I appreciate Dr. Ha Hoang for raising this interesting question. Furthermore, Prof. Ronán Michael Conroy manipulate the issue perfectly when he said:

"They are trying to invent questions to fit the data they have instead of collecting data to answer the questions they have."

Really, he evidently thinks very deeply and hit the target!

I get this from time to time. A person comes in with a database with lots of information, keen to write papers. The problem is that the are trying to invent questions to fit the data they have instead of collecting data to answer the questions they have. It's like looking at a lot of ingredients and wondering what you can cook with them. The better solution would have been to choose a recipe and then go out to get the ingredients!

You should start by itemising the areas in which you have data. Read the literature in these areas and see what are the current 'hot' questions. Now go back and see if you have data that can throw light on these questions. Your data can be helpful either

a) by providing a lot of data and so increasing the precision of our knowledge,

b) by providing measurements of variables that are seldom measured in the same study, thereby allowing relationships to be explored, or

c) by providing data on a country, culture or population group that has not so far been studied much, or at all.

For example :

A clinical colleague assembled a database of 1,000 hospital admissions, carefully documented. Although there was nothing really new in the data, the size of the study allowed us to publish a series of papers that provided much more precise estimates and to control for potential confounders.

On the other hand, I once found that a large survey of older people had an item on boredom. This interested me. I discovered that there was almost no literature on boredom, and went on to write a paper on it, looking at the factors that were correlated with it and finding some interesting patterns. Strangely, no-one on the original research project knew why they had asked about boredom or who had put it on the survey.

As an example of the last one, we published some papers on health anxiety, which had previously been studied in people who were clearly anxious about their health and were frequent attenders at their doctor. We carried the research into the realm of ordinary patients, and discovered that the relationships that had been found in the very anxious were there in a similar way among routine patients.

Dear Ronán Michael Conroy, thank you for your answer. It's true that if we try to invent the question to fit the data, it's not acceptable and unethical. However, when design a survey for collecting the responses, we could integrate the questions/items (that may not really need for the current study but for another study). So in this case it is another story?

I don't agree that inventing a question to suit your data is unethical or unacceptable! Many excellent papers have leveraged data collected for other purposes.

When designing a survey, beware of putting in questions just because they might be interesting some day. There are two reasons why not. The first is practical : the longer your survey, the lower the compliance rate, and the higher respondent fatigue so you get people just checking responses absent-mindedly without thinking any more.

The second reason is that when you actually want to do a study of a particular factor, you will know which questions you need to ask and what format you need the answers in. By putting in questions 'just in case' you ever need them, you risk not being able to use them because you didn't put in a question that later turned out to be important, and/or you asked the question the wrong way.

For example, you might have asked about medications but not asked why the person was taking them – I encountered a study where we needed to differentiate between people taking the same medication but for different symptoms and we couldn't because no-one had noted the information at the time – end of a promising research idea!

So in a survey have a purpose for every question, make it as short as possible and – here's a tip – ask the interesting questions first, then the sensitive questions, and last of all ask the demographics, which are dull and intrusive and don't seem all that relevant to most people.

Dear Ronán Michael Conroy , Thank you Sir for your tip, and I agree that the survey should be kept as short as possible!

It is an excellent question. Personally I see no problem writing multiple papers from your "very good database" as long as you avoid 'salamization' where you divide the data into finer and finer 'slices' simply for the sake of multiplying papers artificially.

To answer your question in detail would require knowing the type of data, the sample, the original hypotheses (if there were any), any other reasons for collecting the data, the number of variables, the statistical analyses used, etc etc. It is difficult to answer in the abstract, but in principle should not be a problem.

The data have to be interesting and relevant to a particular question, issue or point of view. For example, standing on a motorway bridge and counting the number of red, blue, green...etc cars passing under it could be a waste of time.

Unless you have a hypothesis about an association between car colour and driving speed, e.g red car drivers like driving at a faster speed than blue car drivers.

If you don't know what to do with your data, one approach would be to identify a theory, set of hypotheses or testable predictions and to put your data alongside the theory. The outcome could be rather magnificent and leads to a large number of publications pro or con the theory.

Without knowledge of the dataset, however, I cannot say more at this stage about the theory or set of hypotheses you might need to be looking for.

Dear David F Marks

Thank you very much for your kind contribution.

Regards.