Bivariate Correlational Research Association claims with 2 variables Uses measured variables How do you know the strength of a correlation? Use Cohen’s guidelines… [insert table of Cohen vales] Only use this with quantitative data (i.e. has a range of scores) What if you have ‘categorical data’ (ex: yes/no variable)? Not useful to graph with scatter plot… better to use bar graph Don’t use r with categorical data Construct validity: how do we evaluate this? 5 ways to evaluate statistical validity: 1) effect size 2) statistical significance 3) presence of ‘subgroups’ 4) outliers 5) curvilinear relationship 1) effect size: strength of the evaluation If you have a strong effect size A) Can make more accurate predictions B) Can help you how important the effect is 2) Statistical significance: result is extreme enough that is unlikely to have happened by chance Logic: looking for a difference in the sample that reflects true difference in population… Sometimes find differences in sample by chance when not actually true… Significance evaluates probability that the sample result came from population with no association (i.e. probability it’s a fluke) If there is a very small probability (p) of this happening (less than 5%) then we say its significant Means there is a very low probability that effect is a fluke p<.05 Connection between effect size, sample size, and significance 3) Subgroups: if there are particular groups within the overall sample Can create a ‘spurious’ association: overall effect is attributable only to differences between subgroups 4) outliers: extreme scores EX: (A): r = 0.49; (B) r = 0.15 Outliers can have a large effect on r, especially if there is a small sample size 5) Curvilinear relationship, r will look nonsignificant EX: r = 0.01 EX: negative correlation between small talk and well-being Three criteria: 1) Covariation As A changes then B changes 2) Temporal precedence Correlations have directionality problem 3) Internal validity Correlations have ‘3rd variable’ problem How do we evaluate this? Who are participants? How were they selected? Moderating variables: when relationship between 2 variables changes depending on level of 3rd variable EX: no correlation between maternal employment & child’s vocab… This depends on whether mother is married or single So, relationship depends on marital status (i.e. marital status is a moderating variable) Means correlation does not generalize to all
View Full Document
Unlocking...