Vous êtes sur la page 1sur 112

Chapter 11 (Smith and Davis)

Designing, Conducting, Analyzing, and Interpreting


Experiments with More Than Two Groups
Overview of this Chapter
 The Good News and the Bad News
First up, the Bad News. Once more, we look at statistics. Here,
that means a One Way ANOVA, statistic that once again relies
on interval or ratio scales, but with more than two groups

The Good News? This chapter duplicates a lot of information


that we covered in Chapter 10 (SD) on two group designs
Overview of this Chapter
 This chapter does add a layer of complexity, focusing on studies
that have one IV but more than two groups.

Part One: Experimental Design: Adding Basic Building Blocks

Part Two: Statistical Analysis: What Do Your Data Show?

Part Three: Interpretation: Making Sense of Your Statistics

Part Four: An Eye Toward The Future


Part One

Experimental Design:
Adding to the Basic Building Blocks
Experimental Design
 Recall that the experimental design refers to the general plan for
selecting participants, assigning them to experimental conditions,
controlling extraneous variables, and gathering data
Chapter 10 focused on One IV, Two Group designs (correlated
or independent)
Experimental Design
 Experimental Design
One IV, Two Group Designs

Convertibles with their tops open or closed present a good two-


group comparison, but sometimes researchers are interested in
looking at more than just two levels of the IV
– This is where a multi-level (three or more levels) IV study
comes into play …
Experimental Design
 Experimental Design
One IV, Three or More Groups
– Researcher Daniel Linz exposed participants to one of three
kinds of movies with an “adult” element to them
Slasher Films
Teen sex films
Non-violent Porn
Experimental Design
 Experimental Design
One IV, Three or More Groups
– Researcher Daniel Linz exposed participants to one of three
kinds of movies with an “adult” element to them
Linz was interested in looking at participants’ empathy
levels for a (fictional) victim of rape after they had just
viewed one of the three types of films
Thus his DV was empathy while his IV was type of film
(three levels), and participants were randomly assigned
to view one of the three films
Experimental Design
 Experimental Design
One IV, Three or More Groups
– Researcher Daniel Linz exposed participants to one of three
kinds of movies with an “adult” element to them
Linz found that participants who viewed the teen-sex
films and the soft-core pornography films had MORE
empathy than those who viewed the slasher film.
In other words, participants found rape more acceptable
after viewing a slasher film (Kind of gives you second-
thoughts about watching those movies, huh!)
Experimental Design
 Experimental Design
One IV, Three or More Groups
– As in Chapter 10, we have several questions to ask before
beginning our analysis, one related to the number of IVs,
one related to the number of levels of each IV, and one
related to whether we use independent assignment to
groups versus correlated designs assignment to groups

– Before starting that discussion, let’s see how much review


you need of independent vs. correlated designs …
Experimental Design
 Experimental Design
One IV, Three or More Groups
– In this section, we will cover multiple-level studies in more
detail, focusing on

– 1). The Multiple-Group Design


– 2). Comparing Multiple-Group and Two-Group Designs
– 3). Comparing Multiple-Group Designs
– 4). Variations on the Multiple-Group Design
– 5). Post Hoc Tests in Multiple-Group Designs
Experimental Design
 1). The Multiple-Group Design
Similar to Chapter 10, we need to ask several questions about
our research design before analyzing the study data
– A. How many IV’s are there?
– B. How many groups are there?
– C. How do we assign participants to groups?

