General steps in correlational analysis
|
View the accompanying screencast: [1]
|
|
In correlational analysis we are asking the general question:
"To what extent are two variables related (dependent) or un-related (independent)?"
More specifically, in the case of linear correlation, we are asking the question:
"To what extent do two variables exhibit a linear relationship?"
Recommended steps for conducting correlational analyses are:
- Determine the level of measurement for each variable.
- Examine univariate descriptive statistics and graphical displays for each variable, paying particular attention to the sample size (N >= 50; VanVoorhis & Morgan, 2007), frequencies, central tendency, and distribution of responses.
- Recode if necessary (e.g., if there are too few responses for a response option or data is not normally distributed).
- Examine a bivariate graph (e.g,. clustered bar graph or scatterplot)
- Examine bivariate descriptive statistics if variables are discrete (i.e., cross-tabs or contingency tables)
- Examine relevant correlational statistics, including indicators of size (e.g., Phi, Cramer's ν, Point biserial, Spearman's Rho or Kendall's Tau, or product-moment correlation (r)) and statistical significance of the relationship
- Interpret/conclude - Explain, in clear, simple language, what the correlational analysis indicates about the relationship (direction, strength, and significance) between the constructs of interest.
|
