PSYCH257 Chapter Notes - Chapter 4: Measuring Instrument, Internal Validity, Scatter Plot
CHAPTER 4: Research Methods
1. Considerations in Research design
Virtual reality – made for people who were very afraid to diagnose; scientists wants to know if this is as effective
as a real life treatment
Balancing internal vs. external validity
a) Internal validity – Confidence that effects are due to the independent variable; any confound that makes us
think that the results is not valid – threat to internal validity
b) External validity – Extent to which the findings are generalizable
Ways to increase internal validity by minimizing confounds
a) Use of control groups
b) Use of random assignment procedures
c) Use of analogue models
Relation between internal and external validity – one comes at the cause of the other.
- If you have high internal validity, you have a great control to your
I. Different types of Research designs
a. THE CASE STUDY
* Not a scientific Method
* get a researcher to report in detail about the case they see
* Often done when the disorder is very rare
* Limitations: 1.) very low external validity because we have no idea if these things do really applies 2.)
Relies heavily on subjective interpretation, it depends on the researcher how to interpret the result: no
apriori hypothesis; 3.) low sample size; problems because no control set up; 4.) low internal validity
b. CORRELATIONAL RESEARCH
Descriptive questions (e.g.: is the severity of trauma associated with thye severity of PTSD
Scatter-plots of negative and positive correlations
Perfect Correlation - When two variables are perfectly correlated, knowing the value of one
variable allows you to exactly predict the value of the other variable
- Few, if any, psychological variables are perfectly correlated with
- Many non-psychological variables do have a perfect correlation
- E.g. Time since the beginning of class and the time remaining in the
class are perfectly correlated
*Factors that can limit a Pearsons’ Correlation coefficient
• Homogenous group -
• Unreliable measurement instrument
• Nonlinear relationship
• Ceiling or Floor with measurement
* Example 1: Imagine that we created a scatterplot of first graders’ weight and height. Notice how the correlation is around
*Now let’s add data from second graders (assuming second graders are generally heavier and taller than first graders but the
relationship between their weight and height is similar to first graders).
*We now have added third graders. Notice how the total scatterplot for first through third graders resembles r=.80 while each
grade resembled r =.60.
*As we add fifth graders, we can see that the correlation coefficient is approaching r=.95 for first through fifth graders.
*The purpose of this demonstration is to illustrate that homogeneous groups produce smaller correlations than heterogeneous
E.g.: when you have truncated or narrowed participants, you’ll get a low correlation
Unreliable Measurement instrument
*Assume that the relationship between Variable 1 and Variable 2 is r = - 0.90.
*If the instrument to measure Variable 1 were unreliable, the values for Variable 1 could randomly be smaller or larger.
*Image that each year couples were married they became slightly less happy.
*Image that after they are married for 7 years, they slowly become more happy each year.
*Pearson could only detect linear relationship
Ceiling or floor with measurement
Why it is not possible to infer causality from a correlation?
Statistical attempt to overcome the directionality problem: cross-lagged panel correlation
Take two sets of correlations separated by a time interval
If two things are associated, it doesn’t mean that they cause each other; it would be hard to know if one
thing causes the other
2. Third-variable Problem:
- What is it?
X <----> Y
- Statistical attempt to overcome the 3rd variable problem: partial correlations
The experimental Method
An experiment is a research procedure in which a variable is manipulated and the manipulation’s effect on another
variable is observed
Manipulated variable = independent variable
Variable being observed = dependent variable
Allows researchers to ask such questions as: Does therapy X reduce symptoms of disorder Y?
Causal relationships can only be determined through experiments
* Researchers must eliminate all confounds – those variables other than the independent variable that may also be affecting
the dependent variable
* Guard against confounds:
- The control group
- Random assignment
To avoid bias by the participant, experimenters employ a “blind design,” in which participants are kept from
knowing what condition of the study (experimental or control) they are in
One strategy for this is providing a placebo – something that looks or tastes like real therapy but
has no key ingredient
Double blind (thrapists and Ps don’t know which group they’re in); Single blind (only the Ps don’t
know which group they belong)
To avoid bias by the experimenter, experimenters employ a “double-blind design,” in which both
experimenters and participants are kept from knowing what condition of the study participants are in
Often used in medication trials
Alternative experimental designs
Clinical researchers often must settle for designs that are less than ideal and include:
o Quasi-experimental designs
o Natural experiments
o Analogue experiments – when you do animal studies
o Single-subject experiments – you need a group to compare to
Single Subject Design Characteristics
Experimental control is demonstrated within the individual
Evaluate behavioral data series for each individual, NOT the mean of all individuals.