PSY395 Exam 4 Flashcards
Within-Subjects Design Pros
- Control individual differences (ultimate matched group because everyone is in both conditions)
- More likely to find effect (reducing individual differences - decreasing noise).
- Fewer participants needed
- Efficiency gains (less time-consuming/expensive) (esp. if task is unfamiliar/complex, participants are scarce).
Within-Subjects Design Cons / Confounds
Carryover: responses to one condition influence the responses to a later condition (impact of one condition on another).
- Order effects: conditions can change meaning depending upon the condition they follow.
- Practice: participants’ experience in one condition makes it easier to perform in a later condition.
- Interference/Fatigue: participants experience in one condition makes it more difficult to perform in a later condition.
- Differential carryover effect: an effect from a particular treatment(s) does not end before the next treatment begins.
Counterbalancing
Critical for removing W-S confounds.
Subject-by-subject: each subject gets multiple orders of presentation (reverse or block randomization)
Across-subjects: each subject is randomly assigned to one order (latin square, balanced latin square).
Latin Square
Each condition appears once in any order position in the sequences. Protects against order effects.
Balanced Latin Square
Each condition appears once in each order position in the sequence and each condition precedes and follows every other condition an equal number of times.
Protects against order effects and maybe differential carryover effects.
Experimental vs. Relational
Relational: attempts to determine how 2+ variables related
Experimental: varies one factor while all else held constant and some result is measured (evidence of causal relationship between IV and DV)
Assuming Linearity
When someone says r = 0, it only means that there is no linear relationship (there might still be a non-linear one).
Restriction of Range
?? No fucking clue
Cross-Lagged-Panel Correlational Procedure
Does watching violent TV programs cause aggressive behavior?
Four corners, everything correlated where two corners represent one time frame and the other two represent a later one (X, Y)
X at time 1 correlated with Y at time 2
But
Y at time 1 not correlated with X at time 2
Then
X might cause Y but Y probably does not cause X.
Mediator
The mechanism b which the foca independent variable affects the dependent variable of interest.
Full mediation
Partial Mediation
How to test:
Estimate association between IV and DV, estimate association between mediator & IV, estimate association between mediator and DV.
Association must be reduced when the Mediator is accounted for.
Regression
Correlation describes the relationship between two variables. Gives best fit line through data in terms of predicting Y from X.
Allows predictions.
Regression line represents predicted values of Y.
Y = a + bX (a and b are constants, X and Y are variable).
b is slope: how much value of X influences the value of Y. Will tell you if the predictor is significant and how important the predictor is.
Partition of Variance
total variability = residual variability + explained variability
(how much variability around the mean = variability left in Y that X does not explain + how much variability is explained by knowing something about X)
Coefficient of Determination
Total variations = explained variation + residual variation
Prop Expalined = Explained Variability/Total Variability
Multiple Regression
multivariate approach - examines relationship between more than 2 variables. Multiple predictors (at least one is continuous) and one outcome variable.
Y = a + b1X1 + b2X2
Strengthening the evidence for a causal relationship
A correlation is not sufficient to conclude causation.
Run different correlational studies
-Cross-lagged panel (directionality)
-Mediation analysis with multiple regression (3rd variable problem).
One Factor Design
1 IV
Two group design (two levels of IV, use independent samples t-test)
Multiple groups design: more than 2 levels of IV, use one-way ANOVA
Multi-Factorial Design
2 or more IVs
Number of levels of IV #1 by number of levels of IV #2, etc. (like 2X2 design, 2X2X2 design)
use ANOVA
Analysis of Variance (ANOVA)
Answers question: do scores on a dependent variable (DV) differ across levels of an independent variable (IV)?
(tests for null hypothesis)
Omnibus test: tells us if there is an effect, but not where.
IV must be nominal or ordinal (categorical) DV must be interval or ratio (continuous)
Treatment Effect
If H1 is true, scores on the DV will vary systematically because of the effect of the IV (occurs only between treatment groups). Called treatment effect.
Error Variance (always present)
The scores on the DV will vary due to unsystematic factors (random factors, extraneous variables). Most often these are subject variables (individual differences).
Occurs within and between treatment conditions.
Error variance
F-Ratio
By comparing the variance within groups ((error) to the variance between groups (error + treatment), we can see if the treatment had an effect.
F = differences among treatment means / differences among subjects treated alike
F = between-groups (BG) differences / within-group (WG) differences
If H0 true, F =
F = (error variance) + (treatment effect) / (error variance)
F = (error variance) + (0) / (error variance)
F = (error variance) / (error variance) = 1
If H0 true, there should be NO treatment effect (all between-groups variance is error variance).
If H0 false, F =
(error variance) + (treatment effect) / (error variance) = >1
(because treatment effect not equal to 0)
Post hoc
Require alpha adjustment (downward) to compensate for doing many unplanned tests. (after the fact, unplanned)
Only use parities comparisons
Planned comparisons
Limited set of comparisons; derived from theory.
Do not require adjustment (more deliberate).
May use pairwise or complex comparisons.
Factorial Designs
Designs with more than one Independent Variable (IV) or Grouping Variable (GV). More than one Factor (IV/GV).
Better approximates “real life”
More efficient, economical (sometimes informative) than doing many single-factor studies.
Can test for interaction effects between factors.
Factorial Design Nomenclature
Each number represents one IV or GV (if three numbers, there are three factors)
Number itself tells the number of levels for that variable.
Then say BS, WS or Mixed (some of BS and some of WS)
To get total number of conditions in the study, multiply out the numbers.
Main effect of Factorial Design
The overall effect of a particular IV/GV (factor) collapsing across all levels of the other factors.
The number of main effects equals the number of factors.
Interaction Effect
The effect of one IV/GV (factor) depends on the level of the other IV/GV (factor). The differences between levels on one factor are different across levels of the other factor.
Often explain main effects.
# of interactions = # of unique combinations of factors.
