Experimental Methods 8.
Topics of Lecture
- Understand Humans and their Behavior
- to find Causal (Cause-Effect) Relations
- Variation as a Prerequisite to Measure Differences
- What is Experimental Design
- Validity in an Experiment
Understand Behavior
Uncertainty
makes us Uncomfortable
we want Control and
Predictability
Research Design
3 main
Design Approaches:
Statistical Control
- Descriptive Design (Describe)
- Correlation Design (Describe, Relation and Predict)
Experimental Control
Experimental Design
Science
Goals:
- Describe
- Predict
- find Causes
- Explain and Understand
- Explain Causes
- Explain Effects
- find Cause-Effect Relations
Relations
Causal Relation
(Cause-Effect)
a Variable Directly or Indirectly
Influences Another
Correlational
Relation
Changes in a Variable are Only
ASSOCIATED w Changes in Another
Cause
what Produces
a Modification, a Result or a Consequence
Effect
what has been Produced
the Condition
to talk bt Causal Relation
we have Control
over the Involved Variables
Problems w
Causality
Different Causal Relations
X is Sufficient for Y
- if X then Y
- there can be Y wout X, if Y can be also Caused by Z
X is Necessary for Y
- there is No Y wout X
- there can be X wout Y
X Contributes to Y
X is a Contributing Cause to Y
- X Can be part of a Causal Chain that Can lead to Y
Abolsute Causality
is Problematic
DIfferent Degrees of Probability
Probabilistic Causality
Sum up
Variation as a Prerequisite to
Measure Differences
Classical Test Theory
(aka: True Score Theory)
Spearman, early '900
Goal
Increase Reliability
Each Observation (X) is a Sum of:
True Score (T) +
a Measurement Error (e)
True Score (T)
Measurement Error (e)
Systematic E.
Ex.:
- Social Desirability, Demand Characteristics,
Unclear (/Bad) content, Experimenter Bias,
Lying, Motivation
Affects Validity
we are Not Sure that
we have Measured what we Intended to Measure
Ex. if we want to Measure Alcohol Intake during Pregnancy
and everybody Lie bc they know it's Wrong
have we rellay measured Alcohol Intake
during Pregnancy or Smth else?
to Control
Systematic Errors
Very Difficult
History of Researchers Headache :)
Unsystematic
(Random) E.
Ex.:
- daily shape, Situational conditions (noise in the room...),
error by test administrators, memory
Affects Reliability
you do Not get the
Precise Scores you wanted
Does Not Influence Data
in a Particular Direction
to Control
Unsystematic Error
- just Replicate the Test again and again
- or Replicate it w Other Groups
- See Reliability Testing
Reliability Testing #
- allows to Discover Unsystematic Errors,
- while Systematic E. are Camouflaged
as a part of the True Score
can be Calculated Out:
Value from 0 to 1
1 = all Variation
is due to True Scores
we Measure Only the Same Smth,
which equals to True Score (since we Ignore Systematic Errors)
0 = all Variation
is due to Errors (Unsystematic)
we do Not Measure the Same Smth. at all,
we Measure Only Unsystematic/Random Errors
Systematic and Unsystematic Variation
in Experiments
Casual Unsystematic Variance
Systematic Variation
we have Control on Variation
we Intervene in the Experiment
by Manipulating a Variable
the Variation is due to the Manipulation
the Variation is Not Natual,
is Artificial
it may also be Due Confounding Variables
but we try to Avoid them
all the Variation Not due to
our Manipulation
we hope is due to
Unsystematic Variation
and not due Confounding Variable
What is Experimental Design
Experiment
Definition
Formal
Introducing an Event (Indipendent Var.) in a Controlled Environment,
to Measure the Effect of the Variable on Other Variables
Scientific Method used to
Test Hypotheses of Cause and Effcet
Control Group
Randomized Participants
Each Participant has the
same Chance to be Selected
No Bias in Selection
click to edit
Types
Between-Subject
(At least) 1 Experimental Group
and 1 Control Group
Within Subjects
the Same Group is Measured Before and After the Manipulation
N=1 Design
1 Individual is followed Over Time
Quasi-Experimental
Same as Experimental Design,
but Lacks Randomized Assignment of Participants
can Talk bt Causal Relations
when a Participant that receives the Manipulation
can be Switched w Another Participant
and we get the Same Results
Things we should
have Control over:
- Sample: Representative
- Control and Experimental Groups
- Indipendent Variable
- Dependent Variable
- External Conditions: control over External
Variables that may Possibly have an Effect on Y
Validity in Experiments
Test Validity
- Content Validity
- Criterion Validity (Concurrent, Predictive)
- Construct Validity (Convergent, Divergent)
Inferential Validity
- Internal Validity
- External Validity
Statistical Validity
External
Validity
if the Findings are Generalizable
(to the Real World)
External to the Controlled Environment
if we have the Statistical Basis to draw
our Conclusions
Concerns:
- Samping
- Analyis Method
- Number Precision
- Representativity
- etc.
