The way a research question is asked
can make a great difference. It's only natural that a substantive research question may need a bit of "translation" when
it comes to data analysis. Moreover, many people are not aware of the range of questions that are
answerable using quantitative analysis (which one also might call statistical analysis, predictive
modeling, data mining, or machine learning).
The following list illustrates common types of questions. (Don't see your question?
Email me and we can talk about research approaches that will be most meaningful in your context.)
***
 What sort of change has occurred over time?
 How valid or reliable are the indicators used for a certain purpose?
⤖
Does this accurately represent that?
 How do groups differ?
 Is one group's average, or incidence, or level of risk, higher than the other's?
 Does one group have much more variability than the other?
 What differences might we find among geographical areas?
 How can we best characterize the relationships between or among variables?
 What relationships do we see at face value?
 How strongly is quantity A correlated with quantity B?
⤖
Does more education tend to mean greater prosperity?
 Can we effectively predict Y if we know A, B, C, and D for a given case?
 Does the relationship between A and B change depending on the group or region in question?
⤖
Are there "different slopes for different folks?" (statistical interactions)
 Can we go beyond mere correlation to assess cause and effect?

⤖
Even though greater education tends to mean greater prosperity, which is causing
which?

⤖
What kind of educationprosperity connection do we see if we account for (control
or adjust for) as many other relevant variables as possible?
 Can we establish webs of relationships?
 Is there a basis for taking a large set of variables/factors/characteristics and summarizing
them by, or distilling them into, just a few?
 Is there a basis for clustering (people or schools or nations) to create groups in which each
case has a similar profile?
Then, for most of these types of analyses, it is possible to test whether a phenomenon is statistically significant: whether
it goes beyond what chance would normally produce. This can be helpful in establishing the reliability
or definitiveness of a finding, but researchers are increasingly reluctant to place too much emphasis
on statistical significance, for good reason. See
my commentary
and a
position paper
of the American Statistical Association
(once there, click on PDF  Free Access).
Homepage
~ Copyright 20072018 by Roland B. Stark. ~

Visit my instructional/enrichment site at
yellowbrickstats.com
My reputation points at the
Stack Exchange
Statistics Q & A site
