Data

Abstract

This chapter portrays the economic analysis of data as a sophisticated way of “seeing like a state,” a perspective which highlights the inherent limitations of most economic data. It articulates the data qualities the researcher should be familiar with, and the econometric consequences of failing to understand these qualities. These ideas come to life in applications to patents, informal markets in Peru, school accountability ratings, drug dealing, medical coding, and the employment effects of the minimum wage.

Food for Thought

• 1. The introduction to this chapter gave two examples in which intuition trumped formal measurement, and two counterexamples that went the other way. This discussion side-stepped a book, excerpted in an earlier chapter, whose entire theme was the tension between subjective intuition and objective measurement. What book? What did it conclude?

• 2. Table 5.1 is one of many similar entries in the classic book, The Philadelphia Negro: A Social Study (1899). (I’m not sure why the totals in the table are occasionally off a bit.)

  • (a) Using this table, detail the ways in which this book’s author uses measurement to legitimize an overlooked people, as did DeSoto.

  • (b) Who authored this book? Was it important to this author to legitimize these people?

• 3. In a recent cross-sectional analysis of partisan primaries in Texas (Grant 2017), I used the following county-level controls: the fractions of the population that are Anglo, black, and Hispanic; the percentage of adults with at least a high school diploma and with a college degree; the percentage of housing that is owner-occupied; median age; per capita income ; the unemployment rate; mean annual rainfall; and the logarithms of the value of agricultural production, population, area, the number of registered voters, the number of voters in the election being analyzed, and the number of votes received by that party’s 2012 Presidential nominee.

  • (a) Sketch out the control space spanned by these variables. Which elements of the control space should vary the most across Texas counties? Which should vary the least?

  • (b) The key independent variable in this paper is the order in which each candidate is listed on the ballot. By law, this must be determined at random within each county. What, then, is the primary benefit of including controls in the analysis?

  • (c) If your answer to part (b) was “by accounting for more variation in the dependent variable, you get more precise estimates,” then you are seeing (or thinking) like a state. Scan the paper’s introduction, and then answer this question again.

• 4. The North American Industrial Classification System (NAICS) extends, in some cases, to six digits: a pretty fine level of detail. This gives analysts the option of controlling for industry using broader, higher-level one- or two-digit dummies, or finer, lower-level three-or-more-digit dummies. Review the NAICS online at Statistics Canada’s website, and then answer the following questions.

  • (a) In deciding the level of detail to use in controlling for industry, one must trade off data precision against degrees of freedom and data accuracy . Expound on this tradeoff . What levels of detail are used most commonly in papers in labor economics? Industrial organization ?

  • (b) Consider how this tradeoff might differ for industry dummies used in the following contexts : (1) to control for working conditions in a wage analysis, (2) to control for market power in a pricing analysis, and (3) to account for industry-specific systemic risk in estimating stock returns using the capital asset pricing model.

• 5. Sometimes the easiest way to recognize the issues that can arise with formal measurement is to compare multiple measurements of the same general thing.

  • (a) In addition to the formal unemployment rate, the Bureau of Labor Statistics (BLS) reports five other measures of labor underutilization. Speculate on what these other measures might be, and then check your work on the BLS website. Does the formal unemployment rate see like a state more or less than these other measures?

  • (b) Three sources measure the number of homicides in the U.S.: the Uniform Crime Reports, the Supplemental Homicide Reports, and the Fatal Injury Reports. Identify the main differences in the way these data are compiled, and compare the strengths and weaknesses of each measure.

  • (c) Consumer inflation is measured using the PCE Deflator and the CPI. Identify three important differences in how the two measures are created. Which measure do you prefer?

  • (d) The text analyzed a binary variable that was measured in error 2% of the time. Do the homicide measures in part (b) differ by more than 2%, or less? What about the inflation measures in part (c)? Are estimates using alternative measures of homicides or inflation likely to differ by more than 2%, or less?

Table 5.1 “Colored Methodist Episcopal Churches in Philadelphia, 1897” (title of the original)