How to write a Master thesis
How to write a Master thesis
1. INTRODUCTION
The purpose of this paper is to give some advice to students writing their Master thesis.Typically students spend too much time in trying to .nd an interesting topic for a thesis. Thestudent counsellor has a set of topics that will attract the interest of potential supervisors. Manycompanies also have an inventory of problems that they are interested in. This can be a good .rstcontact with a potential employer. However, DO NOT CONTACT companies unless they havevoluntarily stated that it is OK after your .rst e-mail contact. Selecting a topic for a thesis istherefore normally not a problem, all topics are interesting if treated with imagination and based ondetailed knowledge. The probability of success is therefore more dependent on the use of conceptsthat the student has internalised rather than some new technique.The best way of getting an idea about the ideal "look and feel" for your thesis, is to read severalarticles in recognised journals. However, remember that the target audience of these articles is theprofessional research community in a speci.c .eld. Whereas, your audience is your fellow students.This di¤erence will therefore de.ne which concepts that you will have to clarify and typically addto the length of the paper. Methods or ideas covered in mandatory courses, eg. CAPM or standardeconometric techniques should not be explained.
2. PLAN
Familiarise yourself with the present research by reading a good survey article that is focusedon your topic. Prior to starting with the thesis, you should provide the student counsellor with astructured plan (max. 2 pages) for your thesis. This should contain: What is the problem thatyou want to address and why is this important? What kind of results do you expect to .nd? WhatSections do you want to include and what kind of questions do you want to answer in each Section?If the paper is empirical, how do you .nd data? Are you sure that this data is available to you,This guideline is to a large extent based on my seminar series to future PhD students titled - "How to write apaper". The usual disclaimer applies.
what kind of software packages are needed and do you have a working knowledge of them? Do youhave a suggestion for a supervisor?The plan should be .nalised and approved three months prior to when you expect to start withthe work. We strongly suggest that you co-write your thesis with a fellow student. When makingthe choice of co-author, try to .nd someone with similar rather than complementary competenciesto yourself. If for some reason, your colleague decides to abandon the project, you will be in a muchbetter position to .nalise the paper. Having similar interests will also make it easier to adhere tothe original plan. One month after you have started your work, you should present your mid-thesis report (max.10 pages), which will be commented upon by a fellow student. The .nal report should be presentedno later than 10 weeks after you started. After this, only minor adjustments should be made. Tobe able to .nish a project according to the original time plan is of outmost importance. #p#分页标题#e#
3. LAYOUT AND STYLE
Do not spend too much time on this issue until the .nal stage. Successive changes will makesome of your intermediate work obsolete. However, writing becomes much simpler if you useautomated references to Equations, Sections, Figures and Tables as well as to citations. Keep eachobject (Figure and / or Table) centred on a single page and after the Section - References. Makesure that objects are self-explanatory. Put markers in the text where you want the object to beinserted. Use a style that is generally accepted, cf. the style guide of Journal of Finance, which isavailable on the American Finance Association homepage. To facilitate proof-reading, leave amplespace, at least two centimetres of empty space surrounding the text on the page and the textdouble-spaced.
Decide which English language you should use and adhere to it. American-English is a separatelanguage, not a dialect, mixing languages is therefore rude to both nationalities. Many pitfalls aredescribed in the highly recommended (and very funny) The Economists Style Guide, which isavailable on the www.The Fundamental rule of good writing: Revise, revise and revise. You have to convince potentialreaders quickly, therefore convey your message e¢ ciently! Most readers will scan the paper .notread it, you must therefore help readers minimise time spent on your paper. This is done by,catching the reader.s eye with an interesting title and clear structure. 4. STRUCTURE The structure of the paper should at least contain: Title, Abstract, Introduction, Conclusions,Acknowledgements and References.
4.1. TitlePaper
titles should attract attention: Pollution Havens and Foreign Direct Investment - DirtySecret or Popular Myth?, What determines why and how much people tip in restaurants?, IsHungary ready for in.ation targeting? Do public investments crowd out private investments?Fresh evidence from Fiji. Nord Pool: A Power Market without Market Power.
4.2. Abstract
The Abstract should attract potential readers. It should not be longer than half a page and max.100 words and should contain: Background, Purpose, Model-approach, Results and Conclusions.About one sentence on each point approximately .ve sentences. An example:In recent years several models have been proposed to estimate time-varying tech-nical e¢ ciency. These models di¤er to a great extent in speci.cation and estimation.This paper undertakes a comparison between di¤erent speci.cations proposed in ear-lier research. The models are used to estimate the technical e¢ ciency of 15 Colombiancement plants observed during 1968-1988. The e¢ ciency scores and the time path ofe¢ ciency are found to vary substantially across models. www.ukthesis.org
4.3. Clear structure #p#分页标题#e#
For an applied paper: Introduction; Previous research (short if not in introduction); Method-ology; Theoretical model; Calibration (Econometric speci.cations, Stochastic assumptions, Esti-mation procedures); Data with Summary statistics; Empirical results (Statistical .t, Speci.cationtests, Robustness-sensitivity); Conclusions; References. 4.4. Body of the paperKeep it down to a few sections. Try to get it short and keep the structure clear and well-balanced. Typical problems: paper too long: max 30 pages double-spaced, excluding Tables andFigures; the string of paragraphs do not give a clear structure; too many Tables and / or Figures;policy conclusions (if any) are missing.
