Beruflich Dokumente
Kultur Dokumente
new knowledge. Learn to write but also condition you introduced, or directed
write to learn. 2 you to a pertinent reference. Your ap-
portionment of credit among the vari-
2. General Principles ous people who helped you, however,
should be commensurate with the time
Convey your message efficiently. 3 By and effort they spent and the usefulness
leafing through your article, a reader of the suggestions they made. The
should be able to easily spot the main referee who sent you five pages of
results, figure out most of the notation, comments deserves recognition in a
and locate the crucial definitions separate sentence.
needed to understand the statement of In your introduction, briefly place
each theorem. your work in the context of the existing
Readers who have found your central literature and describe your main find-
points interesting and want to know ings. Do not start with a two- or three-
more, but have little time to invest in page survey of the field; your reader
your work, should then be able to get will want to know what your contribu-
an idea of your methods of proof by tion is sooner than that. Use plain lan-
visual inspection. It is often quite infor- guage, and skip the technical details.
mative just to glance at the way an Your literature review should not be a
argument is structured and to identify mere enumeration of previous articles.
the central assumptions and the known In describing the work on which you
theorems on which it is based. Think build, give priority to the development
about the way you read a paper. You of the ideas rather than to telling us
probably do not proceed in a linear way. who did what, although this information
Instead, you scan it for the formal re- should be included, and where you
sults and look around them for an expla- stepped in should be unambiguous. You
nation of the notation and terminology need not repeat in the body of the pa-
that you do not recognize or guess. You per all of the points that you made in
do not like having to spend too much the introduction, although some repeti-
time to find what you need. Your read- tion is unavoidable. On the other hand,
ers probably feel the same way about I do not generally favor relegating proofs
your work. to appendices (more on this later).
The Components of a Paper. Your ti- Your conclusion should not be a re-
tle should be as descriptive of your hashing of the introduction. However, a
topic as possible. Devote time to your compact summary of your results and a
abstract, as it is on that basis that many statement of the main lesson to be
potential readers will decide whether to drawn from your analysis is a good lead
continue. In your acknowledgment foot- to a list of specific open questions and a
note, be generous. Include the seminar general discussion of promising direc-
participant who suggested a name for a tions for future work, all of which do
2 I owe this formula to William Zinssers 1989 belong there. In your bibliography, give
pedagogical essay. Writing to Learn, a book that I the relevant background papers. If a
strongly recommend. good survey is available, mention it. You
3 This paper is longer than the average, but ex-
cept in Lake Wobegon, not all papers can be may have to include papers that you did
shorter than the average. Actually, I do not have a not use, and papers that you discovered
recommendation on how long a paper should be, only after you completed yours. Check
except for Make it as long as it needs to be, no
longer, and no shorter. If its structure is clear, references carefully, and update them
length by itself is not a problem. as papers get published.
Thomson: The Young Persons Guide to Writing Economic Theory 159
The structure of your paper should Also, motivate your work, but do not
be clear, as should the structure of each over-motivate it, or your readers will
section, each subsection, and each para- get suspicious.
graph. To better see how your para- Do not forget the process by which
graphs fit together, summarize each of you made your discovery. By the time
them in one sentence. Does the string your paper is finished, it will cover an
of these sentences make sense? It arbitrary number of goods and agents,
should. Perform this exercise also at the general production possibilities, uncer-
level of subsections, and then sections. tainty, and so forth, and nobody will un-
Show that what you did is interesting derstand it. If you read it several
and has not been done before. To show months later, you will not understand it
that your results are significant, the either. You got to your main theorem in
temptation is great to present them small steps, by first working it out for
with the utmost generality, with big two agents, two goods, linear technolo-
words, and in gory detail. Resist it! Try gies, and with no uncertainty, and by
instead to make your argument appear drawing lots of diagrams. It is also by
simple, and even trivial. This exercise in looking at simple versions of your
humility will be good for your soul. It model that your reader will understand
will also give referees a warm feeling the central ideas, and it is most likely
about you. Most importantly, it will these central ideas, not the details of
help you prove your results at the next proofs, that will help her in her own
level of generality. work.
Because the refereeing process and Reproducing the process of discovery
publication constraints often have the in a paper is not easy, but try. In a semi-
unfortunate effect of wiping out from a nar, quite a bit more can be done be-
paper most of what could make it easily cause of the informality of the occasion.
understandable, you may think that if Explaining how you came to the formu-
yours does not contain at least one re- lation you eventually chose and to your
sult that looks difficult, it is not ready results, however, is not a license to a
for submission. You are rightly proud of rambling discussion in which notation,
the sophisticated reasoning that led you definitions, assumptions, and motiva-
to your findings. Nevertheless, work tion are all mixed up, like the ingredi-
hard to make them look simple. 4 ents of a big salad. Even worse is
To show that what you do has not adding some semi-formal algebraic ma-
been done before, explain how your as- nipulations (tossing the salad?), and
sumptions differ from the assumptions suddenly confronting us with the sen-
used in related literature, and why tence: We have therefore proved the
these differences are significant, both following theorem: . . . As a reader, I
conceptually and technically. Demon- feel as if I have been mugged when I
strate your knowledge of this literature find myself in that situation.
by citing the relevant articles and tell- Another principle that has wide valid-
ing us how they pertain to your subject. ity is that good exposition means going
back and forth between the general and
4 As a young economist, it is natural that you the particular, and I will give several
should be proud of the complicated things you illustrations of it.
achieve; as you get older, you will become proud Learn from your errors. There is
of the simple things you do. (Of course, it is not
because you will not be able to handle the compli- nothing like having misunderstood
cated things anymore.) something to really understand it, and
160 Journal of Economic Literature, Vol. XXXVII (March 1999)
there is nothing like having seriously its members z and z, perhaps x, y, and
misunderstood it to really, really under- z, but certainly not b, or . Upon en-
stand it. Instead of being embarrassed countering z and z, your reader will im-
by your errors, you should cherish mediately know what space they belong
them. I will even say that you cannot to, how many components they have,
claim to have understood something un- and that these components are called zi
til you have also very completely under- and zj. If is a family of functions, re-
~
stood the various ways in which it can serve the notation and , (perhaps
be misunderstood. It has been said be- or even f ) for members of the family,
fore, and better: Erreur, tu nes pas un but certainly not or m.
mal. (Gaston Bachelard 1938) If R i is agent is preference relation,
Your readers are likely to be victims you may have to designate his most pre-
of the same misunderstandings that you ferred bundle in some choice set by
were. By remembering where you had bi(Ri), his demand correspondence by
trouble, you will anticipate where you di(Ri), and so on, but dropping this func-
may lose them, and you will give better tional dependence may not create ambi-
explanations. In a seminar, quickly guities. For instance, you may write bi
identifying the reason why someone in and di, provided that you designate
the audience is confused about some as- agent js most preferred element in the
pect of your paper may save you from a choice set and his demand correspon-
10minute exchange that otherwise dence by bj and dj, and the comparable
would force you to rush through the concepts when agent is preferences are
second half of your presentation. changed to R i by bi and di.
