Tutorial for beginning graduate students. Hypothesis or research question serves as the compass that gives direction to the project. Posing it poorly guarantees poor results, so here is some guidance.
2. Executive summary
• The Scientific Method starts with a
hypothesis, but in applied science it is often
easier to take the research direction as a
question
– Hypothesis tends to be yes/no or on/off type, but
the question can be “yes, but how much…”
• There are good and poor choices for that
direction, we summarize some guidance for
successful thesis project
3. Types of research
• Exploratory
– Mapping something on which there is little prior literature,
unpredictable results
• Testing-out
– Test prior knowledge outside its original scope (can use prior studies
as templates, just change some parameter or choice of material)
• Problem solving
– High risk, because if it is straightforward, then it is not a problem at
all… It is a problem if nobody knows what needs to be done.
– This can be an ultimate goal, but a thesis project must be doable – you
can’t promise to solve a problem, you can test a candidate solution
• Theory only
– Normal in math, theoretical physics, theory of fluids, thermodynamics,
etc. MS level usually is about explaining existing theories,
demonstrating learning and understanding. PhD expects new
contributions.
4. Suitable as “safe” projects
• Exploration of an inventive idea
– Combining existing tools in a novel way, seeing what happens
• Testing-out
– Use the prior publication as template for plan and analysis,
instead of materials A & B (published) you use C & D.
• Trouble with safe projects
– If they are too safe and predictable, then any student will
perform about similarly. No room for an exceptional student to
show abilities.
– Need to add a component where the student has some
challenge… or provide opportunities to expand from initial plan.
5. Extension and replication
• Testing-out is an example
– Replicate prior work with slight changes, that
extend outside earlier results (this is also called
approximate replication)
• Conceptual replication
– Test if earlier findings can be corroborated by a
different approach (change methods used)
– Particularly fit with novel measurement devices or
techniques – compare new estimates to “old”
measurements
6. Elastic questions are good
• The idea is that the project can be shrunk or
expanded, as necessary – must have several
points where it could stop (because of time limit)
• This again gives safety, but allows a fast student
to go far, and a slower student to finish in time
• It is common that answering one question leads
to another question
– To have that planned ahead of time can be difficult
7. Hypothesis vs. question
• In fundamentals of science, hypotheses could be
used to test the “axioms” needed for theory
– We can leave that to the next Einstein
– Hypothesis tends to have on/off or yes/no character
– Often you already know “yes, this affects”, and the
question is “how much, is this knowledge useful?”
• So, for planned research it is typical to pose a
question (or a few of them)
– How much can we improve extraction yield by
sonicating? Should it be used before or during
extraction?
8. A good question is
• Up-to-date
• Has originality or novelty
• Is important or at least interesting
• Is small and precise
• Is practically testable
9. Recall the approach to planning of
experiments
• List all factors that can affect your results
• To get a short list, remove some by selecting
– This gives the scope of your study
• Choose the main manipulated variables
– This means your hypothesis is that these are the most useful or
most influential ones
– If you have many variables, test with Plackett-Burman to
determine the most important ones. Perhaps continue to
optimize these, with another set of designed experiments.
• What are the measurements?
– Targeted output of course
– Also disturbances that you can’t control, i.e. factors not
eliminated by scope and not manipulated. Include in statistical
analysis…
10. Why only main factors are
manipulated?
• Usually you want to have
– Two levels per factor, to select which factors are
unimportant and can be ignored
– Then three levels per factor, to see curved responses (two
points only define a line)
• Two manipulated factors would give 3*3 = 9
experiments (without replicates)
• Three would give 3*3*3 = 27, four would give 81, etc.
– You have limits on time and cost, so at most 3 manipulated
factors in practice
– With clever statistical designs the number of experiments
comes down from full factorial design
11. The experimental design is coupled
with statistical analysis
• Statistics can show
– Significant differences between alternatives, and
effect size of the choice
– Regression models for continuously manipulated
variables (not only alternative levels, but you can
choose a value from, say, 1.0 to 4.5), or for
measured disturbances
– Models that help find an optimum (maximum or
minimum), so-called “response surface models”
12. In a research proposal
• If you propose specific experiments, you should
also explain the analysis and what you expect out
of it
– You have a goal to answer some question, the
combination of experiments and analysis is how you
reach that goal
• You will need to estimate time and cost of the
experiments also: somebody must have the
money or budget you need
• Next presentation is about the research proposal
13. Reference
• Parts of the discussion were influenced by
Chapter 2 in Lowe: A first textbook of research
methodology… , 2016
Available online at
www.scientificlanguage.com