Sample Paper in Econometrics

This is a sample research paper for an introductory course in econometrics. It shows how to communicate
econometric work in written form. The paper integrates many writing instructions and rules into a single example
and shows how they all fit together. You should pay attention to the structure of the paper: how it is divided into
sections and how each section serves a distinct purpose. You should also note how the descriptive statistics and
empirical results are presented.

The paper includes numerous notes in the margins. These notes explain the purpose of each paragraph, and provide
comments on tables and other aspects of the paper. The margin notes are there to make you aware of the writing
process. They are designed to help you bridge the gap between reading and understanding on one hand, and writing
and creating knowledge on the other. The readings which have been assigned in your economics courses are finished
products which you are able to read and understand. However, in order for you to be able to create a finished product
yourself, you need to become aware of how such a product is created. The notes in the margins reveal the thinking
and consideration that go into each section, paragraph and table, and should therefore help you in writing your own
paper.

It is worth emphasizing that you should use this paper only as a guide. You should not copy the paper and simply fill
in your own names, words and numbers. You can deviate from the order and purpose of each paragraph in order to
meet the needs of your own work. You can add separate sections on prior literature, methodology or theory. Such
sections would normally come after the introduction. The sample paper includes the discussion of prior literature in
the introduction. The theory and methodology are folded into the Introduction, Data and Empirical Results sections.
The absence of a separate theory or methodology sections is not uncommon in applied empirical papers. However,
theory or methodology sections are a must when the empirical question is derived from an explicit theoretical model
or when the methodology requires a longer explanation. You are also welcome to include additional tables or graphs.
What should remain the same, though, is that each section, paragraph, table and graph has a purpose, and that they
are organized in a logical manner.

For printer friendly version click here.

Concepts illustrated in the paper

(The list is expandable. Click on the concept to see its description.)
Structure: Writing Style:

Introduction (see example) Citation Style (see example)

The Introduction should convey four things. The citation and bibliography styles most
First, what is the question that the paper asks. commonly used in economics are detailed in
Second, why is the question important. Third, the Chicago Manual of Style.
how is the paper going to answer the Use of acronyms (see example)
question. Finally, how is the paper related to The first time an acronym is used it should be
existing work. The introduction is the most written out, followed by the acronym in
important part of any paper. No one will parenthesis.
continue to read any further if the Use of first person (see example)
introduction is confusing or poorly written. It is acceptable to use first person (I) in an
Data (see example) economics paper.
The Data section should accomplish three Coherence (see example)
things: First, state the sources of data. Make each sentence linked to the previous
Second, discuss the variables used and how one.
they relate to the concepts that they are Tense (see example)
supposed to measure. Finally, present the It is appropriate to use past tense when
:
 data’s descriptive statistics. describing the construction of your variables.
Empirical Results (see example) However, use present tense when referring to
The Empirical Results section should present tables or your results.
and discuss the empirical results. The
presentation of results is usually done with a
table. The discussion of results typically
includes a statement of whether the results
support or refute the hypothesis, a statement
of whether the results are statistically
significant, interpretation of the magnitude of
the coefficients and a comment on functional
form.
Conclusion (see example)
The conclusion should accomplish three
things: summarize the results, explore the
implications of the results, and point to future
research.

Conventions in an Empirical Paper: Other:

Descriptive Statistics Table (see example) Title (see example)