Designate time by t, land by , alter-
3. Notation natives by a, mnemonic notation by mn
and so on (and make sure that no two
Choose notation that is easily recog- concepts in your paper start with the
nizable. If you have no problem remem- same letter).
bering what all of your variables desig- Some letters of the alphabet are used
nate, congratulations! But you have in certain ways so generally in your field
been working on your paper for several that their common interpretation may
months now. Unfortunately, what you get in the way of other uses that you
call x is what your reader has been want to make of them. You will prob-
calling m since graduate school. ably be better off accepting tradition.
The best notation is notation that can Do not designate just any quantity by .
be guessed. When you see a man walk- Reserve this letter for small quantities
ing down the street with a baguette un- or quantities that you will make go to
der his arm and a beret on his head, you zero. 5 Call your generic individual i, his
do not have to be told he is a French- preference relation R i, his utility func-
man. You know he is. You can immedi- tion ui, and his endowment vector i.
ately and legitimately invest him with The production set is Y. Prices are p,
all the attributes of Frenchness, and
this greatly facilitates the way you think 5 I like the fragile look of my , especially when
and talk about him. You can guess his my printer is running out of toner. How could one
childrens namesRene or Edmond doubt that the quantity it designates is about to
and chuckle at his supposed admiration fade into nothingness? However, as a referee re-
minded me, in econometrics, the error term is
for Jerry Lewis. not necessarily a small quantity, but rather a quan-
Similarly, if Z designates a set, call tity that one would like to be small.
Thomson: The Young Persons Guide to Writing Economic Theory 161
quantities q. Calligraphic letters often out from the text and is perceived glob-
refer to families of sets; so, a is a mem- ally, as a unit: it is not read syllable by
ber of the set A, which is chosen from syllable. An alternative way to achieve
the family A. this important visual separation of the
Choose mnemonic abbreviations for axioms from the text is to capitalize them.
assumptions and properties. Do not re- Never use abbreviations in a section
fer to your assumptions and properties heading.
by numbers, letters, or letter-number Do not bother introducing a piece of
combinations. Since you state your first notation if you use it only once or twice.
theorem on page 10, it will be virtually There is no point in defining a new
impossible for us to remember then piece of notation if you hardly ever use
what Assumptions A1A3 and B1B4 it. How many times should a concept
are, but the fact that Assumptions Diff, be used to deserve its own symbol?
Mon, and Cont refer to differentiabil- Three times? Four times? I will let
ity, monotonicity, and continuity will be you decide. Certainly, do not bother
obvious to a reader starting there. introducing notation that you never use.
Choose these abbreviations carefully: If I feel the same way about utility nota-
you write Con, we may not know tion when only preferences are in-
whether you mean continuity or convex- volved. It is wonderful, of course, that
ity, so write Cont or Conv. The cost to preference relations satisfying certain
you is one extra strike on your key- properties can be represented by nu-
board, but your small effort will save us merical functions, and these repre-
from searching through the paper to sentations are sometimes useful or even
find which property you meant. Admit- necessary. But it has become a common
tedly, naming each assumption in a way excuse to use them even in situations
that suggests its content is not always where in fact they only clutter the text.
possible, especially in technical fields. Suppose, for instance, that you want to
It is common to introduce in paren- write that the allocation rule S is strat-
theses an abbreviation for a condition, egy-proof. This means that for every
next to the full name of the condition at agent i, announcing his true preference
the time it is formally stated. When the relation Ri is preferable to announcing
abbreviation is used later on, the paren- any false preference relation R i inde-
theses are no longer needed. 6 pendently of the announcements made
In axiomatic analyses, many authors by the other agents. Then (here I will
refer to axioms by numbers or abbrevia- skip the quantifications) you can write
tions, but I do not see any advantage to ui(S(u)) ui(S(ui,ui)), but is such an
that. The argument that numbers and expression preferable to S(R) R i S(R i,
abbreviations save space is not very con- R i)? If your paper involves long strings
vincing given that they will not shorten of terms of that form, as may well be
a 20page paper by more than five the case, utility notation will contribute
lines, and they certainly will not save to an unnecessarily messy look.
time for your reader. If you use differ- Matters are worse if you discuss cer-
ent typeface for your axioms, which I tain normative issues of welfare eco-
strongly recommend (for instance ital- nomics, social choice, or public finance,
ics, or slanted type), each axiom stands because in these fields utility functions
have cardinal significance. Even though
6 When you begin a proof, write Proof: and your theory may only involve the under-
not (Proof:). lying preference relations, some of your
162 Journal of Economic Literature, Vol. XXXVII (March 1999)
readers will come from a different tra- cate them. Do not write in= 1 xi, i N
x i,
dition and be tempted to compare utili- i xi, N xi, or i = 1, ,n xi when, in most
ties, or equate them, or maximize their cases, xi is perfectly clear. I assure
sum, and so on. On the other hand, if you, upon encountering xi, your read-
you address some problem of demand ers will be unanimous in assuming that
theory and you need to calculate matri- you are summing over i when i runs
ces of partial derivatives, then of course over its natural domain. Similarly, and
you cannot avoid utility notation. although the set consisting of agent i
Do not define in footnotes important alone should be denoted by {i}, if you need
notation that is unlikely to be familiar to refer to it on multiple occasions, you
to your reader, and that you will use in are better off dropping the curly brack-
the body of the paper. More generally, ets. Do apologize for the abuse of nota-
do not refer in the main text to terms, tion though. Similarly, if O designates a
ideas, or derivations introduced in a list of objects indexed by agents in the
footnote or in a remark, since the set N, you should refer to the shorter
reader may have skipped it. There is a list from which the i-th component has
hierarchy here that you have to respect. been deleted as O N\{i}, but it has become
Save on mathematical symbols. Do standard to write Oi. I welcome the
not use symbols that are not necessary. shortcut, and I used it earlier. Expres-
For instance, try to avoid multiple sub- sions can be considerably lightened by
scripts and superscripts. If you have using such tricks. Imagine that you are
only two agents, call their consumption on a diet and that each symbol is worth
bundles x and y, with generic coordi- one calorie. You will quickly discover
nates xk and yk (instead of x1 and x2, with that you can do with half as many. You
coordinates x1k and x2k). In a text, com- will improve the readibility of your text
binations of subscripts and superscripts and lose weight.
