WIDENING WAGE GAP

Investigative Reporters and Editors, Inc. The IRE Journal, Summer 2009 by Knox, David

Fellowship offers tools for economic analysis

Bad economic news from Wail Street to Main Street has dominated the media for months, but the broader debate over whether the middle class is endangered has raged for decades.

In this brawl, newspapers largely sit on the sidelines. Most reporters - myself included - don't have the education or expertise needed to evaluate the reams of statistics and the conflicting interpretations of academic and think-tank studies.

That frustration motivated me to go at the story directly by doing original research as part of a six-month Kiplinger journalism fellowship at Ohio State University.

The immediate aim of the research was to track median wage and salary earnings, which make up three-quarters of all household income, by generations across more than a half century of census data. But the ultimate goal was to build a solid statistical foundation for an in-depth examination of the state of the middle class in rust-belt Ohio and the nation.

The resulting yearlong project, "The American Dream: Hanging by a Thread," found that the middle class is literally shrinking as a percentage of all workers - squeezed on two fronts by steadily decreasing earnings and dramatically increasing costs of the hallmarks of the middle class: home ownership, higher education, affordable health care and secure retirement.

In systematically exploring these issues, the series produced previously unreported findings, including a dramatic loss of one-in-four managerial jobs in Ohio in the past decade, the failure of high-tech industries to generate jobs, and a widening health care gap between Ohio and its geographically and demographically close neighbor, the Canadian province of Ontario.

How the project started

I first learned the power of looking at demographic data through the prism of generations while reporting on the 2000 Census. Although Ohio had grown in population by less than half the national average since 1 990, it was far from the bottom of the list.

But when I grouped the population counts into generations according to year of birth, I found Ohio had lost more baby boomers and members of Generation X - the core of the work force - than any state.

The resulting story drew on Internal Revenue Service migration data, which is available from the IRE and NICAR Database Library, that showed those who left Ohio reported higher household incomes the following year. This led me to suspect that poor job prospects for younger workers were driving the exodus. Other projects I did in 2005 and 2006, which were based on U.S. Bureau of Labor Statistics data, provided more evidence of eroding paychecks across the job spectrum.

Pursuing the story further required more resources than I had available at the Beacon lournal. The Kiplinger fellowship provided the tools, time and expertise I needed.

The project's dataset was drawn from census microdata samples, as compiled in the Integrated Public Use Microdata Series of the Minnesota Population Center.

Unlike the usual census summary tables, IPUMS data are the answers given by individuals on the census "long-form" questionnaires, which are stripped of names, addresses and other information for privacy.

For the earnings study, the IPUMS samples came from the decennial censuses of 1950-2000 and the American Community Survey for 2001 -2006.

After downloading the datasets, I used SAS statistical software to filter the 51 million records to include only persons 1 6 years and older who lived in a household and who reported some wage or salary income. The datasets were further limited to those who reported working at least 40 weeks during the sampled year.

To track how paychecks changed throughout the years, a simple methodology was used that could be explained to readers: A series of SAS programs sorted the annual inflationadjusted wages to determine the median earnings of workers - half taking home more and half less - grouped into five-year brackets from ages 20 to 64 and also by gender, race and level of education.

The age groups for each demographic group were then classified by birth years as belonging to the "Cl generation," born from 1905-1924; "depression kids," born from 1925-1944; "baby boomers," born from 1945-1964; or "Generation X," born from 1965-1984.

The main reason to use IPUMS is the large sample size 18 million records were examined, which allows narrowly focused cross-tabulations with small margins of error.

For example, the 2005 ACS dataset included more than 21 ,000 women, ages 30-34, with a four-year degree or better. Their median annual earnings: $42,083, with a margin of error of plus or minus $510- less than 1.2 percent - at the 95 percent confidence interval.

The downside of such large datasets is the need for more advanced, therefore expensive, software to handle the data. One of the many advantages of an on-campus fellowship is the availability of virtually any software tools you need and researchers willing to help you use them.

I followed the advice of several journalists and social science researchers and used SAS statistical software for this projectand never regretted the choice. In addition to filtering, recoding and sorting the large datasets, SAS programs were used to generate frequency distributions needed to calculate margins of error for each earnings median, using procedures recommended in the Census Bureau's IPUMS technical documentation. Unless I win the lottery, though, SAS won't be returning to my toolbox anytime soon. After finishing the fellowship, I checked into a single-user commercial license for SAS. I was told it would cost $7,200 up front, plus an annual fee of about $2,400.