A descriptive statistics table should include The title should concisely express what the
the list of variables and the mean, median, paper is about. It can also be used to capture
standard deviation, minimum and maximum. the reader's attention.
In cases where the number of observations Searching for existing literature (see example)
varies from variable to variable, a column EconLit is the most commonly used database
specifying the number of observations is for searching published papers in Economics.
necessary. The orientation of the table should Working papers can be found via IDEAS,
be such that the variables are in rows and the SSRN, NBER or even google.
statistics in columns. This way, even if a large Effect vs. affect (see example)
number of variables are used, the table will fit "Effect" is usually a noun (that is, it could be
on one page. preceeded by "the"). "Affect" is usually a
Discussing Descriptive Statistics (see example) verb.
Discussing the minimum and maximum and Appeal to authority (see example)
the corresponding data points makes the data It is appropriate to cite other studies when
“come alive.” It also reassures the reader justifying the use of a variable or technique.
that the data was put together correctly. This also makes the comparison to other work
Rounding numbers in the text (see example) easier.
When discussing quantities in the text, use Acknowledge shortcomings of data (see example)
round numbers. It is appropriate to acknowledge the
Presentation of regression results (see example) shortcomings of your data. The shortcomings
Regression results are typically presented in could come from unreliability of the source,
this compact form. The columns show results lack of observations or, as in this case, lack of
from 6 different regressions. The rows show time to properly adjust the data for inflation.
the intercept, independent variables and the
R-squared. The estimated coefficients and
their associated standard errors in
parentheses appear inside the table. Some
authors prefer to show each coefficient’s t-
statistics in parentheses; therefore it is always
necessary to specify this in the table’s
footnote. If the independent variable is not
included in a specification, the cell
:
 corresponding to that independent variable
and specification is left blank. If the number
of observations varies across specifications, it
can be included as the last row. The asterisks
are for easy identification of the significance
level - the more asterisks, the higher the
significance.
Converting variables to convenient units (see
example)
In order to be able to present regression
results in a compact and readable form, it is
necessary to convert the variables to
appropriate units. For example, the
appropriate units for payroll are millions of
dollars. This is because if payroll were in
dollars, the coefficient in specification (3)
would appear as 0.0000001 which is more
difficult to fit in a table and more difficult to
read.
Interpreting estimated coefficients (see example)
It is very important to include the units of
both the independent and the dependent
variables.
Assessing economic significance (see example)
Assessing economic significance requires
judgment. Unlike statistical significance,
there is no "official" benchmark for assessing
economic significance.

Does pay inequality within a team affect performance?

Tomas Dvorak*

1. Introduction

The business of sports draws considerable attention from the

media and the general public. Fans and sports writers frequently
speculate about the effects of money on athletic performance.
There is general agreement that more financial resources usually
lead to better athletic performance. In team sports, higher pay
can be used to lure better players from other teams and therefore
improve performance. However, performance can also be
affected by pay inequality among players within a team. On the
one hand, pay inequality could have a negative effect because it
may hinder cooperation among team members. In many sports,
team cooperation is critical for good performance. If pay
inequality creates tensions or animosity among team members,
:
 performance is likely to suffer. On the other hand, inequality
could have a positive effect on performance by providing
incentives. The prospect of a very large salary could be a
powerful drive behind an athlete’s performance. Pay inequality
might also enhance performance if low paid players learn from
high paid players. This would happen when pay inequality is
associated with skill inequality. For example, if a highly paid
superstar can teach other players, the overall performance of a
team may improve. Given that arguments can be made both
ways, it is not surprising that there is little agreement on the
effects of pay inequality on team performance. The purpose of
this paper is to determine whether, on balance, the effect of pay
inequality on performance is positive or negative.

Understanding the effect of pay inequality on a team’s

performance is important for at least two reasons. First, team
managers can use this information to make decisions about
which players to hire. For example, should they hire one
expensive superstar and two inexpensive players, or three
medium-priced players? If we find that pay inequality leads to
poor team performance, then the team may perform better with
three medium-priced players than one superstar and two low-
priced players. Second, because salaries are a large part of
contract negotiations between player associations and team
owners, understanding the effects of pay inequality on
performance can help determine optimal policies. For example,
if pay inequality has a negative effect on performance, an
argument for a higher minimum salary could be made.

There are a number of studies that look at the effects of pay

inequality on performance. DeBrock, Hendricks and Koenker
(2004) study the effects of pay inequality on performance in
Major League Baseball (MLB) . They find that pay inequality is
associated with poor performance. Frick, Prinze and
Winklemann (2003) look at the effects of pay inequality in all
four major leagues in North America. They find that inequality
improves team performance in basketball and worsens team
performance in baseball. They find no statistically significant
effect of inequality on performance in football and hockey.

This paper looks at the effects of inequality on performance in

MLB. It differs from that of DeBrock, Hendricks and Koenker
(2004) in that it uses the most recent data. While the previous
authors use data from 1985 through 1998, I use data from the
latest two seasons: 2003 and 2004. Another difference is that I
use a different measure of pay inequality. Rather than the
Herfindahl index, I use the percentage of payroll earned by the
best paid 20% of players. I chose the share earned by the top
20% players for two reasons: it is somewhat easier to calculate,
:
 and its magnitude is easier to interpret.