look a little better than only subscripts, Do not let the reader guess or infer
but in a blackboard presentation, watch from the context what your inequality
out for the sliding superscripts that end symbols mean. Define them the first
up as subscripts. If F is your generic no- time you use them. Doing that in a foot-
tation for a solution to the bargaining note is acceptable. 7 Alternatively, you
problem, you can certainly refer to the can give them in a preliminary section
Nash solution as FN , and when you ap- of notation.
ply it to the problem (S,d) with feasible
set S and disagreement point d, you will 4. Definitions
get FN (S,d). But why not simply desig-
nate the Nash solution by N? If you can Be unambiguous when you define a
choose the disagreement point to be the new term. Make it immediately clear
origin, as is almost always the case that indeed it is new. Do not let your
without loss of generality, ignore it in reader think that you may have already
the notation. Altogether, you will calcu- 7 x > y means x y for all i; x y means x > y
= i i =
late N(S), a much lighter expression and x y; x > y means xi > yi for all i. You could
than FN (S,d). If you systematically also use x => y, x > y, and x >> y. It is a very common
convention to define these symbols in a footnote,
search for such notational simplications, and this is where most of us will look for them
your text will be much cleaner. when we need them. It is therefore a good idea for
Bounds of summation or integration you also to define yours in a footnote. Some peo-
ple have an aversion to footnotes, but personally, I
are often (I agree, not always) unambi- love them. In academic writing, they are often the
guous. There is then no need to indi- only place where you will find evidence of life.
Thomson: The Young Persons Guide to Writing Economic Theory 163
Figure 1. Examples of increasing functions and of functions that are not increasing. (a), (b), and (c): These functions
are increasing. (d) and (e): These functions are not.
given the definition but she missed it, there is no other allocation z that all
or that you are assuming she knows the agents prefer to z.
definition. To emphasize certain aspects of your
Here are three possible ways of in- paper, such as important conclusions,
troducing a definition: 1. A function is exploit the typographical choices at
monotone if . . .; 2. A function is your disposal. Italics is a good one.
monotone if . . .; 3. A function is However, if everything is emphasized,
said to be monotone if . . . I prefer nothing is.
the first format and use it throughout When introducing a novel definition,
this essay, because its phrasing is direct give illustrative examples. If the defini-
and its different typeface will facilitate tion is a property that an object may or
its retrieval, if needed. Concerning the may not have, exhibit: 1. Objects that
typeface, I recommend boldface or satisfy the definition; 2. Objects that do
boldface italics over italics or plain text not satisfy the definition; 3. Objects
between quotation marks, neither of that satisfy the definition but almost do
which makes the new terms stand out not; 4. Objects that do not satisfy the
sufficiently. You should probably dis- definition but almost do. Examples in
play the crucial definitions separately, Categories 3 and 4 are particularly im-
and you may precede each of them by portant as they are responsible for most
the word Definition in boldface (see of the work in the proofs. Conversely,
the examples below). But do not intro- they may be the ones that allow the
duce all definitions in this way, espe- proofs to go through! In a paper, giving
cially if you have many of them, as it a range of examples that are repre-
will get tedious. Focus on the critical sentative of all four categories is, once
ones. again, not easily achieved because of
To avoid repeating quantifications space limitations, but in seminars this
that are common to several definitions, can sometimes be done. Here are two
you can group these definitions and illustrations:
state the quantification once: An allo-
Definition. A function f: [0, 1] R is
cation rule is efficient if for all prefer-
increasing if for all t, t [0, 1] with t >
ence profiles R, and all allocations z
t, we have f(t) > f(t).
that it selects for R, there is no other
allocation z that all agents find at least Figures 1a and 1b are dangerous, be-
as desirable as z and at least one agent cause they may plant in your readers
prefers; it is weakly efficient if instead mind the seed that you will work with
164 Journal of Economic Literature, Vol. XXXVII (March 1999)
R6
R4
R3 R5
R2
R1
a b a b a c d b
Figure 2. Examples of single-peaked and of non-single-peaked preference relations. (a) These relations are single-
peaked, with peaks at a for R1 and at b for R2. (b) These relations are single-peaked too, but they are not sufficiently
representative of the whole class due to their symmetry. Readers who have not worked with such preferences often
assume that symmetry is part of the definition, so you should emphasize that most single-peaked preferences do not have
that property. (c) These relations are not single-peaked, since R5 has two local maxima, at a and b, and R6 is maximized at
any point of the non-degenerate interval [c, d].
functions that are linear, or perhaps ing single-peaked because its repre-
concave. Figure 1c is what you need: it sentation achieves its maximum at a
represents an increasing function in its corner, or may think that R6 is admissi-
full generality, with a kink, a convex ble, although its representation has a
part, a concave part, and a discontinu- plateau and not a peak. You should
ity. Figure 1d is useful too, as it shows also make her aware of the fact that you
a typical violation of the property. include preferences that do not exhibit
Figure 1e is very important because the symmetry illustrated in Figure 2b.
it makes it clear that you want All of these examples will be very useful
more than that the function be to ensure that she fully perceives the
nondecreasing. 8 boundary of your domain.
Definition. The continuous preference Write definitions in logical sequences.
relation R defined on [0,1], with asym- Introduce terms in such a way that the
metric part P, is single-peaked if there definition of each new one only involves
exists x [0, 1] such that for all x, terms that are already defined, instead
x [0, 1] with either x < x x or of asking your readers to wait until the
x x < x, we have x P x. end of the sentence or paragraph for
everything to be clarified.
