discussions and we should not
strictly conclude whether the instrument really is exogenous or not.

2) Relevant
We need the instrument to have some effect on the variable we are trying to
instrument for. In other words: class_size has got to be correlated with
assigned_small_class. From the regression output we can clearly see that this
is the case, as the correlation between the two is highly statistical significant.
It signalisez that if a student is assign to a small class, he/she is more likely to
have a smaller class size, which makes perfect sense. This condition could
also have been verified by common sense, as these two variables are totally
dependent.

Answer 2:

Task 1

It is natural to think that institutional strength is very much correlated with the error
term, and therefore we will have problems with endogeneity. There are many
examples, and here is one. For example, institutional strength could be argued to
correlate with whether the country is a democracy or not, this in term will also affect
the GDP. The variable “institutional_strength is hard to measure exact and
numerically, and may therefor correlate with a lot of other variables.

Task 2

Task 3
In the plot we can see a cleat correlation between early mortality and institutional
strange. The correlation is negative, which means the either the early mortality rates
get, the less institutional strength the country will have.

Task 4

(1) (2) (3)


institutional_stren
logGDP (OLS) logGDP (2SLS)
gth (1st stage)
institutional_streng
0.0517*** 0.0922***
th
(0.00608) (0.0152)

log_early_mortality
-6.151***
_rates
(1.299)

_cons 4.679*** 93.87*** 2.043*


(0.405) (6.296) (0.994)
N 63 63 63
Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
The result using OLS method has a coefficient of 5.17% whereas the coefficient
using 2SLS method is 9.22%

Task 5

(1)
logGDP
institutional_s 0.0922***
trength
(0.0152)

_cons 2.043*
(0.994)
N 63
Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001

Endogeneity problem occurs when there is a correlation between independent


variable and the error term in the model. The results from the IV estimation are
similar to the 2SLS estimation. Furthermore, Hausman test have been used to
test if the institutional_strength is endogenous. The null hypothesis explains that
there is no correlation between two variables and the result of p-value in this
case which is 0.000 less than p-value we can reject the null hypothesis.

Task 6

(1)
logGDP
institutional_ 0.0893***
strength
(0.0126)

_cons 2.232**
(0.827)
N 63
Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001

To conduct the over identification test, we have null hypothesis: all instruments
exogenous and alternative hypothesis: at least one instrument is not exogenous.

You might also like