Ch8: Bivariate Correlational Research (Final Exam Info)

How do you interrogate association claims

look at statistical validity, internal validity, and moderator variables

statistical validity

how well does the data support the conclusion

• involves: using the appropriate test statistic and using the appropriate sample size

• looks at: effect size, outliers, restriction of range, and curvilinear associations

effect size (what is it? what do researchers use to measure it?)

describes the strength of an association; generally only calculated in experiments

• strong effect size enables more accurate predictions

• Cohen’s criteria:

  • 2% - small effect size

  • 5% - moderate effect size

  • 8% - strong/large effect size

outliers (what are they? how to detect? what are the types? what to do with them?)

• an extreme score

• can detect them through standard deviations (at least 3 standard deviations from the mean) or visually from a scatter plot

• types: online outlier and offline outlier

• has a greater impact with a smaller sample size

online outlier vs offline outlier

• Online outlier - follows the pattern of data and makes the correlation coefficient appear stronger 

• Offline outlier - doesn’t fall in line with the data and makes the correlation coefficient look weaker

restriction of range (what is it? give an example)

only showing scores in a specific range; primary effect is it deflates the correlation/makes correlation appear less strong

• Ex. a college shows scatter plot between SAT scores and college GPA, but the college restricted the range between 1200 to 1600, with this correlation coefficient was .33; when they looked at the whole range (400 to 1600), the correlation coefficient was now .57 (stronger relation)

curvilinear association (what is it? give an example)

can be u-shaped or inverted u-shaped; as x increases, Y does more than 1 thing; correlation coefficient = .01

• Ex. kids/babies use the health care system a lot at first, as they get older they use it less, but after a certain age (20-40), people use the health care system at a more increasing rate

internal validity

3 criteria for establishing causation

• Covariance - do results show variables are correlated? A ←→ B

• Temporal precedence - directionality problem; does the method establish which variable came first in time; did A → B or did B → A; if we can’t tell which came first, we can’t infer causation

• Internal validity (confounds) - 3rd variable problem; is there a C variable that’s associated with both A and B independently?; C → A and B; if there’s a plausible 3rd variable, we can’t infer causation

moderator variables

when the relationship between 2 variables change depending on the level of another variable (a moderator)

Describe an example of a moderator variable

Looking at sports teams success and game attendance: they found that in Phoenix (has high residential mobility - people come and go) there was a positive correlation and in Pittsburg (low residential mobility) there was a -.16 correlation (even when the team was losing, fans still went to games); the moderator was residential mobility