correlational research—strength of a relationship between two or more variables is quantified

--variables most be quantifiable and usually represent at least ordinal scale of measure

--variables not usually manipulated

--cannot be used to determine causality but still useful in science

--a consistent relationship can be used to predict future events

--provide data that are either consistent or inconsistent with some currently held scientific theory (correlation cannot prove a theory but can negate a theory)

--accurate prediction is a principal goal of correlation research

differential research—observe two or more groups that are differentiated on the basis of some preexisting variable

--group differences existed before any research study was conducted

--group difference become (nonmanipulated) independent variable

--cannot infer causality

--uses similar statistical techniques to correlation

--cohort effect—shared life experiences result in a group behaving similarly

--cross-sectional versus longitudinal (panel designs)

artifacts and confounding variables
--confounding occurs when two or more variables (that move together) can explain the results

--holding a variable constant is best way to avoid confounding

--the variable of greatest interest is the one that is allowed to vary

artifact—is any apparent effect of an independent variable that is actually the effect of some other variable (a result of confounding)

differential research is higher constraint than correlational because researcher exerts some control (by selection of group members) to minimize confounding

--e.g., place subjects in groups that are consistent on confounding variables, but differ on independent variable

--"What is the strength and direction of the relationship between variable X and Y?"

--may also want to develop regression equation for prediction

measurement—multiple measures are best

moderator variable—a variable that seems to modify the relationship between other variables

--best to compute correlations for different subgroups to test for presence of moderator variables

Pearson product-moment—when you have two interval or ratio scale variables

Spearman—when variables are at least ordinal

mutliple correlation—relationship between one variable and a set of variables

canonical correlation—relationship between two sets of variables

partial correlation—correlation of one variable with another after statistically removing the effects of a third variable

coefficient of determination—squared correlation—explains the variability in the first variable (proportion of variance accounted for)

--"Does group A differ from group B?"

--must pick the right two groups and these should have theoretical relevance

--focus on a problem statement where two groups differ on only one variable or dimension

--best to rely on several comparisons when attempting to draw a conclusion about the role of a variable

control group—any group selected in differential research as a basis of comparison with the primary of experimental group

--difficult to find a single, ideal control group matched with the experimental group on all potential confounding variables

--common to use more than one control group in differential studies

--do not draw causal inferences from correlational data