2. Data

The data on pay inequality was constructed in the following

way. From the USA Today salary database, I collected annual
salaries for each player in all MLB teams during the 2003 and
2004 seasons. I summed the salaries of all players for each team
and each season to obtain the total payroll. The active roster in
baseball is 25, but the database includes salaries of disabled
players as well. Therefore, the number of players for each team
ranges from 25 to 31. As the measure of pay inequality, I
calculated the percentage of payroll earned by the highest paid
20% of players. For example, for a 30 player team I summed the
salaries of the highest paid 6 players and divide that amount by
total payroll. If every player earned the same amount, the best
paid 20% would earn exactly 20% of the payroll. When pay is
unequal, this measure is higher than 20%. The higher the share
of payroll earned by the top 20% of players, the higher the pay
inequality.

To measure performance I use the percentage of games won in

the regular season. This data comes from
BaseballReference.com. It does not include performance during
league championships or the World Series. However, with 162
games per regular season, the winning percentage can be
regarded as a reasonable measure of performance. This is also
the measure used by DeBrock, Hendricks and Koenker (2004).

In addition to pay inequality and performance, I use data on the

total payroll of each team. This is a measure of financial
resources which could be an important determinant of
performance. I measure payroll in current dollars and do not
adjust for inflation. While 2003 dollars are not exactly
comparable to 2004 dollars, 2003 inflation was low enough not
to influence the results significantly.

Table 1 shows the descriptive statistics of each variable. In the

first row we see that on average the highest paid 20% of players
earn about 61% of the total payroll. This implies that on a 30
player team, the six best paid players earn more than the
remaining 24 combined. According to this measure, the team
with the most equitable pay is the New York Yankees during the
2003 season when the top 20% of players earned only 42% of
total payroll. The team with the highest inequality was the
Colorado Rockies during the 2004 season. On that team, five
players earned more than 78% of the team’s total payroll.
:
 The second row in Table 1 shows that the average winning
percentage is 50% which has to be the case since for every game
won there is a game lost. The Detroit Tigers have the lowest
winning percentage in the data with only 26% of games won
during the 2003 season. The maximum winning percentage in
the data is for the St. Louis Cardinals, who won nearly 65% of
their games during the 2004 season. Finally, the last row in
Table 1 shows that the average payroll is about 70 million
dollars. The range of payroll is quite striking. It goes from less
than 20 million dollars for the Tampa Bay Rays to over 184
million for the New York Yankees.

Table 1: Descriptive Statistics

mean median st.dev. min max

Top20share (in %) 61.0 61.4 8.0 42.2 78.3
Games Won (in %) 50.0 51.6 8.2 26.5 64.8
Payroll (in mil. USD) 70.0 65.3 30.3 19.6 184.2

3. Empirical Results

I estimate three different specifications. The dependent variable

in each specification is performance, as measured by the
percentage of games won. Pay inequality and total payroll are
the independent variables. Table 2 shows the results. In the first
specification, I regress performance on the share earned by the
top 20% of players. The coefficient on the share of top 20% is
negative and statistically significant. This indicates that teams
with higher pay inequality tend to win fewer games. A one
percentage point increase in the share of payroll earned by the
top 20% of players is associated with about half of a percentage
point decline in the percentage of games won.

Table 2: Regression Results

Dependent variable: winning percentage (in %)

(1) (2) (3)
Intercept 77.3 59.9 37.35
(7.43)** (9.51)** (15.37)*

Top20share (in %) -0.45 -0.27 -0.28

:
 (0.12)** (0.13)* (0.13)*

Payroll (in mil. USD) 0.10

(0.04)**

Log of Payroll 7.09

(2.43)**
R-squared 0.19 0.29 0.29
Adjusted R-squared 0.18 0.26 0.27
Number of observations is 60.
Standard errors are in parentheses.
** significant at 1%, * significant at 5%

In the second specification I include total payroll as an

independent variable. Payroll is a measure of the financial
resources which can affect performance - the higher the payroll,
the higher the quality of players and, generally, the better the
performance. Therefore, including payroll may increase the
precision of the estimated coefficient on pay inequality. More
importantly, it is possible that pay inequality is correlated with
total payroll. If low payroll teams tend to have more pay
inequality, then the coefficient on pay inequality in specification
(1) is biased. Indeed, the correlation coefficient between the
share earned by the top 20% of players and total payroll is -0.5.
Teams with high pay inequality may perform worse not because
of pay inequality, but because they are also the teams with a
lower payroll. Therefore, in order to measure the effect of pay
inequality on performance, I need to control for total payroll.

