Chapter 9 - variate correlational research

Establishing temporal precedence with longitudinal designs 

measure the same variables in the same people at several different times (long term research)

• Provides evidence for temporal precedence

  • Ex. one variable has changed at time 1 which changes the next variable → variables change based on the order

• Used by developmental psychologists

• Gets us closer to a causal claim 

• Note: longitudinal design is time consuming → effects like cohort effect may influence data 

  • Cross sectional designs can be a better choice BUT → can still have cohort effects 

Longitudinal design yield 3 diff types of correlations

1. Cross-sectional correlations

2. Autocorrelations

3. Cross-lag correlations 

Cross-sectional correlations

• Whether 2 variables measured at the same point in time are correlated 

• Cross-sectional referring to the measuring but it is still part of longitudinal

Autocorrelations

• The correlation of each variable with itself across time 

• Usually looking at things that should consistently correlate across time

• Looking for strong positive correlation bc of consistence

  • Ex. intelligence would be relatively consistent across time, mood would not be

•  

Cross-lag correlations

• Correlation of how much an early measure of a variable is related to a later measure of another variable 

Longitudinal designs can provide some evidence for causation by fulfilling three criteria:

1. Covariance

2. Temporal precedence

3. Internal validity

Why not do an experiment?

• In many cases, participants cannot be randomly assigned a variable

  • Cannot be assigned to preferences

  • Unethical to assign participants

    • Ex. can’t be assigned to have stress

    • Ex. can’t be assigned to have a certain opinion

Multiple regression (multivariate regression)

• Ask if a relationship is still present, even when you statistically control for one or more third variables 

  • Helps address questions of internal validity by ruling out some third variables 

• Ex. pregnancy risk and exposure to sextaul TV content

  • There may be a third variable that can affect data

  • Accounting for age? 

• Regression results indicate if a third variable affects the relationship

Criterion variable

dependent variable

• Variable that we are most interested in understanding

• Measured

Predictor variable

independent variable

• Other variables of interest

• Manipulated 

Beta

given for each predictor variable 

• Tells us ab the relationship between criterion and the specific predictor variables 

  • when all other predictor variables are held constant

• Similar to r 

  • Calculated differently 

• Can be positive or negative

• The number that comes out is the strength of the relationship between that criterion and predictor 

Positive beta

there is a positive relationship between the criterion and predictor variable when all other predictor variables are held constant 

• May occasionally see b (beta)

• Number can possibly get higher than one in certain circumstances 

Statistical significance of beta

• Reported similarly to r 

  • Column labelled p or sig

  • Footnotes with * indicate significance levels 

• Usually report anything smaller than 0.05 to be significant 

What if beta is not significant?

• Will usually state if it is not significant 

• If not significant → not related to criterion 

Regression in popular media

• “Controlled for”

• “Taking into account”

• “Correcting for”

• “Adjusting for”

Regression does not establish causation

• multiple regression is not foolproof

  • Cannot rule out all kinds of variables 

Getting at causality with pattern and parsimony

• What if we can't do an experiment? 

• Pattern and parsimony describes using a variety of correlational studies that all point to a single causal direction

  • Pattern of results that are best explained by a single, parsimonious causal theory

Getting at causality with pattern and parsimony - example

Ex. 

1. The longer a person has smoked cigarettes, the greater are the chances of getting cancer

2. Ppl who stop smoking have lower cancer rates than ppl who continue to smoke

3. Smoker’s cancer tend to be in the lungs and of a particular type 

4. Smokers who use filtered cigarettes have somewhat of a lower rate of cancer than those who use unfiltered ones

5. Ppl who live with smokers have higher rates of cancer from secondhand smoke 

• Pattern of results from one cause

• More confident in making a causal claim 

Pattern, parsimony and the popular press

• Journalists do not always fairly represent patterns and parsimony

• When journalists report only one study at a time, they are selectively presenting only part of the scientific process 

  • Only one study is useless to make pattern and parsimony claim 

  • Lacking context in most popular press due to their own agendas/intended audience 

Mediators vs. third variables 

• Similarities:

  • Both involve multivariate research designs  

  • Both can be detected using multiple regression

• Differences:

  • Third variable are external to the bivariate correlation (problematic)

  • Mediators are internal to the causal variable (unproblematic)

    • Help explain how two things are related to each other

Mediators vs. moderators

• Mediators - ask why does this relationship work

• Moderators - ask for whom does the relationship hold and when does it work 

Multivariate designs and the 4 validities 

• Internal validity (discussed)

  • Can't measure everything - there will always be smth missed

• Construct validity

  • Still the same questions of construct validity

• External validity

  • How have we selected our samples? generalization?

• Statistical validity 

  • Beta - reporting significance levels