PSYB04H3 Lecture Notes - Lecture 6: Effect Size, Construct Validity, Statistical Significance
PSYB04
LEC 6 - Correlational Research
February 9, 2017
Chapter 8 : Bivariate Correlational Research
→ Introducing Bivariate Correlations
● Association claim: relationship found b/w two measured variables
● Bivariate correlation : association that involves exactly two variables
● To investigate associations researchers need to measure first variable and then measure second
variable → IN SAME GROUP OF PPL
○ Then use graphs and statistics to describe the type of relationship b/w the variables
● Analysis of bivariate correlations looks at only 2 variables at a time
→ Describing Associations B/w Two Quantitative Variables
● Describe the relationship between the 2 variables using scatterplots and correlation
coefficient r
→ Interrogating Association Claims
● w/ association claim the two most important validities to interrogate are construct validity and
statistical validity
→ Construct Validity: How Well Was Each Variable Measured?
● Ask questions about how researchers measured the variables
→ Statistical Validity: How Well Do the Data Support the Conclusion
● What is the Effect Size?
○ All associations are not equal → some stronger than others
○ Effect size: strength of an association
○ Larger effect sizes give more accurate predictions
■ Errors of prediction get larger when associations get weaker
○ Larger effect sizes are usually more important
■ When all else is equal, larger effect size is often considered more
important
■ Depending on the context, a small effect size may be important → ex:
when sample is extremely large
■ however , when outcome is not as extreme as life or death, a very small
effect size is not so important
● Is the Correlation Statistically Significant?
○ Statistical significance : conclusions a research reaches regarding how likely it is
they’d get a correlation of that size just by chance
○ Determining it is a process of inference
find more resources at oneclass.com
find more resources at oneclass.com
■ Researchers study only one sample at a time and make an inference
from the sample about the population
■ Sometimes even if there is 0 association between two variables in a pop,
by chance a study happens to use a sample in which an association
shows up
○ If probability (p) associated with the result is very small (p<0.5) → result is very
unlikely to have come from a “zero-association” population
■ CORRELATION = STATISTICALLY SIGNIFICANT
○ If probability of getting some correlation just by chance is relatively high (p>0.5)
→ result is considered nonsignificant
○ Larger the effect size (stronger a correlation), the more likely the correlation will
be statistically significant
● Could Outliers be Affecting the Association?
○ Outlier: a single case that stands out far away from the pack
■ Extreme score
■ Depending on where it sits in relation to rest of sample, a single outlier
can have strong effect on the correlation coefficient r
■ Potentially problematic for an association claim → may exert
disproportionate influence
■ In bivariate correlation, outliers are especially problematic when
they involve extreme scores on BOTH variables
■ Outliers matter most when a sample is small
● Is There Restriction of Range?
○ If there is not a full range of scores on one of the variables in the association →
can make the correlation appear smaller that it really is
○ Underestimates the true correlation
○ Can apply when one of the variables has very little variance
● Is the Association Curvilinear?
○ Curvilinear association: relationship b/w two variables is not a straight line
■ Ex: relationship might be (+) up to a point, and then become (-)
■ As ppl’s age increases, their use of health care system decreases up to
a point, then as they approach age 60 and beyond, health care use
increases again
→ Internal Validity : Can We Make a Causal Inference from an Association?
● Temptation to make a causal claim is pervasive
● Correlation is NOT causation
● Applying the 3 Causal Criteria
○ Covariance of cause and effect
■ There must be correlation, or association, b/w the cause variable and effect
variable
○ Temporal precedence
■ Causal variable must precede the effect variable; it must come first in time
■ Aka directionality problem
○ Internal validity
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Association claim: relationship found b/w two measured variables. Bivariate correlation : association that involves exactly two variables. To investigate associations researchers need to measure first variable and then measure second variable in same group of ppl. Then use graphs and statistics to describe the type of relationship b/w the variables. Analysis of bivariate correlations looks at only 2 variables at a time. Describe the relationship between the 2 variables using scatterplots and correlation coefficient r. W/ association claim the two most important validities to interrogate are construct validity and statistical validity. Ask questions about how researchers measured the variables. Statistical validity: how well do the data support the conclusion. All associations are not equal some stronger than others. Larger effect sizes give more accurate predictions. Errors of prediction get larger when associations get weaker. Larger effect sizes are usually more important. When all else is equal, larger effect size is often considered more important.