Internal
Validity
Threats
Between-Subjects Exp.
Selection Bias
Diffusion of
Indipendent Variable
Definition
there are Systematic Diffrences
btw the Experimental and Control Groups
Before the Manipulation
Selective Drop-Out Rate
Within Subjects Exp.
Carry Over
Effect
Previous Measurements
have a Transissible Effect
ex. Irreversibility that Carries Over to
following Measurements
Repeated Testing - Learning
Fatiguing
Tired, do Not Perform as Well
Habituation
Repeated Stimulus receives
a Weakened Response
Sensitization (more Psychological)
- Used to Stimuli
- Expecting Stimuli
Constrasts
Adaptation
Physical Adaptation
History
Contrasts Betwee different Conditions may make us more Attentive to Conditions that come After
so that these Later Conditions actually get More Attention than they should have
ex. Tolerance to medicaments
Changes in the Circumstances
that are not due to Changes in X
ex. Changes in Society,
in your Family, in You,
Death in Familiy etc.
Changes in the Features
of the Measuring Instruments
Statistical Regression
Scores, especially Extreme ones,
have a Tendency to move toward the Average
(in the next Measurement)
Maturing
Natural Changes in the Person
during the Time Interval
Threats
Observation Effects /
Reactivity
- Experimenter Bias
- Hawthorne Effect
- etc.
how to Tackle them
Invert the Order of Conditions
for Each Participant
Change the Order in which you
Introduce Indipendent Variables
ex. you have 4 situations A, B, C, D.
You Change their Order
Pre-Test Design
have a Pre-Test Post-Test Group
Test Before Manipulation,
Control Baseline Behavior
Solomon 4 Groups Design
Pre-Test (Pr), Post-Test (Po):
- Pr ... X ... Po
- ... X ... Po
- Pr ... ... Po
- ... ... ... Po
- All have a Post-Test
- Only one has both Pr, X and Po
- one has only Pr
- the last has only X
Formal
- whether Researchers have Drawn the Correct Conclusions on the Conditions of Cause-Effect in the Experiment
- Inferences have Internal Validity if a Causal Relation btw 2 Variables is Demonstrated
- the extent to which a Causal Conclusion based on a study is Warranted, which is determined by the degree to which a study Minimizes Systematic Error (or 'Bias')
- the Cause Precedes the Effect
- the Cause and the Effect are Related (Covariation)
- there are No Plausible Alternative Explanations for the Observed Covariation (Nonspuriousness)
Personal
- when we Control all Extraneous Variables and the Only Variable Influencing the Results is the One being Manipulated (Indipendent V.)
and Not other Extraneous Variables - the Degree to which the Experiment supports Clear Causal Conclusions
the Effect of the Manipulation is Carried Over
to the Control Group
Ex. In the study of ppl. running more on the tapis roulant when seeing their progress on the diplay
is whether the increase of performance (running) was only due to the display (indipendent variable) and not other extranoeus variables
Ex. in the study of the tapis roulant
whether the findings (the relation between the ind. variable of seeing the progress on a display and the dep. var. of increased running) apply to other contexts, like the impotance of measurement in systematic goal setting