Task 8: Someone like you. Flashcards
What are the three basic types of experimental design?
- Between-subjects design (different group randomly assigned to the levels of the ind. var.)
- Within-subjects design ( a single group + all levels of treatment)
- Single-subject design (similar to within subjects design, except you don’t average data across subjects)
Error variance is the
variability among scores caused by variables other than your independent variables.
Sources of error variance can be:
subject differences
environmental conditions
even the same subject cannot be exactly the same from moment to moment.
Steps for reducing error variance:
- hold extraneous variables constant by treating subjects similarly
- same procedures for all subjects
- subjects matched on characteristics
- random assignments.
With statistical analysis one can
estimate the probability with which error variance alone would produce differences between group. You can do this using inferential statistics.
Inferential statistics is the process of
making predictions about the population based on a sample.
If the probability (due to statistical analysis) of the effects of error variance is low enough, your results are said to be
statistically significant, from which you can conclude that your results are mostly due to manipulation of the independent variable and not error variance.
The types of between-subjects designs are:
- Randomised two-group design
- Randomised multi-group design
- Matched-group design
- Matched pairs design
- Matched-multigroup design
Randomised two-group design:
- two groups => expose them to different levels of treatment ;
- compare the two means => determine whether they differ ;
- simple statistical analysis ;
- assed the reliability of any difference.
also
- no pretesting necessary
- provides a limited amount of information about the effect of the independent variable.
The randomised multi-group design:
- add as many levels of the independent variable as needed to test you hypothesis ;
- is used to expand the randomised two-group design
- multiple control group design => 2 or more experimental groups, does not have a control group.
Parametric design =
manipulating your independent variable quantitively (mean)
Nonparametric design =
manipulating your independent variable qualitatively (median), using a rather small sample.
Matched-group design:
matched sets of subjects are distributed at random, one per group, into the groups of experiment => group the subjects whose characteristics match and distribute them randomly.
Because of the matched-group design..
the effect of the characteristic on which the subjects were matched gets distributed evenly across the treatments => effect of the error variance has been minimised.
Characteristics of matched group designs:
- Control over subject variables ;
- Simple to carry out (only two levels of your independent variable) ;
- Requires pretesting ;
- If the matched characteristic has a large effect = > more sensitive experiment ;
- Larger subject pool.
The matched-multigroup design involves
multiple levels of a single factor or multiple factors.
In an experiment, a factor is the
independent variable (explanatory). Each factor can have multiple levels.
e.g. Dosage of Vitamin C = factor
0 mg - level 1
40 mg - level 2
100 mg - level 3
Combinations of levels are called
treatments.
Characteristics of within-subjects design
- One group ;
- Each subject is exposed to all levels of treatment ;
- The researcher looks at changes in performance within each participant across treatments ;
- Because the behaviour is measured independently => also called a repeated-measures design ;
- No error variance ;
- Fewer subjects ;
- Demanding on subjects ;
- Subject attrition (drop out)
- Carryover effects.
A carryover effect occurs when
the first treatment alters the behaviour observed in the subsequent treatment = effect that “carries over” from one experimental condition to another.
Carryover effects can arise from:
- Learning - subjects learn the treatment => better performance in a similar task ;
- Fatigue - performance deteriorates ;
- Habituation - stimuli more familiar => reduced responsiveness ;
- Sensitisation - exposure to one stimulus => responds more strongly to another stimulus ;
- Contrast - exposure to one condition => altered response in another ;
- Adaptation - 1st result simply differs from last result.
You can deal with carryover effects in three ways:
1- Counterbalancing
2- Taking steps to minimise carryover
3- Separate carryover effects from treatment effects.
Counterbalancing happens when
you assign the various treatments of the experiments in different order for different subjects.
Characteristics of complete counterbalancing:
1- provides every possible ordering of treatments ;
2- the minimum number of subjects = number of different orderings ;
3- small number of participants.
Partial counterbalancing:
1- includes only some of the possible treatment orders ;
2- order is chosen randomly ;
3- each treatment appear equally in each position ;
4- Latin square design ensures that each treatment appears an equal number of times in each ordinal position.
Counterbalancing may be ineffective when
we discover differential carryover effects (the magnitude of the carryover effects differs for orders of treatment presentation).
Taking steps to minimise carryover:
- minimising carryover effects => reduced error variance + stronger design ;
- not all sources can be minimised ;
- in case of learning => retrain subjects before your experiment ;
- adaptation and habituation => before introducing the treatment, allow subjects to adapt or habituate;
- habituation, adaptation, fatigue => breaks between treatments.
You can separate the carryover effect from the effect of your experimental treatment using..
factorial design.
Using factorial design =>
two or more independent variables with two or more levels of treatment should affect the dependent variable separately.
In a 3x4 factorial design we have:
two factors, where
the first factor has three levels
the second factor has 4 levels
How many groups do we have in a 2x2 factorial design?
4
Characteristics of factorial design:
- you can measure the size of any carryover effect ;
- the result is a complex, demanding experiment ;
- only practical with a small number of treatments.
The single-factor two-level design:
- simplest form of within-subjects design ;
- two levels of a single independent variable ;
- all subjects receive both levels of the variable ;
- half of the subjects = one order ;
- other half of the subjects = opposite order ;
- the scores within each treatment are averaged ;
- the two means are compared.
Single-factor multilevel design:
a single group of subjects is exposed to three or more levels of a single independent variable.