Transcript Two-Level Designs

CHAPTER 7
Single-Factor,
IndependentGroups Designs
Copyright 2005, Prentice Hall, Sarafino
Single-Factor, IndependentGroups Designs
• A Single-Factor, Independent-Groups
Designs is one that:
– Has only one independent variable (IV)
– The IV has two or more levels
– Separate groups are expose to only one level
of the IV and compared
• These designs are also referred to as
independent groups designs or betweensubjects designs.
Copyright 2005, Prentice Hall, Sarafino
Two-Level, Single-Factor
Designs
• A two-level, single factor design is one that
has a single independent variable (IV) and
the IV has two levels.
– One of the levels is often a control condition.
– If one level is not a control then the design
has a serious weakness.
– E.g., IV = Drug, and this Drug IV has 2 levels
– a no drug level and a drug level.
Copyright 2005, Prentice Hall, Sarafino
Two-Level, Single-Factor
Designs: Non-Manipulated IV
• Recall that a non-manipulated IV is often a
subject variable (e.g., gender, SES).
– Subject variables give rise to non-equivalent
groups and as such are considered quasiexperimental designs.
Copyright 2005, Prentice Hall, Sarafino
Maximizing IV Variance
• Recall that researchers try to maximize variance
from the levels of the IV and try to minimize
variance from extraneous variables and
nonsystematic error.
• To maximize variance from the IV researchers
typically select levels for the IV that are very
different (e.g., no dose, high dose).
– How do researchers know what levels to choose?
• Research!
• Trial and Error
• Common sense, predictions, theories.
Copyright 2005, Prentice Hall, Sarafino
Minimizing Other Variance
• How can researcher minimize error variance and
variance from extraneous variables?
– The answer is simple – by being careful.
– Things to do to minimize unwanted variance:
• Be sure to create equal groups (use randomization,
matching, or repeated measures)
• Guard against differential attrition (dropout)
• Watch for history, maturation, and regression to the mean.
• Double blind procedure.
• Keep experimental conditions constant.
Copyright 2005, Prentice Hall, Sarafino
Control Groups
• You’re familiar with the concept of a
control group: A group that is not exposed
to the IV and serves as a baseline
comparison.
• Let consider some control group variants:
Copyright 2005, Prentice Hall, Sarafino
Placebo Control Group
• Placebo control groups are typically use in drug
studies.
• A placebo control group receives identical
treatment to that of the experimental condition
except the substance they receive is inert.
– Placebo groups allow for false expectations – which
may themselves be beneficial.
• The placebo effect is any effect the placebo
condition has on the dependent variable.
– Placebos have been shown effective in pain
reduction – they likely cause the release of the
body’s own endogenous opioids.
Copyright 2005, Prentice Hall, Sarafino
Yoked Control Groups
•
•
•
A yoke is a frame that links two animals, at the
neck, together to equalize work.
A Yoked Control Group means that each
member of a pair of participants experiences
simultaneously the factors in an experiment –
the control participant does not receive the IV
manipulation.
The key feature here is the experience of the
same events, expect for the IV, at the same
time.
Copyright 2005, Prentice Hall, Sarafino
Waiting List Control Group
• A waiting list control group is exactly what it
appears to be. Persons in the waiting list control
condition do not initially get exposed to the IV,
but after some delay they will be exposed to the
IV.
• Two key features:
– The effect of a IV can be compared to a standard
group of similar individuals
– Control participants eventually get exposed to IV and
if the IV is a life-saving treatment they will eventually
get the treatment. Reduces the ethical dilemma of
not treating someone – especially if the treatment
works.
Copyright 2005, Prentice Hall, Sarafino
Multilevel, Single-Factor Studies
• The designs have a single IV, but more
that 2 levels of the IV exist.
– E.g. Drug Dose with four levels (0 g, 0.01 g,
0.05 g, 0.1 g).
• Why use a multilevel design?
– See if a nonlinear effect exists
– The hypotheses may require more than 2
levels to be supported.
Copyright 2005, Prentice Hall, Sarafino
Analyzing Data: Single-Factor
Independent-Groups Designs
Two-Level Designs
Parametric Analysis
• Must have interval or data.
• Statistical procedure: independent-groups t-Test
–
t-Test assesses the mean differences between two groups.
Nonparametric Analysis
• Must have nominal or ordinal data
• Statistical Procedure:
–
–
If nominal: Chi-Square Test
If ordinal: Mann-Whitney U
Copyright 2005, Prentice Hall, Sarafino
The p Value
• The p value is the likelihood that the
difference you found might have occurred
if no difference actually does exist.
• Generally speaking, if the p value is less
than 0.05 we reject the null hypothesis
and conclude that the difference we found
was real.
– A p value less that 0.05 means that there is
less that a 5% chance that the results you
obtained occurred by chance.
Copyright 2005, Prentice Hall, Sarafino
Effect Size
• Effect size is a measure of the separation
between two populations.
– Put another way, effect size is the effect that the
experimental procedure had on separating the
experimental group from the control group.
• There are various statistics used to estimate
effect size and all are based on correlation.
Some examples of effect size correlations
include:
– Cohen’s d
– Cramer’s V
– rES:t
Copyright 2005, Prentice Hall, Sarafino
Multilevel Design: Statistics
Parametric Analysis
• The most common statistical procedure for a multilevel
design with interval or ratio dependent measures is an
analysis of variance (ANOVA).
• There are various forms of ANOVA – when only one
independent variable exists with more that 2 levels, a
“one-way ANOVA” is typically used.
• One-way ANOVAs determine if the differences observed
in the sample means are large enough to draw a
conclusion that the population means are different.
– Interestingly, the way we test this is to measure the variance
amongst the means.
Copyright 2005, Prentice Hall, Sarafino
Multilevel Design: Statistics
Nonparametric Statistics
• If your data from a multilevel design is
nominal then a chi-square test is used.
• If your data is ordinal then a multilevel
design requires either a:
– Kruskal-Wallis H test, or a
– K-sample median test.
Copyright 2005, Prentice Hall, Sarafino