What Do You Mean the Data Are Suppressed? Understanding the Ins and Outs of QCEW Disclosure Rules

Perfect information: it's an assumption we would all like to make when looking at a dataset; but life is not neat and tidy and neither is our data. In order to protect the confidentiality of company-specific information, not all data collected by the government are released to the public. You can call it suppression, nondisclosure or withholding, but whatever the name, it can lead to inaccurate analysis if you don't know what to watch for. This article will look at the disclosure rules for one of the most frequently used data sets: the Quarterly Census of Employment and Wages (QCEW), from the Bureau of Labor Statistics (BLS).

Suppression and What to Do about It

You are most likely to run into suppression issues when looking at a small geographic area (such as a county or a region that is built using county-level data), an industry with few employers, an industry that is dominated by a single employer, or one where state and local government plays a large role (such as educational services).

There are various levels of suppression, so the first step in properly analyzing your data is to know why the information is being withheld. This is especially important if you are doing comparisons among places that you are not familiar with and, therefore, might not know if they have a major employer in a given industry.

One quick way to check that out is by using a tool developed by the Indiana Business Research Center called the Simple Business Lookup available on the Hoosiers by the Numbers website (www.hoosierdata.in.gov/BusLookup.asp). This tool uses publicly available data to show employers in Indiana's regions and counties by size and industry.

Primary Suppression

Primary suppression (dubbed the 80/3 rule) occurs when one of the following conditions is true:

However, it is important to also be aware that, “at the request of a state, data are also withheld where there is reason to believe that the ‘fewer than three' rule would not prevent disclosure of information pertaining to an individual firm or would otherwise violate the state's disclosure provisions. Information concerning federal employees, however, is fully disclosable.”²

Another issue to consider is that the 80 percent rule applies at the account level, so if a company has multiple establishments, the entire firm's employment in a given geography is combined across establishments to see if primary suppression applies.

As an example of how the Simple Business Lookup can help you parse out these issues, Table 1 shows that Martin County's mining sector (NAICS 21) is not disclosed because Indian Creek Stone Products is the only employer in that sector (criteria #1); meanwhile, the data for furniture and related product manufacturing (NAICS 337) in Jasper County is most likely nondisclosable because of the dominance of Sealy Components (criteria #2).

Table 1: Martin County Mining and Jasper County Furniture Businesses

If the data are not disclosed because there are a small number of establishments with relatively few employees, it should not negatively impact your overall analysis if you basically ignore them. However, if you're dealing with a nondisclosed major employer, you will want to keep that in mind when calculating percent of total employment or when comparing to other geographic areas.

Secondary Suppression

Secondary suppression occurs when the value of certain withheld data may be discernable through simple calculations of other released data. For example, if data for one industry group in a county are suppressed, then data for a second industry group in the county (the one with the smallest non-zero employment figure) must also be suppressed. 

Moreover, “total covered employment” is really an aggregation of the four different types of ownership: private, local government, state government and federal government. Ultimately, for the data coming from BLS, the disclosure rules kick in at these individual ownership levels and not at the total covered employment level to prevent you from using the pieces to calculate the missing values.

The educational services sector illustrates this issue nicely. Table 2 shows that Marion County has 93 local government-owned establishments and six state government-owned establishments in the education sector; however, BLS does not disclose these data because one state employer surpasses the 80 percent threshold. Therefore, they do not publish the data for local government-owned establishments either to prevent one from calculating the suppressed state-owned numbers. (One can argue that government employment is public record so it should always be disclosed, but that is a topic for another day.)

Table 2: Educational Services in Marion County, 2007:3

This suppression by ownership is especially important to consider in sectors where government plays a prominent role (think education and health care), as well as when you are comparing a county to another geography, such as the state or the nation. The U.S. or state numbers will be a true calculation of total covered employment with no disclosure issues. However, if some of the underlying ownership codes are not disclosed, the county's total covered employment for a given industry may really only include private data.³ As a result, it would be inappropriate to compare the two and it would be better to compare total private employment for each geographic area instead. You can access data by ownership from Hoosiers by the Numbers or the BLS website (see Table 3).

Table 3: Online Sources of Quarterly Census of Employment and Wages Data

Not as Straight-Forward as It Sounds

BLS does not disclose their detailed methodology for suppression because they want to prevent anyone from reverse-engineering the data in order to get to the suppressed numbers. So basically, we don't really know all the little things they are doing to protect confidentiality.

Each state's workforce agency, such as the Indiana Department of Workforce Development (IDWD), has its own disclosure rules that may differ from those of BLS. Therefore, it is quite possible to see different things disclosed depending on where you look. For example, STATS Indiana uses both QCEW data based on the BLS guidelines, as well as data based on the IDWD guidelines. This is not to introduce confusion but to provide users with the best output targeted to their needs. BLS data allows comparisons across the nation, whereas the data based on IDWD guidelines is subject to less withholding in some cases, thus providing a more detailed picture if numbers within the state are all you care about.

Table 4 looks again at Marion County's education sector, and compares what you get depending on which guidelines are used. Obviously, the total employment figures based on the IDWD guidelines does not encounter the same disclosure problems for the ownership classes as the BLS data because it applies the disclosure guidelines to total covered employment and not at the individual ownership levels (since those are not reported, it is unnecessary to do it at that level of detail); therefore, this gives you a better picture of what the industry actually looks like. (If you happen to also be curious why the number of establishments is lower in this STATS Indiana output, it is because establishments that report zero employment and wages are excluded from the data set.)

Table 4: Educational Services in Marion County from Two Sources, 2007:3

What's a Data User to Do?

To make a long story short, it pays to look a little closer at the data that are not disclosed. See if you can figure out why they are not disclosed and your analysis will be the better for it (is it because of few establishments, a dominant employer or perhaps secondary disclosure issues?). Understand why you might get different numbers from different places and know which output you need to use when. After all, the devil is in the details.

Notes

Rachel Justis, Managing Editor
Indiana Business Research Center, Kelley School of Business, Indiana University