Once I control for total payroll, the coefficient on the share of

the top 20% remains statistically significant but the magnitude
drops substantially. Holding payroll constant, a one percentage
point increase in the share earned by the highest paid 20% is
associated with a 0.27 percentage point decline in the percentage
of games won. The impact of inequality on performance does
not seem enormous. For example, a five percentage point
increase in inequality for the team with median inequality would
shift the team up 13 spots in the inequality ranking, but its
performance ranking would drop by only 2 spots. The
coefficient on total payroll is positive and statistically
significant. A one million dollar increase in total payroll is
associated with about 0.1 percentage point increase in the
percentage of games won. This indicates that greater financial
resources tend to improve performance. Adding payroll as an
independent variable led to an increase in R-squared from about
0.19 to 0.29.

Finally, in specification (3) I include the logarithm of payroll

instead of payroll. I want to verify that the result in specification
(2) is robust to different functional forms. In addition, the effect
of an additional one million dollars may be smaller for a team
with a 100 million payroll than for one with a 20 million payroll.
Thus, including payroll in logarithm seems appropriate. The
:
 coefficient on the share of the top 20% remains statistically
significant with roughly the same magnitude. The log of payroll
is statistically significant. A one percent increase in payroll is
associated with about 0.07 percentage points increase in the
percentage of games won.

4. Conclusion

The analysis in this paper shows that pay inequality within MLB
teams has a negative effect on performance. The effect remains
statistically significant even after controlling for total payroll.
The result is the same as that of DeBrock et al. (2004) who use
data from 1985 through 1998. My paper confirms their finding
using the most recent data and using a different measure of pay
inequality.

The fact that pay inequality leads to worse performance implies

that managers should strive for pay equality in their teams. For
example, instead of hiring two low-priced players and one
superstar, performance may be better if three medium-priced
players are hired. Given these results, it is surprising that there is
not a more equal distribution of pay in baseball. One possible
explanation is that managers may care about attendance as well
as winning. They may be willing to sign up an expensive
superstar who will attract fans even though it will increase pay
inequality and may hinder performance.

The conclusions above are subject to a number of limitations.

First, it is unclear to what extent the results can be generalized to
other sports. Each sport requires a different degree of
cooperation among team members. Therefore, the relationship
between pay inequality and performance is likely to differ across
sports. Second, the error terms for each team could be correlated
over time. For example, if a team wins a lot of games one year
given its payroll and pay inequality, that team is likely to win a
lot of games the next year as well. Therefore, the estimation
procedure may need to correct for this autocorrelation. Finally,
there may be other variables that affect performance, e.g. coach
salary or quality of training facilities. Including these in the
:
 regression would increase the precision of my estimates as well
as eliminate potential omitted variable bias.

The channels through which pay inequality affects performance

are not clear. I can think of two possibilities. One is that pay
inequality leads to tensions within the team and impairs
performance. The other possibility is that baseball requires
players of similar quality. Pay inequality is probably associated
with skill inequality, and it may be the skill inequality that drives
down performance. An excellent pitcher cannot win the game
when the outfielders cannot catch or throw. It may be possible to
distinguish these two channels empirically. Using statistics on
individual player skill level, one could construct a measure of
skill inequality for a team and include it as an additional control.
The coefficient on pay inequality in that case would capture the
effect of pay inequality on performance while holding skill
inequality constant. A negative impact of pay inequality would
then support the idea that pay inequality leads to tensions which
affect performance. This investigation, however, is left for future
research.

References:

DeBrock, Lawrence, Wallace Hendricks, and Roger Koenker.

2004. Pay and performance: The impact of salary distribution on
firm-level outcomes in baseball. Journal of Sports Economics 5
(August): 243–261.

Frick, Bernd, Joachim Prinz, and Karina Winkelmann. 2003.

Pay inequalities and team performance: Empirical evidence
from the North American major leagues. International Journal
of Manpower 24: 472-491.

Appendix:

Data with documentation and results: MLB.xls

(back to the top)

* I would like to thank Mary Mar, Youghwan Song, Stephen Schmidt and two anonymous referees for their helpful
comments. I am also grateful to many Union College students for their useful feedback.
:
: