Because data collection can be repetitious, time-consuming, and tedious, its importance can be underestimated. Care should be taken to assure that those responsible for collecting data are adequately trained and motivated, that they employ methods that limit or eliminate the effect of bias, and that they keep records of what was done by whom and when.

The best model for recordkeeping will not be the same for all areas of research. However, nearly all types of research in the physical and natural sciences include records that should be kept in bound lab or field notebooks. At a minimum, such notebooks can provide a listing of the date of research, the investigators, what was done, and where the corresponding research products can be found. The lab or field notebook should be supplemented as needed by specialized methods of recordkeeping such as digital files, audio/video recordings, and gels.

The use of statistical methods varies widely among research disciplines. Although the successful cloning of a gene may require no statistical analyses, biology is characterized by variation and therefore most inferences will depend on statistical methods to quantify confidence in accepting or rejecting hypotheses. Such testing depends on many experimental and statistical assumptions. Violation of those assumptions, or a misunderstanding of the methods of analysis, can result in significant, even if unintentional, misrepresentation of the results of a study.

Because it is not possible to report everything that has been done, researchers must make decisions about which studies, data points, and methods of analysis to present. Although some of these decisions are easy, many are not. For example, should an anomalous data point be excluded from analysis when there is no known reason for the discrepancy? Most researchers would favor retaining the data point, but some fields have criteria for excluding such outliers. It would be irresponsible to exclude such a data point without clearly reporting the use of the exclusion criteria.

It is a laudable ideal to analyze and report all data; however, in practice some data must be excluded. The selection should be based on objective criteria, preferably ones specified before data collection. Critically evaluate the reasons for inclusion or exclusion of data, the measures taken to avoid bias, and possible ways in which bias may nonetheless influence selection. Clearly document how the data was obtained, selected, and analyzed-- especially if the methods are unusual or potentially controversial.

Because the products of research involve creative contributions to new knowledge, it is easy to assume that the resulting data are different from the routine products of employees in any other private or public institution. Although the language and practice of science seems to suggest otherwise, the equipment, materials and reagents, and even resulting data all belong to the institution in which they are purchased or produced. The issue becomes especially salient if a marketable product is produced, but it is also an issue when someone moves from one institution to another. If the principal investigator is moving, she or he can normally expect to take the data. However, exceptions do occur, and equipment transfer is nearly always a matter for negotiation. Absent some explicit agreement or ruling to the contrary, the principal investigator has primary responsibility for decisions about the collection, use, and sharing of data. Student or postdoctoral researchers should assume that their original data will stay with the principal investigator. However, most institutions have the expectation that graduating students may take copies of their research records. If regulations preclude researchers from taking such copies, then the principal investigator is responsible for making this clear to research group members before work begins.

This is a good discussion of data retention issues.

The quality of data supporting published work becomes moot if the data is lost. This raises issues of what should be retained, how it should be stored, and for how long.

What should be retained?
This depends in part on the nature of the products of research. Some materials, such as thin sections for electron microscopy, cannot be kept indefinitely because of degradation. It is also impractical to store extraordinarily large volumes of primary data. At minimum, enough data should be retained to reconstruct what was done.

How should it be stored?
Any stored data will be rendered useless if there are insufficient records to locate and identify the material in question. Ease of access must be balanced against security, for instance if the study involved human subjects with a reasonable expectation of confidentiality. Although the institution is the legal owner of the data, it is usually the responsibility of the principal investigator to ensure that records are stored in a secure, accessible fashion.

How long should it be kept?
Under current Health and Human Services requirements, research records must be maintained for at least three years after the last expenditure report. Federal regulations or institutional guidelines may require that data be retained for longer periods. These formal requirements are minimal constraints. Decisions about retention of records should take into account the extent to which a line of research is still being pursued, the likelihood of ongoing interest in the research, continued assurances of confidentiality for any human subjects, and the space and expense necessary for storage.

 

Although sharing of data is generally in the best interests of science and the individual, it is clear that such sharing can place an individual scientist at risk. It is reasonable to fear that sharing data before publication can result in loss of credit or opportunity. Other concerns include exposure of data to the prejudiced scrutiny of competitors or detractors, risk of compromising confidentiality of human subjects, and expense of time and resources to meet requests for sharing of data. However, reasonable strategies to minimize potential problems should make it possible to choose sharing over secrecy. Before publication, it is best to maintain an open data policy with appropriate caution. After publication, be prepared to grant reasonable access to the raw data; that is, honor requests that are in the interest of scientific inquiry and can be accomplished without inordinate expense or delay.