Dissertation milestones timetable Semester 2

Milestone Completion of Dissertation

target times
This will take  Extension of Introduction chapter to further develop the general focus /
the whole of motivation of dissertation that should demonstrate a coherent
the Semester economics-based approach to your chosen topic. The research
to develop. question(s) to be investigated should also be clearly explained in this
section. [approximately 750 words]

If you did  Background / historical overview chapter to provide the reader a

not write this greater perspective of your topic. [approximately 1000-1500 words]
in semester
1, then you  This might not be appropriate for all dissertations, but if you want to
should treat discuss the content / structure of this chapter before starting to write it
this as an then email me a plan so I can see the proposed balance of the chapter.
assignment
and so this
should take
2-3 weeks.

Depending  Methodology chapter to further develop the method and data to be

how well you used in your analysis. In particular, this should include the justification of
completed the empirical techniques that you intend to undertake through outlining
this section
their advantages / how they will resolve your research question(s), this
of your
Semester 1 should come from the previous chapters on Economic Theory and the
Research Contextual/Empirical Literature Review. [approximately 750 words]
Outline, then
this should  This should contain the following four points. Numbers 1-2 and 4 can be
take you less written into the Methodology chapter, while number 3 can be an
than 2 appendix to your dissertation or in this chapter.
weeks.
1. Produce a formal model (i.e. regression equation) or empirically based
However, explanation of your research question(s). For example, this might be:
even if this
section was Regression model specifications:
done well in (1) BG = B0 + B1BG(t-1) + B2INT(t-1) + B3GDP + B4INF + B5UN + B6ELE + B7LEFT +
the Semester
1 Research B8FRAG + B9COAG + B10MINGOV + B11MAT + B12EMU
Outline, you
will still need
to complete (2) PD = B0 + B1PD(t-1) + B2INT(t-1) + B3GDP + B4INF + B5GAP + B6ELE + B7LEFT +
numbers 3 B8FRAG + B9COAG + B10MINGOV + B11MAT + B12EMU

1
and 4. Dependent variables:
PD = Government Debt to GDP
Once you BG = Budget Balance to GDP
have your
data in an Independent Variables:
Excel INT = Long Term Interest Rate
spreadsheet, GDP = GDP Growth
send this to INF = Inflation
me with a UN = Unemployment rate
full list of the ELE = Election Year Dummy
variables LEFT = % of Left Cabinet Members
(number 2) FRAG = Fractionalisation Index
so I can COAG = Coalition Government Dummy
check it looks MINGOV = Minority Government Dummy
okay. MAT = Dummy for Maastricht Treaty (TEU)
EMU = Dummy for EMU Accession
In particular
if doing time- A priori expectations:
series
Model (1)
analysis then
you must B2 B5 B6 B7 B9 B10 <0
ensure you
B1 B3 B4 B8 B11 B12 >0
have data in
either real Model (2)
terms, or in
B3 B4 B8 B11 B12 <0
proportional
terms (e.g. as B1 B2 B5 B6 B7 B9 B10 >0
a % figure).

2. This chapter should include a matrix / table as suggested below for all
the variables in your model:

Variable Description Variable A priori Data Citations

name (and construction impact source supporting
abbreviation) inclusion
of variable
Government PD captures The ratio of Positive IMF: A Alesina
Debt as % of the higher domestic Historical and Perotti
GDP (PD) risk of government Public (1995a)
holding debt to Debt
domestic GDP. Database
assets
associated
with higher
level of
indebtedness.
Therefore, an
increase in

2
 PD leads to
depreciation
of the RER.

3. Collect data and correctly arrange it (i.e. in columns) in a spreadsheet.

For example:

Year PD BG INT GDP INF

1980 78.5 -2.8 13.4 1.4 2.7

4. Listing and completion of any data transformation (e.g. current to real

values; calculation of rate of change etc.) required saved to initial
spreadsheet.

