What makes a study correlational?
Measured variables (of both in a bivariate study), not any particular statistic, make a study correlational
What validities are associated with an association claim?
- Construct validity - how well was each variable measured?
- Statistical validity - how well does the data support the conclusion?
- External validity (lower priority) -
Why can a correlational study support an association claim but not a causal one?
As it does not satisfy temporal precedence and establish internal validity, although it may show covariance
What are the variables like in a bivariate correlational study?
They can be quantitative or categorical. If both are quantitative, it can be visualised in a scatterplot. If one is categorical, it should be depicted in a bar graph.
What measures can be used to describe a relationship in a scatterplot and bar graph?
For scatterplot, it is r. For a bar graph, it is the difference between the two group means.
What should the construct validity address in a bivariate correlation study?
It should address each of the two variables.
How do you interrogate the statistical validity of an association claim?
- Effect size (strength of r) - the strength between two or more variables; if everything is equal, larger effect sizes are more important but small ones can compound(increase) over many observations
- Precision (confidence interval of r) - as sample size is small, precision is less likely to be strong
- Replication of the study
- The presence of outliers
- Possible restriction of range - can mean that the study underestimates the true relationship and strength of lower or higher treatment values
- Whether the association is curvelinear
How do we look at internal validity in an association claim?
Internal validity addresses the degree to which a study supports a causal claim. Although it is not necessary for an association claim, it may be useful to see some form of causality in a correlational study.
How do you interrogate the external validity of an association claim?
By asking whether the sample is representative of some population. For example, if it was random or not. However a lack of external validity should not disqualify an entire study, if it fufills the other three validities - it is still good.
A bivariate correlation is…
..sometimes moderated. The relationship changes depending on the levels of another variable such as gender, age or location.
What is the directionality and third-variable problem?
Directionality considers temporal precedence as it struggles to differentiate what comes first in a relationship. The third-variable problem concerns internal validity as a whole, whether or not a third variable is causing both problems
