Sie sind auf Seite 1von 5

ECON212 Basic Econometrics 1

Semester 1 2014,15

Feedback document relating to Mid-Semester test held 28 November 2014


The marks for the Mid-Semester test should now be available. The purpose of this
document is to provide students with some feedback from the exercise. First, some
summary statistical evidence on general performance is provided. Later some specific
issues are addressed.
The histogram presented below in Figure 1 provides a summary of the performance of
the 484 candidates. It is evident from this graph that the general performance was good,
a feature that I was very pleased to note. The modal group gained a mark between 60
and 70 and marks were distributed over the full range.

Figure 1 Histogram of Final student marks

FINAL
120

100

Frequency

80

60

40

20

0
0

10

20

30

40

50
1

60

70

80

90

100

Some summary statistics of the marks are given in Table 1

Table 1 Summary statistics


Mean
Median
Maximum
Minimum
Std. Dev.
Skewness
Kurtosis

63.46694
66.00000
96.00000
6.000000
17.14935
-0.715386
3.176995

The percentage of students failing this part of the assessment for ECON212 is 9.3%.
This equates to a total of 45 failing candidates. Since I do not think that the MidSemester multiple choice test was particularly difficult, each of these 45 students should
regard themselves as having got an early warning that they may be at risk of failing this
course. In particular, they should consider examining their work and study patterns in
conjunction with the course with a view to improving learning outcomes.
As you know, feedback was sought from students through Task 4.1 in
Workshop/Tutorial 4. Each tutor has written a short report relaying your main thoughts
to me as Module Leader responsible for this report. Much of what follows is a synthesis
of your opinions, or my reaction to them.
The main issue that arose relates to Question 4, so lets get that out of the way first! It
related to a question asking you to evaluate a (conditional) variance. Unfortunately,
there was a transcription error that led to none of the offered answers being correct. I
am very sorry about this and apologize for the inconvenience that I know this must have
caused you all. The answer I intended to be correct was answer (b); this was given as
2.49 when it should have been 2.52 (2.49 was comfortably the closest to 2.52 of those
available).
Despite this small error, 68% of students got the answer right and hence received full
credit. But, given that this was clearly my error, I feel a strong need not to penalize
students for this mistake. After some thought, what I decided was a fair thing to do was
to credit all 32% of students who got the choice wrong with an extra 2 marks. So, I
went through every script manually to determine who these 32% of people were and
adjusted marks accordingly. In the normal run of events, all students would have
received a mark that was a multiple of 4. Some students now have a mark that does not
divide by 4 (but is still an even number); if your mark is one of those, you know that you
did not choose (b) for Question 4 and have hence received the 2 extra marks. My
thinking here is that nobody who got the question wrong can score as highly as
2

someone who got it right but, again, nobody gets no marks at all because the
mistake.is mine.
One of the other main comments often followed on from this and it relates to the time
allowed for the Test and the number of items (30) it was comprised of. Some students
felt they were under undue time pressure. I also considered this issue carefully and
scrutinized every script to assess the evidence for this. I found that, of 484 scripts, only
17 failed to provide an answer to all 30 questions. Of these, 12 only failed to answer 1
question and nobody left more than 4 questions unanswered; the exhortation to answer
all questions seems to me to have been heeded. Finally, of the total of 28 missing
answers in all scripts, only 9 were from the last 10 questions (and 3 by one candidate
alone); this does not suggest undue time pressure towards the end of the paper.
Overall, I do not really see much evidence in the light of this investigation (and the
generally high level of the marks given in Figure 1) that there is any strong basis for a
claim of lack of time allowed for completing the Test.
Other comments related to the availability of past papers and more specimen
questions. By their very nature an examiner will typically have a bank of questions that
can be used more than once. If students are confronted in their Test by questions they
have seen before, it is hardly a test, more of a rote learning exercise and I do not think
that this is the purpose of education at university. Moreover, in 2013 a MCQ test was
also used and you do not have access to the questions set then, do you? I think that the
8 or 10 Specimen Questions that I set were sufficient for your purposes and indicated to
you that questions would be asked from all 8 of the lectures that you had listened to by
the time of the Test. But I am receptive to your comments and will try to make at least
some more example questions available to your successors in future years.
A difficulty from the point of view of the setter of the test is to try to find questions that
are neither too trivial nor too difficult. On both counts I will typically err! Some questions
were evidently too easy (almost everybody got them right) and I shall either replace
them or make them less straightforward in future tests. On the other hand, there were
some questions that erred in the other direction and I think it provides you with useful
feedback to discuss them here, since it may aid your future study and revision.
On 3 of the 30 questions, less than 30% of students got them right. These were
Questions 9, 21 & 30; they are now considered in some detail.

Question 9.
Data on the income of law graduates collected at different times during the same year is:

(a) panel data


(b) experimental data
(c) time series data
(d) cross-section data
Feedback
The answer is (d), but the commonest answer given was (c). This is not time series data
because just because the data was not all collected at one time does not mean that you
could use the chronology of the data as an aid in analyzing it. Quite apart from anything
else, we have not studied time series data at any point in the course! The data is crosssection in nature because a slice of data about a population of law graduates is being
collected. Over a single year that population will not change very much.
Question 21.
In the model
standard error,

the OLS estimator of the slope coefficient

, that uses the true error variance

has a true

and an estimated standard error,

, that uses an unbiased estimate of the error variance. Which of the following is true as
a general proposition?

(a)
(b)
(c)
(d) none of the above
Feedback.
The answer is (d), but the commonest answers chosen were (a) and (c), each nominated by
about 30% of students. But the answer in None of the above because
is a random variable.
Hence in any given application you can never know if the estimated error variance is larger or
smaller than the true value. Moreover, the probability of them being equal is zero (because the
random variable is continuous).

Question 30.
You wish to conduct an hypothesis test of the null hypothesis that the true coefficient on the
variable BRRA against the one-sided alternative that it is positive and are required to determine
the p-value (Prob-value) for the test. Which of the following is correct (if none of them, choose
(d))?

(a) 0.00085
(b) 0.00000
(c) 0.00170
(d) none of the above
Feedback.
The correct answer is (a) but the most popular answer chosen was (c). This is because the 2sided p-value is 0.0017 (which would have been (c)). But the alternative hypothesis is one-sided
(positive), so a 1-sided p-value is required and is therefore equal to 0.00085..
I hope that you find this feedback document useful. Please feel free to contact me if you have
any further major questions or comments that you do not think I have dealt with.

Andy Tremayne

9 December 2014

Das könnte Ihnen auch gefallen