Once we  Data analysis / results chapter where the specifics will differ for each of
have agreed you in relation to the type of dissertation you are undertaking and how
on your data confident you are on data analysis, but the emphasis of this chapter is to
(numbers 3 show the analytical process that you decide to follow. As indicated
and 4 below, this will probably range from simple building blocks of graphs and
above), then descriptive statistics through to measures of association through
you can start multiple regression. [approximately 1500-2000 words]
the data
analysis.  You will need to install SPSS for your laptop/PC see:
https://unibradfordac.sharepoint.com/sites/it-services-intranet/Shared
I would %20Documents/spss.pdf
suggest you
send me (as  Graphical representation of data:
a Word file
o Line graphs to show trends
by cutting
o X-Y graphs to show relationship between the two variables
and pasting
o Disaggregate the data into key periods (e.g. due to change in
from your
country circumstances) and compare / contrast across these
SPSS output
different time periods
file) the
o Some or all of these might go into your Introduction and/or
results step-
Background chapters
by-step to
review.

 Descriptive statistics:
o Produce descriptive statistics (i.e. mean, minimum, maximum
and standard deviation) of variables across the whole sample or

3
 in the disaggregated time periods/countries etc for comparison if
appropriate

 Measures of association:
o Correlation shows the association between a pair of variables in
terms of whether it is positive (i.e. they move in the same
direction) or negative (i.e. they move in different directions) on a
scale from -1 to +1
o Simple regression again shows the association between a pair of
variables in terms of whether it is positive (i.e. they move in the
same direction) or negative (i.e. they move in different
directions), but also the impact that one variable has upon the
other and elasticity of the relationship
o Multiple regression shows the association between the
dependent variable (i.e. the exchange rate) and a range of
independent variables (i.e. the factors hypothesised to influence
the exchange rate) in terms of whether it is positive (i.e. they
move in the same direction) or negative (i.e. they move in
different directions). Again it shows the impact and elasticity of
the relationship

 Suggested reading:
o Davis, G. and Pecar, B. (2010) Business statistics using Excel,
Oxford University Press. [Management Library 658.00182 DAV]
o Gujarati, D. (2009) Basic econometrics, McGraw-Hill. [330.2 GUJ]
o Gujarati, D. (2011) Econometrics by example, Palgrave. [330.2
GUJ]

Obviously  Conclusion chapter to summarise the key aspects of the dissertation

this is (e.g. the individual conclusions from each chapter), together with
written at findings in relation to your research question(s). Plus what future / policy
the end of implications that might be forthcoming from your research.
your [approximately 750 words]
dissertation
 Email me a draft of the whole Dissertation for feedback at least ONE
week before final submission.

4
SPSS Regression Output1
Model Summary

Model R R Square Adjusted R Square Std. Error of the

Estimate

1 .859a .738 .698 .4918

a. Predictors: (Constant), GDP growth, real INT, CPI （1980=100）, GDP deflator
1980=100
ANOVAa

Model Sum of Squares df Mean Square F Sig.

Regression 17.717 4 4.429 18.316 .000b

1 Residual 6.288 26 .242

Total 24.005 30

a. Dependent Variable: real ER

b. Predictors: (Constant), GDP growth, real INT, CPI （1980=100）, GDP deflator 1980=100
Coefficientsa

Model Unstandardized Coefficients Standardized t Sig.

Coefficients

B Std. Error Beta

(Constant) 2.425 .228 10.625 .000

CPI （1980=100） .019 .004 2.933 4.577 .000

1 real INT -7.967 2.628 -.321 -3.032 .005

GDP deflator 1980=100 -.014 .005 -2.045 -3.122 .004

GDP growth -3.525 1.604 -.304 -2.198 .037

a. Dependent Variable: real ER

1
Also see: http://www.onecaribbean.org/content/files/exploringRelationshipsSPSS.pdf for ideas similar to the above and how to write-up your findings.

5
Results table shown in your dissertation

Dependent variable: Real ER Estimated coefficient

CPI 0.019***
(4.577)
Real interest rates -7.967***
(3.032)
GDP deflator -0.014***
(3.122)
GDP growth -3.525**
(2.198)
Constant 2.425***
(10.625)

R2 0.74
Adj. R2 0.70
F (4,26) 18.316***
n 30

Where: *** = 1%, ** = 5%, * = 10%

Absolute t-ratios in brackets

Note: all of the above information comes from the SPSS output shown on the page 5, but
only these important elements need to be shown in your results tables.