– To begin answering these questions, think back to the chart


we looked at in Chapter 10 …
Experimental Design
 1). The Multiple-Group Design
A. How many IV’s are there?
– Similar to in Chapter 10, we are only concerned with one
IV in this chapter, but here it has three or more levels. Still,
we have ONLY ONE independent variable in Chapter 11
Yet we are still beholden to the principle of parsimony:
“Keep It Simple Student”
Experimental Design
 1). The Multiple-Group Design
B. How many groups are there?
– In multiple group designs, we compare three or more levels
To keep a study practical, you should keep the number
of levels manageable, like four to five levels maximum
(as a rule of thumb, you need 20 participants per level)
If your single IV has five levels, you’ll need 100 p’s
If your single IV has ten levels, you’ll need 200 p’s!
Theoretically, you can have an unlimited number of IV
levels, but there is only one IV in this our current design
Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups?
– Again, like Chapter 10, we can assign participants to groups
using either an …
Independent group design

OR

Correlated group design


Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups? Independent
– Independent group designs involve random assignment, or
a method of assigning participants to groups so that each
participant has an equal chance of being in any group
Random assignment serves a control function: Potential
extraneous variables are controlled because they should
(hopefully) occur in all conditions
Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups? Independent
– Independent group designs involve random assignment, or
a method of assigning participants to groups so that each
participant has an equal chance of being in any group
Random assignment helps us avoid confounding as well,
limiting the impact of extraneous variables on the IV
Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups? Independent
– In Linz’s study, he randomly assigned participants to watch
one of the three videos:
Soft-Core Pornography
Slasher Movies
Teen-Sex Movies

– Participants only watched one of the films, giving them some


“independence” from the other groups in the study
Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups? Correlated
– We can also use correlated groups in a multi-group design
If we have only a few participants (a small n) or the IV is
weak or very too subtle, we may use a correlated group
design (non-random assignment)
Matched Sets (not pairs, but sets)
Natural Sets (not pairs, but sets)
Repeated Measures
Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups? Correlated
– We can also use correlated groups in a multi-group design
We use “matched sets” rather than “matched pairs”
We match participants on a matching variable (any
potential non-IV variable that may affect scores on
our dependent variable)
Such variables might include:
• Demographic characteristics
• Attitudes, beliefs, experiences, etc.
Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups? Correlated
– We can also use correlated groups in a multi-group design
We use “matched sets” rather than “matched pairs”
Think about Linz. Which variable(s) might we want to
match on before having participants to their film?
• Gender: We match three males and assign one to
the teen comedy, one to the slasher film, and one
to the soft-core porn film. Then repeat
• We could look at movie habits, marital status, etc.
Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups? Correlated
– We can also use correlated groups in a multi-group design
In multi-group designs, we might also use “natural sets”
(NOT natural pairs). However, finding natural sets gets
tougher the more levels you have for your IV
Finding three or more “natural” siblings is tough, but
if your IV has 4, 5, or 6 levels, it may be impossible!
Three or more friends / animals might be easier to
get, but it still gets tough
Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups? Correlated
– We can also use correlated groups in a multi-group design
Finally, multi-group designs can use repeated measures.
This gets tough on participants, who must sit through
three or more sessions
• Maybe they watch all three films in Linz’s study
and give their rape empathy assessments after
each. This is very time consuming, but possible!
Experimental Design
 1). The Multiple-Group Design
C. How do we assign participants to groups? Correlated
– We can also use correlated groups in a multi-group design
Finally, multi-group designs can use repeated measures
As with One IV, Two Level studies, we also have to
worry about carryover effects here
• Are participant scores after the third film based on
that specific film, or based on their responses to
the prior two films?
Experimental Design
 2). Comparing the Multiple-Group and Two-Group Designs
A question we haven’t looked at yet but we need to address is,
“Does your research question require more than two groups?”
There are a few things to consider here …
– First, review the literature to see if the simple two-group test
has already been done
If not, try a two group test first, and do a multi-group later
For example, I might want to see how well Methods
Two students do in a Live class methods version vs.
an Online class version first, and then add in “Hybrid”
and “Independent Study” comparison versions later
Experimental Design
 2). Comparing the Multiple-Group and Two-Group Designs
A question we haven’t looked at yet but we need to address is,
“Does your research question require more than two groups?”
There are a few things to consider here …
– Second, “What will the addition of more groups tell you?”
Maybe you want to see if a new ADHD skills class helps
students learn, so you get a baseline score (time 1). You
might want to then get a time 2 score (right after the first
skills class). You might want a time 3 score to see if the
lessons persist a month after the course, or two months
after the course (time 4), etc.
Experimental Design
 2). Comparing the Multiple-Group and Two-Group Designs
A question we haven’t looked at yet but we need to address is,
“Does your research question require more than two groups?”
There are a few things to consider here …
– Third, determine how many participants are available
If there are only a few participants available, then keep
the design simple. Sometimes it is tough to find enough
participants to run a multiple-group study, so a simple
two-group study design might be best
Experimental Design
 3). Comparing Multiple-Group Designs
There are two additional elements will impact your decision to
use either an independent or correlated design, including:
– A. Control Issues
– B. Practical Issues
Experimental Design
 3). Comparing Multiple-Group Designs
A. Control – Independent Designs
– Independent designs rely on random assignment, creating
(hopefully) three or more groups that are equal
Of course (to beat a dead horse yet again), such random
assignment does not guarantee group equality!
Consider our basic formula again
Experimental Design
 3). Comparing Multiple-Group Designs
A. Control – Independent Designs
– Independent designs rely on random assignment, creating
(hopefully) three or more groups that are equal
Our hope is that the error variability is low among our
groups, making the Statistic higher (especially with high
between group variability). But we have less ability to
control such error variability in independent designs
Experimental Design
 3). Comparing Multiple-Group Designs
A. Control – Correlated Designs
– Correlated designs reduce error variability through better by
controlling sources of error (using participants matched on a
key characteristic or using the same participant repeatedly).
We don’t need to hope that our three or more groups are
similar, as we know they are similar!
Experimental Design
 3). Comparing Multiple-Group Designs
B. Practical issues also arise in multiple-group design studies
– Independent groups designs require a lot of participants

