ch8 Flashcards
Bivariate Correlations (bivariate associations)
- associations that examine exactly two variables at a time
- When there are multiple variables, the authors of the study can present the bivariate correlations between different pairs of variables separately
types of bivariate correlations
- Describing associations between 2 quantitative variables
- Describing associations with categorical data
how do you describe associations between 2 quantitative variables
Describe the relationship between variables using scatterplots and correlation coefficient r
qualities of r:
- Direction
- Strength- how close r is to (-)1
Cohens guide to evaluating strength of association
r… would be considered to have an effect size of
(-).10 = small/weak
(-) .30 = medium/moderate
(-) .50 = large/strong
how to describe associations with categorical data
- More common to plot the results of an association with a categorical variable as a bar graph: examines difference between the group means
- t-test statistic
mean
arithmetic avg
t-test
a statistic to test the difference between two group averages (means)
What makes a study correlational? (What is it supported by?)
- Having two measured variables
- An association claim is supported by a study design- correlational research- in which all variables are measured, not supported by statistics or graphs
Which validities are most important for association claims?
construct and statistical
how to interrogate construct validity in association claims
- How well are each of the variables measured?
- Does the measure have good reliability?
- Is it measuring what it’s supposed to?
what kinds of validity are important in the construct validity of association claims
- Face validity
- Concurrent validity (are these results consistent with previous established measures of this construct?)
- Discriminant and convergent validity
how to interrogate statistical validity in association claims
look at:
- effect size
- statistically significance
-outliers
- restriction of range
- curvilinear association
effect size (+ how to compute)
Effect size: describes the strength of an association
compute by taking value for r and square it
how does effect size help us get accurate predictions?
- Stronger effect sizes mean more accurate predictions
- a larger effect size is usually more important, but in life/death situations even a small effect size can have dramatic effects
statistical significance
Statistical significance: how likely it is to get a correlation of that size by chance if there isn’t one in the population
**depends on effect size and sample size
uses p value
When researchers obtain a correlation coefficient (r ) they establish
- Direction of slope
- Strength of relationship
- Statistical significance
Logic of statistical inference
if there is a correlation in a sample there probably is one in the population
Probability estimate (p value or sig)
provides information about stat. Significance by evaluating the probability that the sample’s association came from a zero-association population.
- Less than .05 is statistically significant, considered rare
- more than .05 = not rare not statisitically significant/nonsignificant/ not rare
how does effect size effect statistical significance?
Usually the stronger a correlation is ( the larger its effect size) the more likely it will be statistically significant
- Because, the stronger an association is, the more rare it would be in a zero-association population
- Need to examine the p value to make sure, because it is dependent on both effect size and sample size
when can a small effect size be statistically significant
A small effect size might be significant if the sample is large, but not in a small sample
how is statistically significant info indicated in journal articles
indicated with an asterisk (*), “sig,” or p<.05 or p<.01
Outlier (+ how to find)
an extreme score
can have effect on correlation coefficient
- look at scatterplot
when are outliers most problematic
- Most problematic when extreme scores on both variables
- Most influential and problematic when sample size is small
Restriction of range (+ when does it happen)
when there’s not a full range of scores on one of the variables in an association, it can make the correlation look smaller than it is
- Restriction of range can apply when for some reason one of the variables has little variance (i.e. all upper middle class parents)
- ask about it when correlation is weak
Correction for restriction of range
statistical technique that estimates the full set of scores based on what we know about a restricted set, and recomputes the calculation
how to correct for restriction of range (2 ways)
- Use statistical techniques (correction for restriction of range
- Recruit more people at ends of the spectrum
Curvilinear association
a relationship between two variables that isn’t a straight line- positive up to a point, then negative, or vice versa
- Correlation coefficient may report that there is zero/no relationship , because its meant for straight lines
- look at scatterplot to find
To statistically analyze curvilinear associations
compute the correlation between one variable and the square root of the other
three causal criteria (in relation to association claims)
- covariance of cause and effect
- temporal precedence (directionality problem)
- internal validity (third variable problem)
Directionality problem:
in association claims, we don’t know which came first (no temporal precedence)
third variable problem
Third variable problem: is there a c variable that is associated with both a and b, independently? (internal validity)
how to check if theres an internal validity third variable problem
If you separate results into subgroups of a third variable, is there still a correlation in each subgroup?
- if so, no internal validity problem
- if not, is a spurious association
spurious association
Spurious association: an association only present due to a third variable
External validity in association claims
does the sample generalize to other people, places, and times
- Size of sample isn’t as important as how the sample was selected from the population
- If a bivariate correlational study doesn’t use a random sample, it is still useful- generalizability can be addressed in another study
- sample may still generalize to different populations
moderator (+ what kind of validity does it inform)
Moderator: a variable that changes the relationship of itself and another variable, depending on its level
- Informs external validity- if an association is moderated by a variable, then we know it doesn’t generalize to other situations
