lOMoARcPSD|6106013
Notes Research Planning and Design 16.10.17
Research Design (University of Glasgow)
StuDocu is not sponsored or endorsed by any college or university
Downloaded by Shiyuan Yang ()
, lOMoARcPSD|6106013
Research Planning and Design. Case Selection and Small-N Research Design. 16.10.17
Number of Cases and Indeterminate Research Designs:
Indeterminate – ‘virtually nothing can be learned from about the causal hypothesis.’
This can happen as a consequence of:
More inferences than cases: you are trying to make more causal inferences than you have
cases to draw data from.
Multi-collinearity: several independent variables are related to a very large extent.
Example: Does geographic distance to the workplace explain job satisfaction?
Control variables: nature of area (rural or urban); travel time; salary.
Random sample of 1,000 people drawn from the Scottish population.
1,000 cases – 4 variables = 996 degrees of freedom. (Subtract variables form the cases). This
is first step of calculating indeterminacy. If it’s really low that can be a problem (you get an
indeterminate research design if your degree of freedom is close to zero or negative).
Multi-collinearity. Travel time and distance to work are almost perfectly correlated. This
could be a problem: if we have a lot of cases which are very similar, we cannot compare
them to each other, and so cannot learn anything from the data.
One needs a sufficient number of cases in which the variables differ.
When multi-collinearity is present the degrees of freedom are reduced.
What can we do? Leave out a variable. You have to decide what is conceptually interesting
to you. Travel time or distance? You can’t put in both because they are highly correlated.
Example Two: (few cases).
Does the size of an ethnic minority lead to genocide?
Control variable (you could have more): economic deprivation.
Case selection:
Khmer Rouge in Cambodia.
Contemporary Switzerland.
2 Cases – 3 Variables = 1 degree of freedom. (note: the slides and lecturer said 1 degree of
freedom, but you might want to check this – if the formula is to subtract variables from the
cases then presumably it is -1).
We cannot distinguish between the causal effects of the two independent variables.
We have fewer cases to test than we have variables that we want to examine: this is why we
get negative degrees of freedom here.
Downloaded by Shiyuan Yang ()
Notes Research Planning and Design 16.10.17
Research Design (University of Glasgow)
StuDocu is not sponsored or endorsed by any college or university
Downloaded by Shiyuan Yang ()
, lOMoARcPSD|6106013
Research Planning and Design. Case Selection and Small-N Research Design. 16.10.17
Number of Cases and Indeterminate Research Designs:
Indeterminate – ‘virtually nothing can be learned from about the causal hypothesis.’
This can happen as a consequence of:
More inferences than cases: you are trying to make more causal inferences than you have
cases to draw data from.
Multi-collinearity: several independent variables are related to a very large extent.
Example: Does geographic distance to the workplace explain job satisfaction?
Control variables: nature of area (rural or urban); travel time; salary.
Random sample of 1,000 people drawn from the Scottish population.
1,000 cases – 4 variables = 996 degrees of freedom. (Subtract variables form the cases). This
is first step of calculating indeterminacy. If it’s really low that can be a problem (you get an
indeterminate research design if your degree of freedom is close to zero or negative).
Multi-collinearity. Travel time and distance to work are almost perfectly correlated. This
could be a problem: if we have a lot of cases which are very similar, we cannot compare
them to each other, and so cannot learn anything from the data.
One needs a sufficient number of cases in which the variables differ.
When multi-collinearity is present the degrees of freedom are reduced.
What can we do? Leave out a variable. You have to decide what is conceptually interesting
to you. Travel time or distance? You can’t put in both because they are highly correlated.
Example Two: (few cases).
Does the size of an ethnic minority lead to genocide?
Control variable (you could have more): economic deprivation.
Case selection:
Khmer Rouge in Cambodia.
Contemporary Switzerland.
2 Cases – 3 Variables = 1 degree of freedom. (note: the slides and lecturer said 1 degree of
freedom, but you might want to check this – if the formula is to subtract variables from the
cases then presumably it is -1).
We cannot distinguish between the causal effects of the two independent variables.
We have fewer cases to test than we have variables that we want to examine: this is why we
get negative degrees of freedom here.
Downloaded by Shiyuan Yang ()