– Correlated designs have their own issues


It may be tough to match three or more participants in
matched sets and natural sets designs
For repeated measures, having more levels means more
participants may need to spend more time in the study.
There are also potential confounds
Experimental Design
 4). Variations on the Multiple-Group Designs
Comparing Differing Amounts of an IV
– I mentioned in Chapter 10 that sometimes rather than using
a “control” group, we use a “comparison” group. The same
comparison groups can occur in multiple-group designs
However, in multiple group designs, we might have:
Multiple control groups!
Multiple treatment groups!
Multiple comparison groups!
Experimental Design
 4). Variations on the Multiple-Group Designs
Comparing Differing Amounts of an IV
– Several years ago, I did a study looking at juror’s ability to
comprehend death penalty instructions. I was living in St.
Louis at the time, so I made sure to use Missouri Approved
Instructions (MAI), or actual Missouri capital instructions!
These instructions were notoriously bad, with jurors not
really understanding them.
Could we write better death penalty instructions using
research from psychology about how jurors process
information?
Experimental Design
 4). Variations on the Multiple-Group Designs
Comparing Differing Amounts of an IV
– To see if we could increase juror comprehension, we gave
five groups of participants different instructions, some based
on actual legal language used in Missouri, some based on
everyday English, some based on visual flow charts, etc.
In this study, we had multiple instruction conditions …
Experimental Design
 4). Variations on the Multiple-Group Designs
Comparing Differing Amounts of an IV
– Three control groups and two treatment groups:
1. Skeletal instructions – We gave jurors only basic legal
information with little to no definitions of legal concepts
2. Legal definitions only – Also very basic, no real help
3. Missouri Instructions – Those actually used in Missouri
4. Flow Chart – Our first experimental version, a step-by-
step visual approach to understanding the death penalty
5. Simplified – Second version, everyday simple English
Death Penalty Instructions
Death Penalty Instructions
Can you figure out what we found? Does it surprise you?
Experimental Design
 4). Variations on the Multiple-Group Designs
Comparing Differing Amounts of an IV
– If we hadn’t used as many comparison / control groups as
we did, it would have been hard to understand how bad the
Missouri death penalty instructions were!
Attorneys could say that offering legal definitions would
help (Nope! We showed they wouldn’t)
Attorneys could say the Missouri instructions are better
than nothing (again, nope! This is not true)
The only thing that aids comprehension of death penalty
instructions are flow charts and simplified language!
Experimental Design
 4). Variations on the Multiple-Group Designs
Dealing with Measured IV’s
– The final element of this section focuses on Measured IV’s
Like Chapter 10, sometimes a researcher cannot directly
manipulate an IV, but she can still classify, categorize, or
measure them (especially predetermined participant IV
characteristics, like gender, age, race, or attitudes)
We cannot draw cause-effect conclusions if using
these “quasi-independent” variables, but they still
might be useful in multiple-group designs
Experimental Design
 4). Variations on the Multiple-Group Designs
Dealing with Measured IV’s
– Going back to my death penalty study, I know death penalty
attitudes might greatly impact how jurors perceive the death
penalty instructions
For example, when it comes to the death penalty, jurors
must be “death-qualified” to serve as a capital juror.
This means jurors must be willing to give the death
penalty (compared to those completely unwilling)
Would “death qualification” impact my results? Recall our
data when death qualification wasn’t measured …
Death Penalty Instructions
Experimental Design
 4). Variations on the Multiple-Group Designs
Dealing with Measured IV’s
– But what happens if we ONLY look at participants who are
“death-qualified” (that is, those who can actually serve on a
death penalty panel)?
Time to see something scary: This is what we found …
Death Penalty Instructions
Experimental Design
 4). Variations on the Multiple-Group Designs
Dealing with Measured IV’s
– Results show that our flow chart and simplified instructions
did not help when only “death-qualified” jurors sit on a trial.
– Apparently, our new instructions only help if both death-
qualified and non death-qualified jurors are present.