Figure 2 presents the graphs of the For instance, state the dimensionality
numerical representations of six prefer- of the commodity space before you in-
ence relations. Obviously, R 2 is single- troduce consumers or technologies. In
peaked and R5 is not. But your viewer the standard model, a consumer is no
may not immediately think of R 1 as be- more than a preference relation defined
8 Several readers of this essay objected to sen- over a subset of that space, together
tences such as this function is nondecreasing, with an endowment vector in the space;
which sounds too much like this function is not a technology is simply a subset of the
a decreasing function, but means something else.
Perhaps we should speak of a nowhere-decreas- space. In each case, it is therefore natu-
ing function. ral to specify the space, that is, the
Thomson: The Young Persons Guide to Writing Economic Theory 165
Make sure that the informal descrip- yourself) discover the relevance of your
tions of your definitions match their results to other situations that she (and
formal statements. If you write: A fea- you) had not thought about initially. To
sible allocation is Pareto efficient if pursue the example I just gave, the the-
there is no other feasible allocation that ory of coalitional games is also the the-
all agents find at least as desirable and ory of cost allocation. Some of your
at least one agent prefers, your formal readers are interested only in applica-
definition should not be (still using R to tions, and not in abstract games; others
denote a preference profile and intro- do not care for the applications. You
ducing p for the set of Pareto efficient can catch the attention of all by first
allocations): z p if (i) z Z and (ii) giving general definitions and then
for all z Z such that for some i N , pointing out the various possible inter-
zi Pi zi, there is j N such that zj Pj zj. pretations of your model.
Instead write: z p if (i) z Z and (ii) Present the basic concepts of your
there is no z Z such that for all i N , theory in their full generality. You will
zi R i zi, and for some j N , zj Pj zj. almost certainly use concepts that are
Separate formal definitions from their meaningful much beyond the frame-
interpretations. Formal models can work of your paper. It is preferable to
often be given several interpretations. introduce them without imposing the
It is, therefore, of great value to sepa- extra assumptions that you will need to
rate the formal description of your invoke for your analysis. When you ex-
model from the interpretation you in- plain what a Walrasian equilibrium is,
tend in your particular application. For do not assume convexity, monotonicity,
example, first write: or even continuity of preferences. 12 Of
course, these properties are relevant
Definition. Let V n be a domain of n- when you turn to the issue of existence
person coalitional games. A solution on of these equilibria, but they have noth-
V n is a function that associates with every ing to do with the concept of a Wal-
game v V n a point x R n such that rasian equilibrium itself.
xi v(N).11 When you introduce a piece of nota-
Then explain: If F is a solution on tion, tell your reader what kind of
V n, v is a game in V n, and i is a player in mathematical object it designates,
N, the number Fi(v) can be interpreted whether it is a point in a vector space, a
as the value to player i of being in- set, a function, and so on. Do not write
volved in the game v, that is, the A pair (p,x) is a Walrasian equilib-
amount that he would be willing to pay rium if . . . Instead, first define the
to have the opportunity to play it. Alter- price simplex 1 in the -dimensional
natively, it can be thought of as the Euclidean space and define the alloca-
amount that an impartial arbitrator tion space X. Then, write A pair
would recommend the player should (p, x) 1 X is a Walrasian equilib-
receive. rium if . . . Similarly, do not write
The advantage of this separation is The function is strategy-proof
that it will help your reader (and even if . . . k, but instead, after having
11 Here we have a bit of a notational problem as defined the set of possible preference
the n exponent to V n indicates the n-player case,
whereas the n exponent to R n indicates the n-fold 12 Discontinuous preferences are not easy to il-
cross-product of R n by itself. To avoid it, you could lustrate graphically, so if you give a graphical illus-
write V (n), but I do not think that the risk of confu- tration of your concept, you probably should pre-
sion is sufficiently high to justify the parentheses. sent it for continuous preferences.
Thomson: The Young Persons Guide to Writing Economic Theory 167
14 Besides, if you have to consider changes in the pleonastic initial initial endowments to the
the endowment of a player, to find out for in- oxymoronic final initial endowments, and what-
stance whether the owner of two left gloves may ever benefit he may derive from his clever move
gain by throwing away one of them prior to enter- will be more than cancelled by the embarrassment
ing the market, you will have to make him go from of using bad English.
170 Journal of Economic Literature, Vol. XXXVII (March 1999)
Use technical terms correctly. Do not cisms, nipponisms, sinicisms, and so on,
use the term vector unless you will get the help of a native gardener.
perform vector space operations. If you
have in mind a collection of objects 5. Writing Proofs
taken from some set, the appropriate
terms are lists, ordered lists, or Learn LATEX or Scientific Word. One
profiles. For instance, the notation of the first choices you have to make is
(R 1, , R n) refers to an ordered list of that of a typesetting software. For your
preference relations (or a preference dissertation, I strongly endorse LATEX,
profile), not a vector of preference rela- (or TEX, or Scientific Word, whichever
tions: you will probably not compute one you can handle). LATEX makes plain
(R 1 + R2) 2. On the other hand, it is text look beautiful, and because it un-
often appropriate to present a list derstands the structure of mathematical
(s1, , sn) of strategies as a strategy vec- expressions, its benefits for the writing
tor; for instance, in a game form de- of mathematics cannot be measured.
signed to implement a solution to a Moreover, it is widely used (in mathe-
public goods problem, a strategy for an matics, it has truly become the type-
agent may be a public good level, and setters LATIN , and you will find it very
the outcome function may select the av- convenient when collaborating with
erage of the announced levels. Con- coauthors dispersed throughout the
sumption bundles are usually vectors. world. A reader of a previous version of
You often compute averages of con- this essay suggested that I recommend
sumption bundles or multiply them by the LATEX Graphics Companion of
two. Goosens, Rahtz, and Mittelbach (Ad-
Do not confuse functions with the val- dison-Wesley) and PSTricks of Timo-
ues they take. If f:R R is a function, thy van Zandt, advice that was seconded
f(x) is the value the function takes when by another reader. If you do not know
its argument is x. So f(x) cannot be dif- how to use these softwares, ask one of
ferentiable, or concave, and so on. your younger classmates to teach you
These are properties of f and not of its (knowledge about computers goes from
values. Designate the function simply the young to the old). Also, use a spell-
by f (this is better than f()). By the check. When submitting a paper to a
same token, ui(xi) is not agent is utility journal, respect their style guidelines.
