Research is a particular activity with a methodology. Here are some notes about doing research.
Research is a disciplined, directed way to create knowledge. Knowledge leads to understanding, which puts us in a better position to change how things are to be better. Put another way, research is an activity you are motivated to do to solve a problem.
Getting from a vague sense of a problem to a relatively well-bounded research question is the first step.
A research question is a statement, in the form of a question, that identifies the phenomenon to be studied.
Research questions apply to specific topics (e.g. diagrams in design, FEA for analysis of bone, underwater robots, etc.). We assume here you have already established a suitable topic with your academic adviser.
Three tasks in generating a good research question:
Also, as you carry out these tasks, you should ask yourself (and answer!) these questions:
Characteristics of a good research question:
While you're developing your research question, remember to:
There are different kinds of research questions, not all of which will make sense for particular topics. Still, it can be useful to get a couple of research questions of each kind, and running them by your academic adviser.
Factual research questions: these are popular in “scientific” studies, where you simply look for causal relationships and describe them. For example:
Paradoxical research questions: these describe an apparently contradictory situation and suggest you will resolve the contradiction. For example:
Hypothetical research questions: these help us understand why things are the way they are now, by considering how things would be different if something in the past had happened differently. These kinds of questions are not often used in design research. For example:
Predictive research questions: these help us make decisions by constructing “scenarios” of how things might be in the future. For example:
Problem-solving research questions: these propose specific solutions to existent problems. For example:
Comparative research questions: these identify (usually) two alternatives to a situation and compare the alternatives in actual practise. For example:
Judgemental research questions: these require rational and rhetorical arguments to answer the question instead of scientific or predictive methods (that is, no definitive answer in a scientific sense is possible). For example:
Bad research questions include the following. (Can you tell why they're bad research questions?)
Research questions should always be quite specific, but even then there is great flexibility in just how specific your research question should be.
You can always make your research question more specific, or less specific. Here are two examples:
|Topic||Frank Lloyd Wright and modern architecture|
|A research question||How has Frank Lloyd Wright influenced modern architecture?|
|A more specific research question||What design principles used by Frank Lloyd Wright are common in contemporary homes?|
|Topic||nutrition, children, health education, health risks|
|A research question||How does nutrition education help children?|
|A more specific research question||What are the major health risks related to diet for school aged children?|
|Another specific research question||What are the benefits of nutrition education for school children?|
Many research questions imply causality - cause and effect - between the concepts in it.
Remember, though, that correlation alone does not imply causation.
For example, consider the research question: Do concept maps help design engineers collaborate during early designing tasks?
But since this is only a research question, you don't know for sure if the causality is real, or just an artifact of how you phrased the research question.
There are three main criteria for inferring a cause and effect relationship:
So when you are thinking about causal relationships in your research questions, you have to consider how you will validate that causality, as defined above. This too becomes part of your research project.
Causal analysis can be very hard and take a long time. Causal analysis and validation could be your entire research project! You might have to conduct many experiments to discount alternative causal relations, or conduct a variety of analyses - mathematical or otherwise.
In the example above – Do concept maps help design engineers collaborate during early designing tasks? – you would have to search for other phenomena associated with the introduction of concept maps to a group of engineers that might cause them to collaborate, regardless of the use of concept maps themselves.
Causality can feed back on itself. If concept maps improve collaboration, the increased collaboration may drive further use of concept maps, which further increases collaboration - to a point. You have to be wary of feedback, both positive and negative, because it makes the whole system unstable. For instances, designers can collaborate too much and end up not getting any work done, in which case collaboration becomes a cost and not a benefit.
The second step of a research project is turning the research question into a hypothesis (or a small set of related hypotheses). Rarely do specific projects generate more than three hypotheses for a given research question.
A hypothesis is a specific prediction about the nature and direction of the relationship between two or more variables.
A good research question should lead almost directly to at least one hypothesis. For example:
A good hypothesis has certain characteristics, including:
Other good hypotheses include:
You must be able to justify your hypotheses. That means:
A specific aim is a description of the steps you will take to prove or disprove your hypothesis.
A specific aim must be:
An example of a good specific aim is given below, with respect to a hypothesis.
Prefer a breadth-first presentation: Your research can be thought of as having two dimensions: breadth of coverage, including the different aspects and areas of it, and depth of coverage, including the various levels of detail to which each aspect and area is covered.
It is usually easier for the general reader to understand a breadth-first presentation, where each section or chapter explains matters at a lower/finer level of details than the last.
The breadth should extend well outside the reported work in the Introduction (but obviously not to any significant detail), to show how your research fits in with the rest of the reported work (which you will cover in the Literature Review).
After an oral examination, all the issues, questions, and comments of the examiners have to be merged and integrated into the dissertation. This can be very challenging, especially for very long, complex dissertations. Different examiners will require different changes. Sometimes, the requested changes will directly contradict one another; other times, one change will necessitate various others. In either case, tracking exactly what has to be done to the dissertation can be an onerous project management task entirely unrelated to the research that you did.
One of my PhD students, David Fourney, came up with a chart-based way to manage all those changes that I found greatly simplified the project management aspects and let both David and I focus on making sure that the dissertation was as robust as possible. I describe this method here, in the hope it will help others.
This method works best in Google Spreadsheets or Excel.
Create a table with columns as shown below (and an entirely imaginary sample row is given too).
|STATUS||PAGE||REVIEWER COMMENT||ACTION||SUPERVISOR COMMENT|
|Active||53||Some of the p-values reported indicate non-significant results. This in itself isn't a problem, but you have to admit that the results were not significant and explain what that means for the work overall.||Rewrite discussion for the chapter and overall thesis discussion. Identify which results were significant and which weren't. Add to future work to study the non-significant results.||Hang on. Fisher basically invented 95% confidence level thing; it's completely arbitrary. We need to talk this out. I think it's better to calculate the confidence interval rather than set it, and then use the analysis to rank your results from most to least robust, statistically.|
First, you should go through all the examiners' comments and create an initial table. You will start with only the first three columns filled in. Try to capture the examiners' comments verbatim to preserve as much of the original intent as possible.
The status of each entry can be one of three: Open (not yet started), Active (an item you're currently working on), or Done (having completed thesis updates to address the concern).
Then, go through the table carefully, merging rows that all deal with a given concern. Distinguish between different examiners' comments when you do the merge.
Share the spreadsheet with your supervisor(s) and give them a chance to review it. The table can be used as a source document to guide discussion with your supervisor. But perhaps more importantly, it lets you and your supervisor communicate asynchronously without having to always meet face to face.
Be scrupulous about keeping the table up to date. Any time any progress is made, make sure update the Status and Action cells to reflect the changes. You can also use the Action cells to create a “todo list” of sorts for particularly complex changes.
Carefully track your progress. Note that it's very easy to sort the spreadsheet by the status column, which will group all the open, active, and completed rows together. You can even easily create counters that will total up how many open, active, and completed items you have with the
COUNTIF spreadsheet function - which will give you a sense of your progress.
That's basically all there is to it.
These principles are attributed to Prof. K. Friedman.