Error

There are many horror stories of researchers losing their data. You can just plain lose your laptop or an external hard drive. Sometimes they are confiscated if you are traveling to another country — and you may not get them back. Some errors are more nuanced. For instance, a COVID-19 repository of contact-traced individuals was missing 16,000 results because Excel can’t exceed 1 million lines per spreadsheet.

Do you think a hard drive is the best repository? Keep in mind that 20 percent of hard drives fail within the first four years. Some researchers merely email their data back and forth and feel like it is “secure” in their inbox.

The human and machine error margins are wide. Continually backing up your results, while good practice, can’t ensure that you won’t lose invaluable research material.

Repositories

According to Reid Boehm, Ph.D., Research Data Management Librarian at the University of Houston Libraries, your best bet is to utilize research data repositories. “The systems and the administrators are focused on file integrity and preservation actions to mitigate loss and they often employ specific metadata fields and documentation with the content,” Boehm says of the repositories. “They usually provide a digital object identifier or other unique ID for a persistent record and access point to these data. It’s just so much less time and worry.”

Integrity

Losing data or being hacked can challenge data integrity. Data breaches do not only compromise research integrity, they can also be extremely expensive! According to Security Intelligence, the global average cost of a data breach in a 2019 study was $3.92 million. That is a 1.5 percent increase from the previous year’s study.

Sample size — how large or small a study was — is another example of how data integrity can affect a study. Retraction Watch removes approximately 1,500 articles annually from prestigious journals for “sloppy science.” One of the main reasons the papers end up being retracted is that the sample size was too small to be a representative group.

Replication

Another metric for measuring data integrity is whether or not the experiment can be replicated. The ability to recreate an experiment is paramount to the scientific enterprise. In a Nature article entitled, 1,500 scientists lift the lid on reproducibility, “73 percent said that they think that at least half of the papers can be trusted, with physicists and chemists generally showing the most confidence.”

However, according to Kelsey Piper at Vox, “an attempt to replicate studies from top journals Nature and Science found that 13 of the 21 results looked at could be reproduced.”

That's so meta

The archivist Jason Scott said, “Metadata is a love note to the future.” Learning how to keep data about data is a critical part of reproducing an experiment.

“While this will be always be determined by a combination of project specifics and disciplinary considerations, descriptive metadata should include as much information about the process as possible,” said Boehm. Details of workflows, any standard operating procedures and parameters of measurement, clear definitions of variables, code and software specifications and versions, and many other signifiers ensure the data will be of use to colleagues in the future.

In other words, making data accessible, useable and reproducible is of the utmost importance. You make reproducing experiments that much easier if you are doing a good job of capturing metadata in a consistent way.

The Big Idea

A data management plan includes storage, curation, archiving and dissemination of research data. Your university’s digital librarian is an invaluable resource. They can answer other tricky questions as well: such as, who does data belong to? And, when a post-doctoral student in your lab leaves the institution, can s/he take their data with them? Every situation is unique and deserves a one-of-the-kind data management plan, not a one-size-fits-all solution.