function; ui is. Conversely, if ui is agent The optimal ratio of mathematics to
is utility function, it is not also the par- English in a proof varies from reader to
ticular value that this function takes for reader, but there is a consensus on a
a certain choice of its argument. If F is middle range. A proof written entirely
a solution to a class of bargaining prob- in English is often not precise enough
lems, and S is a problem in its domain and is too long; a proof written entirely
of definition, F(S) is not a solution any- in mathematics is impossible to under-
more, but something like a solution stand, unless you are a digital computer
outcome, the solution outcome of of course. Modern estimation tech-
S. Alternatively, you can call F a so- niques have shown that the optimal ra-
lution concept and refer to F(S) as the tio of mathematics to English in a proof
solution of S. lies in the interval (52%, 63.5%). Pick
Get a good dictionary, and, if English the point in that interval that is right
is not your first language, ask for assis- for you and stick to it. However, the
tance. To weed out from your text galli- theorems themselves should be stated
Thomson: The Young Persons Guide to Writing Economic Theory 171
Figure 3. The ratio of mathematics to English in a proof should be in the interval [52%, 63.5%]. (a) This proof
has too much math. Due to the density of mathematical symbols, it is virtually impossible to understand. (I can
only make out that it states the existence of ducks having certain properties.) (b) This game-theoretic proof due
to William Goldman (1973) has too much English; it is not precise enough and is too long. Not surprisingly,
two paragraphs down, the character who produced it is dead. (c) This proof is just right, said Goldilocks, and
that is the one she read. It is indeed pleasantly short and clean. Wouldnt you like to know what theorem it
proves?
Nobel prize in literature will not im- trate a proof, thank you, but why didnt
prove much by this staccato style. Yet I you say so ahead of time, so that we
could name several grammatically im- could identify on it the variables as you
paired writers who hardly ever used first introduced them and use it to fol-
subordinate or relative clauses and yet low your argument? Warning us of the
got to make the trip to Stockholm! If existence of a figure is especially impor-
you really do not like such choppy tant because, if your typesetting experi-
writing, in your very last draft, recon- ence is as limited as mine, you will find
nect some of your shortest sentences. it hard to control where the figure ends
Similarly, break your text into para- up (my computer always seems to make
graphs of reasonable size, keeping in those kinds of decisions), and a figure
mind that too much of a good thing is illustrating a particular proof might very
a bad thing: a sequence of one-sentence well appear on the page that follows the
paragraphs is not pleasant to read. proof instead of next to the proof.
A certain amount of redundancy is It is often worth explaining very sim-
useful, but do not overdo it. Giving an ple things, especially in seminars where
informal description of the main steps you will not have the time to explain the
of a proof in addition to the formal complicated ones in any detail, and es-
proof is not strictly necessary, but it pecially at the beginning. Indeed, if you
might be quite helpful. Any such expla- lose your audience then, you may have a
nation, however, should not appear hard time retrieving it.
within the proof itself, but outside and After stating an if and only if theo-
preferably before, so as to prepare us rem, do not refer to the if part and
for it. The proof itself should be as con- the only if part, or the sufficiency
cise as you can make it without hamper- part and the necessity part. Most
ing readibility. Similarly, when you people will not know for sure which di-
state a difficult definition, assist us by rection you mean. I have even seen
giving an informal explanation in addi- some of the greatest economists being
tion to the formal statement. Here, too, confused about this, and in my personal
give it before the formal statement, as pantheon, they are people whose ap-
this placement will prepare your read- proach to economics cannot be de-
ers for it. It will also save them 15 frusta- scribed as literary. Restate the result
tion: it is indeed annoying to spend in each direction as you discuss it. Simi-
time trying to understand a complicated larly, would you guess that most of your
concept when it is first given, only to professors really do not know what a
discover two paragraphs down that the marginal rate of substitution is? But it
author was willing to help after all. is true! To most of us, a sentence such
The same comment applies to fig- as Agent 1s marginal rate of substitu-
ures. If you provided a figure to illus- tion at z0 is greater than agent 2s only
means that the two agents indifference
15 Did you notice that I sometimes refer to the curves through z0 have different slopes
reader, sometimes to your reader (in the singu- at z0. We just hope that which is steeper
lar), sometimes to your readers (in the plural), will be clear when we really need to
sometimes to us, your readers? This is an exam-
ple of an inconsistency of style that should be know. Of course, we would never admit
avoided. Just like this should be avoided since I it in public, and I most certainly would
have throughout addressed you, my reader; there- never put such a confession in writing,
fore, I should have written, that you should
avoid. I return to this issue at the end of this for fear of being forever shunned by
essay. my colleagues! Instead, compare the
Thomson: The Young Persons Guide to Writing Economic Theory 173
agents marginal rates of substitution of very well work with linear preference
good 2 for good 1 at the point z0; even relations or non-convex onesthis own-
better, simply talk about their indiffer- ership will not always be so clear. Avoid
ence curves being more or less steep unnecessary arrows. You can most often
at z0. position your labels close to the items
It is a great unsolved mystery of neu- they designate without creating ambi-
roscience that someone can prove the guities. Use arrows only if the figure
fanciest theorems in the most abstract would get too crowded, in particular if
spaces and yet have trouble with some the label is too long. 16
very elementary operations. Remember Have one enumeration for each cate-
that. After all, havent you called your gory of objects. Number definitions
relatives in England when it was 3 a.m. separately from propositions, theorems,
there, after having carefully calculated and so on. Some authors use a single list
that it would be 3 p.m.? You might have for all of their numbered items, so that
failed in such a trivial calculation, for example, Definition 15, which is the
and yet brilliantly passed exams where tenth definition, is followed by Theo-
much more of your intellect was being rem 16, which is the third theorem, this
tested. theorem being followed by Corollary
Use pictures. Even simple pictures 17, which is the only corollary . . . and
can be of tremendous help in making so on. Multiple lists are preferable, as
your seminar presentations more vivid. they help us understand the structure of
Figures are also very important to the paper. If you have two main sec-
lighten a paper, to provide relief from tions, with one theorem in each, label
long verbal or algebraic developments, the theorems Theorem 1 and Theorem
and to illustrate definitions and steps of 2. Having a single list certainly fa-
proofs. Of course, a figure is not a sub- cilitates retrieving a needed item, but
stitute for a proof, and the proof should this benefit is too small. Bringing out
be understandable without it, but it may the structure of your paper is more
give the main idea, and thereby cut by important. 17
half (probably much more than that, ac- State your assumptions in order of
tually) the time your reader will need to decreasing plausibility or generality.
