I am having a discussion with a colleague to define the metadata for data management protocol. The problem is that I cannot define the difference between data and metadata. For example, if I take the measurement of the size of a seagrass plant in a location, in that location I would also take the water depth. So, is water depth a metadatum or a datum? And why?
-
3Why can't you define it how you want? What does your colleague think? Why is it important to distinguish - what are you going to do differently about data vs metadata that makes the difference relevant?– Bryan Krause ♦Sep 15, 2021 at 17:06
-
3Wikipedia actually has quite a useful article about this: en.wikipedia.org/wiki/Metadata– Wolfgang BangerthSep 15, 2021 at 17:17
4 Answers
No, the water depth measurements are data.
Metadata is data about the data. That would include things like the units of the numbers (e.g. cm for length), the equipment you used, the date and time of day when you took the measurements.
The location where you measured the sea grass might be metadata or data - depending on whether you want to compare results from different locations or just document the data from this spot.
-
So what is a data (or metadata) is dependent on the research? Specifically, it is the measurement that would be investigated in the research. In my case, the seagrass length. However, what it would be for me a metadata to describe the location of the data (depth), it can be a data for someone looking at the bathymetry of the area.– TeuzSep 20, 2021 at 9:42
-
@Teuz Yes, sometimes. Metadata for a photograph might include the kind of camera. But if you were doing research on how cameras perform that would be data. Sep 20, 2021 at 11:06
-
@Teuz Very hands-on approach: assume you have all the data but all the metadata was lost. Does your experiment still make sense? Also, sometimes the difference is totally contextual: for an instance, I have images taken by a digital camera and what is called metadata by convention (exposure, location, ISO...) is crucial to my experiment. It is thus all data to me but it's easier to still call it image metadata so others understand what I'm talking about. It really only may make an impact on how you store the experiment data, the rest is semantics - just make sure everyone's on the same page.– LodinnOct 17, 2021 at 2:20
Metadata depends on the context, or the kind of research being done.
As the term suggests, it lies outside the primary data of a study. It consists of data, but of a different sort or type. From the American Heritage Dictionary: "Data that describes other data, as in describing the origin, structure, or characteristics of computer files, webpages, databases, or other digital resources."
For a clinical trial, metadata would include the trial design, including background, definitions of terms, measurement methods, schedule of data collection, analysis, outcomes, etc. as well as some results such as participation rates, dropout rates, etc. This includes anything that others might need to understand the implications of the results.
For a survey, metadata includes the sample and interview design, data linkages, background, definitions of terms, measurement methods, schedule of data collection, analysis, etc, as well as some results, eg, contact rates, refusal rates, etc. Anything that others need to understand the implications.
Think about giving your data to someone else to use for further analysis. You want them to have information that keeps them on track regarding effective analysis. If you don't give them something critical, eg, "80% of the subjects came from one medical center and the rest were spread across 27 centers." This is very different than "the percent of subjects from 28 centers ranged from 3% to 6%." The analyses they do should be different in each case. In short, any documentation you hand them with the data itself is metadata.
Some items of metadata are recorded as data, eg, withdrawal "time" relative to the data collected and missing. It might not be just a time observation. These items can indicate the quality of the execution of a study and might influence the analysis, but they are not directly related to the goals of a study.
Metadata includes a lot of stuff that isn't data and can include stuff that is collected and recorded as data. The latter might be part of your confusion. There is no simple conceptual test as to whether specific information about respondents is data or metadata but it probably doesn't matter a lot if you call some items both data and metadata.
Metadata is not made up to fit a study. It is accurate information about the design and execution of a study.
I don't think there's a general answer to this question (hence the lack of any consensus over the past month). Metadata by definition is "data about the data", but what that means is probably more of a field-dependent convention rather than any deep truth that can be derived from first principles.
Two examples:
- In the experiments at CERN (at least, the ones I worked on), we didn't really have any concept of "metadata." Everything the machine recorded was data, even timestamps and "environmental" conditions. We had simulated data, and results, but we certainly never considered these "metadata."
- When it comes to image processing, pretty much only the images are considered "data." Maybe the labels as well, though I would normally call the labels "labels" rather than trying to group them under data or metadata. Anything else (e.g., the time, location, type of camera, camera settings) would be considered "metadata."
In your case, I think you will have to define what you mean by metadata and then answer your question from there. For example, if you are using an underwater gizmo, you could define "data" to be "records from the gizmo" and everything else to be metadata. Or, you could define metadata to be "the gizmo's operating parameters", and everything else is data. Any reasonable choice should be fine so long as you state it clearly and are consistent.
No, the water depth would also be data and you can look for correlations, for example, between the size and the depth.
But all descriptive statistics are metadata: mean, mode, quartiles, standard deviations, and all the rest. They are data that summarize the basic data. Correlations between size and depth would also be metadata. If you take a lot of samples, then you get the potential for a lot of metadata and it can be analyzed as well. Understanding this is fundamental to the understanding of making predictions with statistics; statistical inference.
You can, of course, make up additional metadata to fit your study if needed.
But in general, metadata is just data that is "about" other more fundamental data or depends in some way on that data. It isn't independent data.
-
1Statistical results themselves are nut usually regarded as metadata. Sep 16, 2021 at 23:49
-
@DavidSmith, perhaps just a semantic distinction. It is what they are, not how they are "regarded". In fact, your own answer says exactly this: "describing the origin, structure, or characteristics...". Descriptive statistics describe the structure and characteristics of some set of data. Please be consistent, at least.– BuffySep 16, 2021 at 23:51