The
National Establishment Time-Series (NETS) Database, a longitudinal database of linked Dun & Bradstreet records, is one of the standout data sources to emerge in the last half decade for the study of entrepreneurship (and businesses, more generally). As an early proponent of the potential of this data, I wanted to offer some commentary now that comes from grantees that have used the NETS data for different purposes such as matching to other data sets, surveys, or just aggregate examinations of firm dynamics. Every data set has its advantages and disadvantages, NETS is no anomaly in this regard, but one of the unique things about my role is that I get to see some of these trends and connect experiences. These comments emerged as the result of an email inquiry I had made to those listed here for a potential new project. I found the comments so valuable I decided they should be published as blog post so that others could see them. All the commentators were allowed to review and modify their comments before posting. I would encourage others who have experiences that can help to hone the use of NETS and hopefully drive improvements to its core to add them as comments to this post.
“The strengths of the NETS are its historical address information, descriptive data (industry, incorporation codes, and CEO and ownership demographics) and survival measures. These traits seem sufficient to conduct event history analysis using quasi-experimental methods.
If your focus is on other types of performance such as growth, then I'm not convinced this is the best dataset. I found in my research that employment and sales figures in the NETS are often estimates and not actual observations of the levels of these measures. It presents a problem when you are dealing with small firms because you lose variation since so many of these small firms appear to reach stability over short periods of time. I'm still using the data for this purpose but I take the results to be more speculative than definitive.
Also, the lack of information on other attributes of organizations (i.e. invested capital, costs, revenue streams, and the education and experience of CEOs) makes it hard to account for unobservable differences that may be the underlying cause of performance when evaluating a policy outcome. There are statistical methods to address these issues but these also open the door to skepticism and criticism.
I'm working on ways to improve on the value of the NETS by merging it with other datasets to add richer data to what I already use. Depending on the types of firms being studied, one may improve on the types of controls that can be accounted for and on the type of performance being evaluated.”
“I agree with Alejandro that NETS is not the best data to see changes in firm level outcomes. In our work we found that the information on sales and credit scores are actually the best of their outcomes variables, and are updated reasonably well. But especially employment variable are very sticky in NETS and do not seem well covered in NETS.
One of the best data set that you could use is probably LBD data at Census, but this data is difficult to access since you need to apply for access.”
“I am less skeptical of the employment measures than Alejandro and Antoinette, although it depends how you are using the data. At the individual establishment level there is clearly a lot of stickiness. But when you are averaging across many establishments (like in our
EZ paper) this doesn't really matter. I think that if you are matching based on geographic location the NETS are very useful although the geocoded information that comes provided with the data might need to be re-geocoded.”
See select other posts dealing with NETS:
Comparing Business Registers;
youreconomy.org Updated