understand it. Again, remember the When introducing your assumptions,
hundreds of little diagrams that you start with the least controversial ones,
drew on your way to your results. and write them in order of decreasing
Label your figures as completely as plausibility. For utility functions, do not
possible. Label the allocations, the sup- write A1: ui is strictly concave; A2: ui is
porting prices, and the endowments. To bounded; A3: ui is continuous. Instead,
indicate the efficiency of an allocation, and here I do not attempt to give names
it often helps to shade the upper con- to the conditions, write: A1: ui is
tour sets in the neigborhood of that al-
location. Label a few indifference 16 Look at the map of the city where you live
curves for each agent (some redundancy there are hundreds of themand you will note
is useful). If you assume convexity of that all the streets are labeled without arrows and
yet without ambiguities! You surely do not need
preference relations and if in fact you arrows in your figures.
draw the indifference curves strictly 17 For long documents such as books, adding to
convex, who owns which indifference the label of a theorem the page number on which
it is stated might be useful: Theorem 3.123 is the
curve will be unambiguous. But if you third theorem of the chapter and appears on page
do not make that assumptionyou may 123.
174 Journal of Economic Literature, Vol. XXXVII (March 1999)
English words: for all i N, xi > yi, left in the box. Either these were re-
does not read as well as for all i N , we placement parts, or I had done some-
have xi > yi, but the formulation xi > yi thing wrong (I will not tell you which,
for all i N , in which the quantification but as a clue, let me say that there
over agents occurs after the inequality, never are replacement parts in the box).
also achieves the desired separation, Similarly, after QED, look in the box for
and it is shorter. stranded hypotheses. You might have
Be specific about which assumptions, made a mistake, but you might also be
or which parts of assumptions, you need pleasantly surprised to find that you can
for each step. Do not write The above actually prove your theorem without
assumptions imply that f is increasing differentiability. Wouldnt you be
if you need only some of the above as- thrilled if your result applied to Banach
sumptions to prove that f is increasing. lattices (which you did not even know
Write Assumptions 3 and 4 imply that f existed two weeks ago), while you
is increasing. Even better, if you do thought you were working in boring n-
not need Part (i) of Assumption 4, write dimensional Euclidean space?
Assumption 3 and Part (ii) of Assump- Sometimes, you will be unable to
tion 4 together imply that f is increas- show that a certain hypothesis is neces-
ing. Similarly, if Theorem 3 follows sary for the proof and unable to con-
from Lemmas 1 and 2, show us exactly clude without it either. This is an un-
how it follows. Do not write A and B comfortable situation that should keep
imply C and D, if in fact A implies C you up late at night.
and B implies D. At a very small addi- A given hypothesis may be the con-
tional typing cost, you can be much junction of several more elementary
more precise. ones. Then, try to proceed without each
When you cite a theorem, be as exact of its components in turn. For instance,
as possible. Refer to a textbook that if you have shown that Under compact-
most of your readers are likely to own ness of the set X, conclusion C holds,
or be familiar with. This is especially do not only check that without compact-
important for theorems that exist in sev- ness, C might not hold anymore. In-
eral forms; we need to know which ver- stead, ask whether Under boundedness
sion you are using. Also, you should of X, C holds and whether Under
probably cite the English edition of a closedness of X, C holds.
classic text instead of the translated ver- Explore all possible variants of your
sion in your native language, even results. If you prove that A and B to-
though that is the one you know well. gether imply C, do not limit yourself to
So write: By the Brouwer fixed point that statement. Find out whether simi-
theorem (Debreu 1959, p.26) . . . Add- lar statements hold with A replaced by
ing the page number is a nice touch. the ~closely related conditions A, A 0,
Verify the independence of your hy- and A , or B replaced by B and B, or C
potheses. For each hypothesis in each replaced by C 0. Knowing statement P is
theorem, check whether you could pro- not enough. Discover as many state-
ceed without it. Do not write Under ments as possible that are close to P and
Assumptions A, B, and C, then D, if A are also true, and statements that are
and B together imply C, or if A and B close to P but are not true. It is as use-
together imply D. ful to understand the multiplicities of
Having put together a toy for one of statements around the one you are
my daughters, I discovered some parts proving that could be true but are not,
178 Journal of Economic Literature, Vol. XXXVII (March 1999)
as the statement that you are proving. It tions to complicated proofs. This is true
may even be more useful. Comment on only if the examples are well chosen. A
the main variants of your theorem but general algebraic expression has in fact
keep to yourself the least significant the advantage of reminding us of the
ones. logic of an argument. If, to fix ideas,
Do not leave (too many) steps to the you choose x1 = 1 and x2 = 8, the number
reader. Give complete arguments. Some 9 will refer to the sum x1 + x2 but
steps in a proof may involve standard it might be helpful to remember this
manipulations and detract from your origin: so write 1 + 8 instead, or
main point. Perhaps they should not be 9 (= 1 + 8). The expression x1 + x2 is
in the body of the paper, but in an ap- often preferable. In a three-player
pendix. Do not take them out though. game, write the number of coalitions
Your reader may not be familiar with a as 23 1; we do not care much if that
derivation that you have seen and per- number is equal to 7.
formed hundreds of times. Just having Also, by using numerical examples in-
the option of assessing the length of a stead of algebraic notation, you lose
step and recognizing the names of fa- track of units of measurement. It makes
miliar theorems on which it is based it harder to check the correctness of
will be helpful to her in checking her expressions.