4.5. How to write Introduction and Conclusion
The Introduction is the most important part of the paper as well as the most demanding partof the paper. It should clearly state: what the problem is, why it is important, what has been done so far, what is your value added is (a better estimation, method, data set, model. . . ), whatyour .ndings are, what the structure of the paper is.Major points: place your work in the context of the existing literature, describe your main.ndings. Do not start with a three-page survey of the .eld .your reader will want to know yourcontribution sooner than that. Give priority to the development of ideas .not who did what,albeit, that must also be included. Your contribution in relation to this should be unambiguous.Be fair and refer to relevant background papers, especially surveys. Only relevant references .noname dropping. Use plain language and general concepts .no technical details.It must contain: the purpose (objective) of this paper is. . . . . . , the contribution of this paperis. . .My (our) major contribution is. . . ., the content of the paper (the paper is organised as follows.).Important to check how your paragraphs .t together .the sequence of paragraphs is very importantin the introduction .try to get a convergence to the main point: your own objective. An example:The speci.cation and estimation of productive e¢ ciency has evolved rapidly duringthe last decades, and there are a large number of di¤erent approaches, deterministic,stochastic, parametric and non-parametric, applied in empirical research. While moststudies are based on cross-section data, there is an emerging set of empirical applicationsbased on balanced or unbalanced panel data. Because such data will provide a muchmore detailed evaluation of the relative performance of micro units, linking e¢ ciencyto productivity and technical change, are of great value.
A crucial issue in panel data modelling is the speci.cation of e¢ ciency and tech-nical change. Such a speci.cation reveals the time path of e¢ ciency for the microunits answering questions about the degree of stability in e¢ ciency scores and whetherine¢ ciency is a transitory or permanent state. Especially in the stochastic frontierliterature, which is of concern here, several approaches are suggested.Panel-data models in the stochastic frontier literature can be divided into two maingroups. The .rst group assumes technical e¢ ciency to be time-invariant; see Pitt andLee (1981), Schmidt and Sickles (1984), and Battese and Coelli (1988), among others.The second group allows technical e¢ ciency to be time-varying; see Cornwell, Schmidtand Sickles (1990), Kumbhakar (1990), Battese and Coelli (1992), Lee and Schmidt(1993). Each of these two groups can be further subclassi.ed depending on whetheror not any distributional assumptions or functional forms are imposed on the errorcomponents #p#分页标题#e#
In the literature on stochastic frontier functions, several models are proposed toallow technical e¢ ciency to be time-varying. However, in the empirical literature onthe estimation of production frontiers using panel data, the issues of model speci.cationand selection of various speci.cation forms are rarely emphasized.
In this paper we consider the issues related to the speci.cation and estimationof various models incorporating time-varying technical e¢ ciency. First, we presentand estimate three di¤erent model speci.cations. Second, we analyse the bene.tsand limitations of the di¤erent speci.cations and quantify the impact of these on theresults obtained. Third, in the empirical part, we use unbalanced panel data fromthe Colombian cement industry, observed over the period 1968-1988, to compare theproductive performance of the plants in terms of input elasticities, returns to scale,technical progress and technical e¢ ciency measures.
A priori, the choice between time-invariant and time-varying technical e¢ ciency isdependent upon technological and the recent surveys of the frontier literature are foundin Lovell (1993), Greene (1993), and Heshmati (1994).
