Doing Economics

2.3 Data and Descriptive Statistics
After developing your theory of change, you have presumably gone in
search of data to test the predictions of that theory. As with writing formal
theoretical models, entire books have been written about the dos and don’ts
of data collection (see Deaton 1997 or Glewwe and Grosh 2000 for survey
data, and Gerber and Green 2012 or Glennerster and Takavarasha 2013 for
randomized controlled trials), so this section will not discuss where the data
come from, and assume that you already have them. Rather, this section will
focus on how to present your data in the context of an economics article.
The best Data and Descriptive Statistics sections answer all of the
reader’s questions about the data. Specifically, a good Data and Descriptive
Statistics section first discusses where the data come from, when they were
collected, by whom, how the observations that compose the sample were
chosen for inclusion (i.e., the survey methodology, or how regions,
communities, firms, households, individuals, etc. were all chosen), what
population the sample is representative of, what the target sample size was
and how that sample size was determined (e.g., via power calculations),
what the actual sample size is, what the nonresponse rate was, what the
attrition rate is if the data are longitudinal, and how missing data were dealt
with (e.g., whether observations were simply dropped, or whether some
values were imputed and, if so, how the imputation was done). Broadly
speaking, the information presented here allows the reader to judge the
external validity of the results contained in a paper (and sometimes their
internal validity, as is the case when the data suffer from attrition), or how
those results might be used for out-of-sample predictions.
After presenting those basics, a good Data and Descriptive Statistics
section introduces all the variables used in the paper (and no variable not
used in the paper) by precisely and concisely explaining what they measure,

---

and how they do so. For instance, people often derive their income from
many difference sources. So if an “income” variable is included in the
analysis, the reader needs to be told what the various income sources are.
This may seem tedious—and if it seems tedious to you as writer, imagine
what it is like to the reader—but it can nevertheless contain crucial
information.
The good news is that it is relatively easy to present that information
when one has access to the survey questionnaires that were used to collect
the data, which is almost always the case. Moreover, one way of presenting
that information optimally is by creating a table of variable descriptions,
where each line is a specific variable retained for analysis, where the first
column gives the name of that variable (and the unit of measurement in
parentheses), and where the second column gives precise measurements.
Figure 2.1 shows one such table. This allows presentation of a lot of
required but tedious information in a compact manner, which minimizes
reader discontent: those who want to know all there is to know about the
data can read the table, and those who do not can just skip it to focus
instead on variable names.

---