understanding of the logic of your argu- When you vary a parameter, as a re-
ment, even if she does not actually read sult of which agent 1s income goes
all the details. In general, I do not like from 5 to 7 and agent 2s income from 8
too much of the work to be relegated to to 5, it will soon be difficult to remem-
appendices. When I first look at a pa- ber which ones are the initial incomes,
per, I skip most of it anyway, and if I which ones are the final incomes, and
decide to study it more seriously, I find whose income is 5 and when. If you use
it annoying to have to go back and forth well-chosen algebraic notation, for in-
between the body of the paper and the stance by calling the incomes I1 and I2
appendix. before the change and I1 and I2 after
If you think a step is obvious, look the change, your reader cannot be
again. Do not think that your errors confused.
necessarily occurred in the hard parts of If you insist on using numbers,
your proofs (I should say, what you choose them so that whatever opera-
think are the hard parts of your proofs). tions you perform on them do not turn
They may very well have hidden in them into monsters. If you will divide x1
(what you think are) the easy parts, tak- by 2, choose x1 even; if you will take its
ing advantage of your overconfidence. square root, do not choose x1 = 10. Actu-
After completing your paper, search for ally, I take this back. It depends: if the
the clearlys and obviouslys and incomes are 5 and 7 initially, and they
make sure that what you claimed was are cut in half, they will be 5/2 and 7/2
clear and obvious is, if not clear and after the change and the fractions will
obvious, at least true.21 make it easier to remember that they
Numerical examples are not always are the new ones. If they were even,
useful. It is commonly thought that nu- you would be tempted to perform the
merical examples provide easy introduc- division to get integers and again, the
21 Do not deduce from this, however, that sim-
new incomes would be hard to tell apart
ply deleting the clearlys and obviouslys will from the old ones.
necessarily eliminate all of your errors. In filling a payoff matrix, take all
Thomson: The Young Persons Guide to Writing Economic Theory 179
native is for you to give entire maps, Alice in a seminar in which I discussed matching
theory, and a member of the audience commented
which in most cases will be by means of that I was showing my age! I was unfortunately not
explicit numerical representations for quick enoughshowing my age once againto
180 Journal of Economic Literature, Vol. XXXVII (March 1999)
Do not collapse two or three simi_ strictly convex and have infinitely dif-
lar statements into one by indicating ferentiable numerical representations.
the variants in parentheses. Consider For the proof of the main theorem, one
the following definition: The func- appeals to the Brouwer fixed point
tion f:R R is decreasing (increasing; theorem. Section 5 concludes. your
non-decreasing) if for all x, y R with readers will think you need psychiatric
x > y, f(x) < f(y) (respectively f(x) > f(y); help. Are you I or we? Is it because
f(x) f(y)). these assumptions are embarrassing
The only way for us to be sure we that you suddenly hide behind the pas-
understand this triple definition is to sive form? Believe me, we have all
read it three times (once for decreasing, made embarrassing assumptions. And
once for increasing, and once for non- why do you let Section 5 conclude when
decreasing), and yet it is pretty simple. you did all the work? The passive form
More complicated statements in that is found awkward by me and our advice
format require a mental gymnastics that here is to have it replaced. I is per-
will unnecessarily exhaust us. Just re- haps too personal. Between I and
state the complete sentence in the we, I choose we, but if you choose
various forms you need. I also have a lot I, we will respect your choice. 23
of trouble with and/or (or is it Similarly, do not travel back and
or/and?). forth between present and future
Do not start a sentence with a piece tenses. Do not write: First, I prove ex-
of mathematical notation. Journal edi- istence. Then I will apply the theorem
tors will red-pencil you if you do, and I to exchange economies. I conclude with
agree with them that it does not look open questions. In most cases, using
good, especially if the notation is lower the present tense throughout, even in
case. x designates an allocation is not describing past literature, is just fine.
pretty. I is the set of individuals is not Choose the sex of your agents once
as bad because I is uppercase (but what and for all. Flip a coin. If it is a boy,
a grammatical provocation!). Let x des- rejoice! If it is a girl, rejoice! And dont
ignate an allocation is what editors will subject them to sex change operations
prefer. from paragraph to paragraph. 24 Two-
Be consistent in your writing style. person games are great for sexual equal-
Do not switch back and forth between ity. Make one player male and the other
first person singular, first person plural, female. This will actually facilitate talk-
and passive forms. If you write: In sec- ing about the game and help your
tion 3, I show that an equilibrium ex- reader keep things straight. It will also
ists. In Section 4, we establish unique- save you from the awkward he or she,
ness. To prove these results, it is him or her,, his or her! Alterna-
assumed that preference relations are tively, you may be able to refer to your
23 As a reader, I rather like the I form, which
is more engaging, but I am not comfortable using
reply that by understanding that I was showing my it in formal papers. I use I here only because of
age, and remarking on it, he was showing his! He the informal style that I chose for this paper. Para-
was right though. I recently asked the students in doxically, the we form is less obstrusive than the
my graduate class whether they understood the I form. We can also be interpreted as you and
allusion. Not one of them did. And yet, Bob & the reader, whom you are taking along, but then
Carol & Ted & Alice (its a movie) came out only be careful if you refer to our previous work.
yesterday (30 years ago, to be precise)! From now 24 For a book, alternating between male and
on, I will use this example only when I give female between chapters might be acceptable
lectures in retirement homes. though.
Thomson: The Young Persons Guide to Writing Economic Theory 181
agents in the plural, or choose one of are < , , and , (and the other symbols
=
them to be a firm, and refer to it as it. of the same kind such as the strict in-
Be consistent in your choice of run- equalities, the strict inclusions, the
ning indices. If N = 1, , n, do not write preference statements, . . .), read as
interchangeably for all i N , or for prepositions or verbs.
all i 1, , n, or for all i = 1, , n.
Blah, blah, blah, since x y, and
Pick one formula and stick to it. In most
x A , and therefore, blah, blah, blah, f
situations, the quantification on the set
is continuous, is fine.
of agents (to take an example) is clear.
situations where it is convenient to
Skip it and write for all i. This
quantify once and 25. For instance,
helps keep down the density of symbols.
you can open your proof by stating: In
In general though, it is good to indi-
what follows, S denotes an arbitrary ele-
cate membership explicitly. For in-
ment of . Then the requirement that
stance, instead of There exists z for
the function F: R 2 satisfies for all
which . . ., write There exists zZ for
S , F(S) > 0 can simply be written as:
which . . . . Therefore, for consistency
of style and esthetic reasons, when Positivity: F(S) > 0. 26
everything else is explicity quantified, it
bothers me a little not to see member- Indicate the end of proofs clearly.
ship indicated for the set of agents, Use QED (for quod erat demonstran-
even if it is pretty obvious that they dum), or Halmos u (I suppose, for
come from N and not from Mars. So quod erat quadrandum. 27). Delete
instead of For all i such that . . . , 25 See the problem with starting a sentence with
I would write For all iN such a piece of mathematical notation (Section 6.10)!
that . . . When I wrote earlier that you should not put
quantifiers in the middle of a sentence in English,
Do not put quantifiers in the middle I should have said: do not put them anywhere in
of a sentence in English. A sentence such a sentence.
26 Or F > 0 . By the way, do not place your foot-
such as
note markers at the end of mathematical expres-
Blah, blah, blah, x such that P(x), sions, as they will look like exponents. Placing
blah, blah, blah y such that Q(x, y) and them beyond the punctuation mark, as the typo-
blah, blah, blah. graphical convention requires, and as I have done
here, helps, although logic would sometimes dic-
does not look good. Write for all and tate that the marker be attached to a word inside
there exists. If the mathematical the clause (or the sentence) that ends with the
statements introduced by the quantifi- punctuation mark. Compare the marker for this
footnote with the marker for the previous one: the
ers are complex enough, pull them out position of that earlier marker did not create any
from the text in English and display ambiguity, as I am sure that you did not think that
them on separate lines, as follows: my intention was to raise the universal quantifier
to any power; still it did not look pretty. The same
Blah, blah, . . . , blah, blah, problem arises with quotation marks. I just wrote
F > 0. The rule is to write F > 0. This is in
x such that P(x), y such that Q(x, y), agreement with logic if you think of the whole sen-
and blah, blah, blah. tence, including the period that ends it, as being
the unit that is being discussed. In other contexts,
The quantifications should always be it may be the requirement F > 0 that is under
unambiguous. Remember also that tak- discussion but here, and given that quotation
marks look a little like double prime, I admit that
ing the negation of a properly written placing them after a needed punctuation mark is
mathematical statement, with no hidden better, so refer to the requirement F > 0, which
quantifications, is a trivial operation. is proved in Section 2.
27 Circulus? What about a little circle to indi-
The only mathematical symbols that cate the beginning of a proof, matching the little
do not bother me in a text in English square that closes it?
182 Journal of Economic Literature, Vol. XXXVII (March 1999)
the redundant This completes the When you pick it up again, it will have a
proof, which precedes u in your freshness that will allow you to see
current draft. immediately where it can be improved.
Good writing requires rewriting, and
6. Conclusion rewriting again. When after many
drafts, your paper has become like a
If you follow all of the above recom- smooth and shiny pebble that fits snugly
mendations, not only will you be in the palm of your hand, treat yourself
pleased with yourself, your seminar to a box of Belgian chocolates. And if
audiences enlightened, your classmates you have found these recommendations
impressed, your parents proud of you, useful, please save me one!
and you will land a job in a top-five de-
partment, but most importantly, your 7. Related Literature
adviser will be happy. I readily admit
that each of them does not amount to As I started circulating this paper,
much. However small imperfections, several readers gave me references to
when added together, will take your pa- similar pedagogical essays written by
per over the line that separates those that mathematicians. I am happy to report
can be understood from those that can- that their recommendations are not al-
not. An Archimedian principle is at work ways in contradiction with mine. I
here. You will lose your readers or your found Nicholas Higham (1993) particu-
seminar audiences much earlier than nec- larly helpful. Paul Halmos essay in
essary. In fact, you too will be confused. the Norman Steenrod et al. (1983) vol-
Do not fool yourself: very few of your ume is often cited and deservedly so.
readers will take the time to understand Leslie Lamports (1986) manual is beau-
your whole paper, and a large fraction tiful. (I will even consider forgiving the
of your seminar audience will not have author for his maxim All axioms are
the faintest idea of what you are talking dull.) William Strunk and Edmund
about when you are half-way through. White (1979) is a well-known general
So, every bit will help in keeping the manual of style. The Merriam-Webster
attention of a few a little longer. Dictionary of English Usage is an
If you are used to certain notational invaluable source, and I am quite
conventions, or terminology, or ways of fond of the American Heritage Diction-
structuring a proof, they almost neces- ary of the English Language. An exam-
sarily seem the best to you, and perhaps ple of a beautifully written text is
the only ones worth considering. You the monograph by Grard Debreu
have to be open-minded and genuinely (1959).
experiment with other formulations. REFERENCES
Only then can you decide what is truly American Heritage Dictionary. 1992. Houghton
best. The first few times you use a new Mifflin Company, Third Edition.
Bachelard, Gaston. 1938. La Formation de lEsprit
piece of notation or a new term or a Scientifique.
new format, it will appear strange to Chicago Manual of Style. 1993. Fourteenth edi-
you. Give it a chance. tion. Chicago and London: U. Chicago Press.
Debreu, Grard. 1959, Theory of Value. New
Let time elapse beween revisions. If York: Wiley.
your paper is so familiar to you that you Goldman, William. 1973. The Princess Bride.
essentially know it by heart, you will Balentine Books.
Higham, Nicholas J. 1993. Handbook for the
never discover your mistakes. You need Mathematical Sciences. Philadelphia: Society
to let it sit in a drawer for a while. for Industrial and Applied Mathematics.
Thomson: The Young Persons Guide to Writing Economic Theory 183
Lamport, Leslie. 1986. LaTeX. Addison-Wesley to Write Mathematics. Rhode Island: American
Publishing Company. Mathematical Society.
Merriam-Websters Dictionary of English Usage. Strunk, William and Edmund B. White. 1979. The
1994. Springfield, MA: Merriam-Webster Inc. Elements of Style. New York: Macmillan Co.
Steenrod, Norman E.; Paul R. Halmos, Menachem Zinsser, William. 1988. Writing to Learn. Harper
M. Schiffer, and Jean A. Dieudonn. 1983. How and Row.