Variance Type of Experiments (Differences)
- Within-subject
- Between-subject
- Between-group
- Within-Group
Design Type of Experiments
- Between-subject
- Within-subject
Within-Subject
- Differences within the same individual (someone is not the same day to day)
X1a ≠ X1b ≠ X1c ≠ X1d
Between-subject
- Differences between people (they are different than others)
X1 ≠ X2 ≠ X3 ≠ X4
Between-group
- Differences across groups (groups that are different)
M1 - M2 ≠ 0
Within-group
- Differences within the same group (individuals making up the group will be different)
X1 ≠ X2 ≠ X3 ≠ X4…..Z1 ≠ Z2 ≠ Z3 ≠ Z4
Between-Subject Group Design
- Different “subjects” in each group
- each measured once
- Hypothesis tested:
Group 1 average ≠ Group 2 Average
Within-Subject Group Design
- Different “subjects” in each group
- each measured twice
- Hypothesis tested:
Average pair difference ≠ 0
Order Effects
- Practice: get better on the dependent Variable test
- Fatigue: get tired and bored with the dependent variable.
- Carryover: effect of an earlier treatment lingers
Example: A drug given earlier in the day may
affect performance on a later trial. - Sensitisation: Participants figure out the hypothesis near the end of the study.
Range Effects
- Best performance near middle of range
- Choose levels that allow generalization to real-world (practical)
Be careful when selecting ranges to be broad enough to catch everything and is practical
Approaches to order problem
- Avoid the problem: use between-subject designs instead
- reduce the practice, fatigue, carryover, and sensitization effects (measure only once and use of distractor sets)
- randomize the sequence of treatment
a.) counterbalance designs (guaranteeing that you
have just as many with treatment A before B or
B before A)
Reduce Specific Order Effects
- Practice Effects: Before starting the experiment, give participants plenty of practice on the DV task.
- Fatigue Effects: Keep experiments short
- Carryover Effects: space treatment far enough apart
- Sensitization Effects: Don’t let participants know what the DV is
Randomize Sequence of Treatments
- Participants get treatments in a random order
a.) Advantage:
i. participants won’t all receive the same
sequences of treatments.
b.) Disadvantage:
i. randomization alone doesn’t guarantee that order effects will be balanced out.
Counterbalance Designs (2 Factorial ANOVA)
- All participants receive each treatment
- Treatment levels are balanced
- ABBA counterbalancing
a.) assumes confounding effects are linear (due to order of effects are reversible.) - treat order Independent variable as between-
subject variable
a.) the between-subject factors manipulate the
order of treatment.
Counter Balance Advantages & Disadvantages
Advantages:
Ensures that routine order effects are balanced across conditions; thus, if there are 2 treatments & 10 participants:
* 5 participants will get T1 then T2
* 5 Participants will get T2 then T1
Disadvantages:
Need to use an ANOVA to analyze results and need more participants
Results from a 2X2 Counterbalanced Design
- Possible Main Effect of treatment
- Possible Main Effect of counterbalancing
- Possible Treatment X Counterbalacing Interaction
How to determine Main Effects or Interaction on Graphs
Main Effects: flat lines and midpoints are separated
Interaction: Lines are not parallel, and mid points are connected.
Latin Square
- Allows for counterbalancing using fewer groups. Looks for sequences of treatments so that:
a.) every treatment precedes and follows every
other treatment the same number of times
b.) and each position in the sequence
Nice for even numbers only; odd numbers produce a mirror of the square, producing 6 unique conditions
Example: 4 treatments A,B,C,D (4 X 3 X 2 X 1)
There are 24 possible combinations; using the Latin square, we can reduce them to 4.
Matched Pair Designs (combined with within-subjects and between-subjects)
- Matched up similar participants that correlate with the DV
- Use random assignment to put one member of each pair in the treatment group(s), the other in the control group(s)
Only effective if matching is appropriate for DV
Match Pair Designs Info
- Matched “subjects” in each group
a. each measured once
b. Hypothesis tested:
Average pair difference ≠ 0
Example: Yolk Control Design (kitten brought up in an environment of only vertical lines and later only associated everything with vertical lines.) - How to match:
a. should correlate strongly with DV
i. use the DV
ii. previous research
b. If the matching variable is unrelated to the DV
i. matching will not reduce the random error
ii. matched designs have half the df (degrees of freedom) of between-subjects design
iii. lower df (degrees of freedom), higher t (or F, or…) needed for significance
Single Variable Experiments
- Advantages:
a.) determine if there is an effect
b.) easy to analyze and interpret
c.) rule out competing theories - Disadvantages
a.) no ability to determine the shape of the
function (only if you have more than 2 dots on a graph)
b.) limits Interpolation (w/in known boundaries) and Extrapolation (outside of known
boundaries)
Multilevel Experiments
- 1 IV with more than 2 groups
Advantages:
a.) better able to tell the shape of the function
b.) common in behavioral and drug research
Disadvantages:
a.) requires more time and effort
b.) harder to analyze and interpret the results
Factorial Design
- Factors can be within or between-subjects
- Mixed factorial designs
Advantages:
a.) Interactives
b.) make random variables into factors
c.) decreases error variances
Disadvantages:
a.) more effect
b.) each additional factor increases the number of
cells
Choosing Designs
- Between-subjects:
a.) if sensitization, practice, fatigue
i. have enough participants
ii. want to generalize, and in real life, people are
exposed to only one level of the treatment - Within-subjects:
a.) if sensitization, practice, fatigue, or carryover
effects are not likely
i. want a powerful design
ii. hard to find participants
iii. want to generate, and in real life, people are
exposed to both treatments - Matched Pair:
a.) easy to obtain scores on matching variable
b.) matching variable correlates highly with DV
c.) hard to find participants