Because of the lack of a well-established theory that generates a certain structure of.rm ine¢ ciency, the speci.cation of ine¢ ciency in frontier modelling is usually ad hocand based on tractability rather than on a theoretically derived ine¢ ciency generatingmechanism. In the literature, however, there are a few old stylized models whichgenerate a certain time path for the development of technical e¢ ciency of a productiveunit; see Hjalmarsson (1973) and (1974), and Forsund and Hjalmarsson (1974) and(1987). First, in a Manne-type capacity-expansion model of the putty-clay type, witheconomies of scale ex ante and a growing demand for output, but no technical progress,a certain productive unit becomes gradually less e¢ cient as economies of scale aregradually more exploited in new capacity; see Hjalmarsson (1974). At the same timethe e¢ ciency structure may be fairly constant. Second, the same holds in a putty-claymodel with embodied .but no disembodied .technical progress as new technologyis picked from the ex ante production function. Third, a putty-clay model withouttechnical progress or scale economies in the ex ante production function generates agradual compression of the e¢ ciency structure, since when all units are renewed theywill pick technology from the same constant ex ante production function. In this case,the e¢ ciency of a certain unit will remain constant until there is a jump in the e¢ ciencyscore from worst practice to best-practice when an old unit is replaced by a new one.Fourth, a "su¢ cient" condition to keep e¢ ciency time-invariant in a putty-clay model isthe presence of disembodied technical progress at a rate that compensates for embodiedtechnical progress and scale economies in the ex ante function.Outside the stylized models, the presence of disembodied technical progress maygive fewer ine¢ cient units a chance to catch up with the frontier units as these aregradually improving.#p#分页标题#e#
Elimination of organisational slack may compress the e¢ ciencydistribution, while generation of slack works the opposite way. Thus a priori, only under strong assumptions could we expect e¢ ciency to be time-invariant. The main reasonfor our choice of the cement industry is that it approximates a putty-clay industryquite well. The basic technology is embodied in the cement kiln, the design of which isdecided at the investment date. Technical progress is characterised by capitalembodiedenergy-saving progress, while labour-saving progress seems to be of a more disembodiedor gradual nature; see Forsund and Hjalmarsson (1983).The paper is organised as follows: In the next section the stochastic frontier pro-duction function models with time-varying technical e¢ ciency are presented. Section3 contains the speci.cation tests and the empirical results, along with a comparison ofthe performance of the di¤erent model speci.cations, while Section 4 summarises anddraws conclusions. The Colombian cement data is described in Appendix. 4.6. Conclusion Repeat what the problem was and give compact summary of your results. Indicate clearly whatyour contribution is and a statement of the main lesson to be drawn from your analysis. Addressopen points and promising directions for future work.
An example:In this paper we have focused on labour productivity and the relative performanceof public and private utilities in Swedish electricity retail distribution during 1970-1990. Because in electricity distribution the choice of output measure is a controversialissue, we consider a hedonic measure of output which is constructed from outputs, theirqualities/attributes, and network variables. The estimation is based on a large sampleof panel data on privately owned .rms, municipal utilities, municipal companies andcompanies with mixed ownership during this time and we examine economies of scale,technical change, and relative e¢ ciency in labour use across ownership categories.Empirical results show that: Privately owned .rms are relatively more e¢ cient and the relative e¢ ciency ofmunicipal companies has deteriorated rather much over time according to the averageproduction function but improved slightly at the frontier. The results are much thesame for municipal companies at the average function, while the time development atthe frontier is rather uneven. Mixed ownership .rms improve relative e¢ ciency at theaverage function but show an uneven development at the frontier. In the www.ukthesis.orgstochasticfrontier models municipal utilities are more e¢ cient than mixed ownership .rms mostyears while the opposite holds in the deterministic DEA models.In general, for the mean of all the .rms as well as for each ownership category,increasing returns to scale are observed most years. An exception is the mixed own-ership category which shows decreasing returns to scale all years in Model 1 and this is also the case in Model 3 after 1979. #p#分页标题#e#
In all models, scale economies are found to behighest for the .rms under municipal utilities, except for some years with very highmixed ownership values.
In contrast to the substantial variation in returns to scale across models, there isvery little variation in the rate of technical change between di¤erent models. Thereis a slowly declining trend in technical progress from the range 2-3% per year downto the range of 1-2% per year. The di¤erences in technical progress between di¤erentownership categories are small.The persistent e¢ ciency di¤erences between private and publicly owned .rms, weinterpret as a strong indication of the impact of yardstick competition. However, thisyardstick type of regulation seems to o¤er weak incentives for cost minimisation in themunicipal .rms located in densely populated regions. 4.7. Acknowledgements
Be generous! An example:The paper is written as part of the Norwegian Research Council program E¢ ciencyin the Public Sector at the Frisch Centre, University of Oslo. It was .nished while the.rst author was a visiting fellow at ICER, Turin. Additional support from the followingsources is gratefully acknowledged: The Bank of Sweden Tercentenary Foundation,HSFR, Jan Wallander.s Research Foundation and Gothenburg School of EconomicsFoundation. We are indebted to Anders Hjalmarsson for carrying out all programmingand calculations. The paper has been greatly improved due to comments from refereesand special issue editor.
4.8. References Be fair and refer to relevant background papers, especially surveys. Only relevant references.no name dropping or .opportunistic. references. Check you statements carefully .who wrotewhat. Refer to good recent surveys, check year, pages, spelling, etc. carefully. Make sure thatthere is a one-to-one mapping between your citations and references.
5. ONCE THE DRAFT IS FINISHED
Do NOT submit it to your supervisor: Read and revise a few times then put it aside for a week.Read and revise again, then submit it. Your supervisor has probably read thousands of articles, ifhe / she does not immediately understands what you want to achieve, no one else will. Good luck.
