Professional Documents
Culture Documents
1) Relation between cause (IV) and consequence (DV). Both designs allow for a
relation between IV and DV.
2) Cause precedes the consequence in time. A change in IV causes a change in
DV (after some time) and it can take a long time (e.g., lung cancer caused by
smoking) or effect can be instantaneous (e.g., pushing a light button). It´s occurs
in experimental and, depends on the type of data collection, correlational
research.
The type of data could be:
- Cross-sectional (collecting all data at the same moment), that´s not possible
to determine whether cause precedes the consequence.
- Panel data (collecting data at different moments in time), where it can be
determined but nuisance variables should be cancelled out.
40
QUANTITATIVE RESEARCH METHODS
3) No nuisance variables that may explain the relation between cause and
consequence. We have to exclude effect of/control for nuisance variables it’s
very important in experimental research. The researcher can also control for
nuisance variables not specified in advance by matching. It is to some extent
possible in correlational research, but only for nuisance variables known in
advance (you have to measure them in order to control for them).
5.2. STANDARDIZATION
His goal is to control nuisance variables. They may influence the IV and/or DV.
Manipulation is the only difference between experimental and control condition.
Change in DV only can be attributed to a change in IV (no alternative explanations for
the change in DV).
There are three aspects that should be controlled (kept constant):
1) Circumstances. Place of experiment, time of the day, amount of sleep.
2) Manipulation/intervention. Should be the same for all subjects (e.g.,
instructions of experimenter) and only the intervention may influence the DV,
not the way an intervention is implemented.
3) Measurement instruments. DV should be measured in the same way in both
conditions and differences in measurement instruments may not be the cause
for the change in DV. For example, in Schachter experiment: the same
questionnaire is administered in both groups.
41
5.3. RANDOMIZATION OF THE SUBJECTS
BETWEEN-SUBJECTS DESIGNS
Condition 1 (experimental)
Random assignment to
conditions
Condition 2 (control)
WITHIN-SUBJECTS DESIGNS
Between-subjects designs
42
QUANTITATIVE RESEARCH METHODS
a) Posttest only
30
Difference between groups for the posttest says if
the groups were equal at the beginning (for DV).
math score
20
Randomization tries to make groups equal at the old method
beginning, but this can go wrong. The solution is 10 new method
pretest-posttest design (controlling for
differences in DV at the start). 0
posttest
b) Pretest / posttest
One advantage is related with the difference between groups for the posttest.
Randomization worked when at beginning groups were equal (for DV). Differences in DV
at posttest cannot be caused by differences in DV at pretest.
The other one is related with the change in scores
30 from pretest to posttest, which can be determined
25 (progression in DV caused by the intervention).
20
When intervention is effective, progression in DV
math score
43
• During intervention experimental group can guess the research topic (prejudices)
• This may influence their posttest score: pretest sensitizated the subjects
• No effect of DVD
Solution: Solomon four group design to determine the effect of the pretest: compare 4
groups. Such studies are scarce
Condition 1
Group 1 Pretest Posttest
pretest / (experimental)
posttest
Condition 2
Group 2 Pretest Posttest
(control)
Randomization
Condition 1
Group 3 Posttest
(experimental)
posttest only:
no sensitization Condition 2
Group 4 Posttest
(control)
Condition 1 Condition 2
Posttest Posttest
(experimental) (control)
Randomization
Condition 2 Condition 1
Posttest Posttest
(control) (experimental)
Advantages Disadvantages
No influence of differences between Nuissance variables (that have nothing to
conditions. All (the same) subjects in all do with your manipulation) may influence
conditions: both conditions are equal. the second posttest.
Because group differences cannot explain
Fatigue effects. Participants may become
the effect of the manipulation and cannot
progressively more tired/bored with task.
be guaranteed with between-subject
So number of mistakes may be greater in
designs.
the second than in the first condition
Larger number of subjects in each because they are tired.
condition. Twice as large for the same n
Learning effects. Participants may
(number of subjects) as compared to
become better at the task. So, number of
between-subject design. More powerful
mistakes may be less in the second than
experiment, one can better demonstrate
in the first condition, because they have
the difference between conditions (cf.
learnt to respond more accurately.
power of statistical tests). Within-subject
designs are in general more powerful and
efficient than between-subject designs.
44
QUANTITATIVE RESEARCH METHODS
Experimental: n=10
Between subjects Randon assignment
design to conditions
n=20 Control: n=10
Experimental: n=10
Control: n=10
Randon n=10
Within subjects
assignment to
design
sequences Experimental:
Control: n=10
n=10 n=10
n=20
0% Posttest
8% Posttest
Randomization
16% Posttest
20% Posttest
45
6.1. MULTIPLE IV - FACTORIAL DESIGNS
Punishment Agression
Main effect of alcohol? Main effect of time? Interaction between two variable?
50
• Main effect of Time: more errors in the
40
morning.
30
• Main effect of Alcohol: more mistakes with
stronger alcohol. 20
46
QUANTITATIVE RESEARCH METHODS
50
• Main effect of Time: more errors in the
40
morning.
30
• Main effect of Alcohol: more mistakes with 20
stronger alcohol.
10
• No interaction. Interactions may be most 0
easily grasped in terms of a graph. No Morning Evening
interactions ≈ the two lines are more or less
6% 16%
parallel.
80
• Main effect of Time: more errors in the
morning. 60
40
• Main effect of Alcohol: more mistakes with
stronger alcohol. 20
50
• No main effect of Time.
40
• No main effect of Alcohol. 30
• But Interaction!!!! More mistakes with 20
stronger alcohol in the morning and less 10
with stronger alcohol in the evening. 0
Morning Evening
6% 16%
47
8% 16% MEAN ERRORS
Morning 20 40 30
Evening 40 20 30
30 30
50
• No main effect of Time.
40
• No main effect of Learning tool.
30
• But Interaction!!!! More mistakes with
textbook in the morning and with Ipad in 20
the evening. 10
0
Morning Evening
ipad textbook
80
• Main effect of Time.
60
• Main effect of Learning tool. 40
20
• But Interaction!!!! In morning no difference
between two learning tools and in evening 0
big difference between two learning tools. Morning Evening
ipad textbook
6.2. MULTIPLE DV
48
QUANTITATIVE RESEARCH METHODS
It is important for designing and evaluating experiments and it´s related with internal
validity: are the conclusions valid? In experiment is (change in) IV causal factor for
change in DV. Controlling nuissance variables to exclude alternative explanations.
When not controlling for nuissance variables, which can explain the effect of IV on dv
away (in reality there is no effect of IV).
Controlling nuissance variables by using a control group and standardization.
Nuissance factors that may play a role (especially in pretest-posttest designs), taking
the same test learning effect may be present or getting bored.
Watch out for differential group composition (especially for between-subject designs &
with a small sample size) and for events that are not part of the experiment (but may
occur during the experiment). E.g.: when children with reading problems in a reading
group (= intervention), parents may send child to speech therapist, and we want to know
if their progress due to the intervention. Also, selective attrition of the sample (= loss of
research participants).
The regression to the mean is a sstatistical regression, reversion to the mean and
reversion to mediocrity.
An extreme score on 1e measurement will tend to be closer to the mean on the 2e
measurement and vice versa.
“Explanation”:
20 questions, y/n: very difficult, hence, everyone is guessing. Mean will be 10/20, some
will have good results (because they were lucky).
Those who perform well on this test, will perform worse on second, but similar test. Best
prediction for this student will be 10/20.
Test normally luck and knowledge. Knowledge more or less constant, but “luck” will
cause regression to the mean.
49