6
Ideas about how to discuss about your regression results

 The sign of the estimated coefficient in relation to that expected/predicted by either

economic theory or that found in previous empirical studies. If your estimated
coefficient’s sign is different to that expected/predicted, you particularly need to discuss
why this might be so.
 The interpretation of the impact of each independent variable upon the dependent
variable, i.e. the estimated coefficient. However, note that how this is interpreted
depends on the functional form (i.e. linear, log-linear, linear-log, double log) of your
estimated regression equation (see table below).
 Also you should note the size of the estimated impact – is it large or small – just because
it is statistically significant doesn’t necessarily mean it is important.
 Similarly, the calculation of the elasticity of each independent variable also depends on
the functional form (i.e. linear, log-linear, linear-log, double log) of your estimated
regression equation (see table below).
 The goodness of fit (R2) shows how well your model performs, but if you change the
number of independent variables such that you want to compare models, then you need
to use the Adj. R2 which deflates the R2 according to the number of independent
variables.

Functional General form of Interpretation of estimated Elasticity (*=

form of equation coefficient elasticity depends on
model the value taken by X
and Y. If no X or Y
values are specified,
then these are
measured at the
mean values)
Linear Y = β0 + β1 X A one unit change in X leads to a β1 (X*/Y*)
β1 unit change in Y
Log-linear Ln(Y) = β0 + β1X β1 represents the relative change β1 (X*)
in Y for a given absolute change
in X, or if you multiply the
estimated β1 by 100 then it gives
the percentage change in Y for an
absolute change in X
Linear-log Y = β0 + β1Ln(X) β1 represents the absolute β1 (1/Y*)
change in the expected or mean
value of Y for a given relative
change in X
Double log Ln(Y) = β0 + β1Ln(X) A one per cent change in X leads β1
to a β1 per cent change in Y

7
8
Regression diagnostic discussion

To raise the level of your empirical analysis, then after completing the discussion of your
regression results, you should consider the potential breakdown of the assumptions of OLS
regression, i.e. whether the estimators are BLUE – best, linear unbiased.

 For all models you need to check for multicolinearity between the independent
variables, this is the undesirable situation when one independent variable is a linear
function of other independent variables. There are several ways you can test for this:
o By examining the level of correlation between the independent variables
o Test when running the regressions by selecting the Collinearity diagnostics test
through the sequence: Analyse → Regression → Linear → (choose your
dependent and independent variables) → Statistics → Collinearity diagnostics.
The collinearity diagnostics confirm that there are serious problems with
multicolinearity if several eigenvalues are close to 0, indicating that the
predictors are highly intercorrelated and that small changes in the data values
may lead to large changes in the estimates of the coefficients
o You can examine the variance inflation factor (VIF) where the established
threshold is, such that VIF should be lower than 10.

 For time series models the issue of serial correlation/autocorrelation, will be a potential
problem. You can test when running regressions by selecting the Durbin-Watson test
through the sequence:
o Analyse → Regression → Linear → (choose your dependent and independent
variables) → Statistics → Durbin-Watson. You will then need to manually check
to see if serial correlation/autocorrelation is a problem in your model (see one of
the recommended textbooks for how to do this test).
o To remove serial correlation/autocorrelation from your model you have to use
the SPSS syntax editor: File → New → Syntax → type AREG → type dependent
variable SPSS name followed by the word with and then type the SPSS name of
your independent variables → press red arrow on menu bar to run the
regression. This will run the model twice, once without correcting for serial
correlation and then correcting for it.

 For cross-section models the issue of heteroskedasticity will be a potential problem.

Unfortunately, SPSS doesn’t not have an option for one of the standard tests to detect
this, so you would need to plot the estimated disturbance terms see:
o http://www.youtube.com/watch?v=3QcX4jqPn14
o http://www.youtube.com/watch?
annotation_id=annotation_803936&feature=iv&src_vid=3QcX4jqPn14&v=bC2rh
8mLkbY