Unfortunately, it appears there is little that researchers


can do to educate death-qualified jurors who actually
serve on death penalty trials, which is quite depressing
Experimental Design
 4). Variations on the Multiple-Group Designs
Dealing with Measured IV’s
– Results show that our flow chart and simplified instructions
did not help when only “death-qualified” jurors sit on a trial.
– Apparently, our new instructions only help if both death-
qualified and non death-qualified jurors are present.

Yet remember that we cannot draw causal conclusions


here, as we did not randomly assign jurors to be death-
qualified (or not death-qualified)
Experimental Design
 5). Post Hoc Tests in Multiple-Group Designs
A Brief Introduction to Post Hoc Tests
– Recall the One IV, Two Group design. If our t-Test shows
significance, all we need to do is look at the descriptive
statistics to determine if Mean A is higher than Mean B

– In a One IV, Multiple Group design, we are now dealing with


Means A, B, and C (and possibly D, E, F, etc.), so looking at
descriptive statistics alone will not tell us which means differ.
We need post hoc tests for this
Experimental Design
 5). Post Hoc Tests in Multiple-Group Designs
A Brief Introduction to Post Hoc Tests
– I’ll talk about post hoc tests later in this chapter, but you
might recall post hoc tests from Methods One when we
covered Salkind (Chapter 13)
– One of the most common in psychology is the Tukey post
hoc test (which we will learn all about shortly!)
Part Two

Statistical Analysis: What Do Your Data Show?


Statistical Analysis
 In this section, we cover three main statistical analysis elements,
including …
1). Analyzing Multiple-Group Designs
2). Planning Your Experiment
3). The Rationale of the ANOVA
Statistical Analysis
 1). Analyzing Multiple-Group Designs
Comparing means: This is easy in a two-group design, where
you look at the mean for Condition A compared to the mean for
Condition B
– Independent and correlated groups t-Tests are appropriate

In multiple group designs, we are dealing with more than two


means, so this entails a different statistical test
– For multiple-group designs, we will focus a lot of attention on
a type of test you should be familiar with … the ANOVA
Statistical Analysis
 1). Analyzing Multiple-Group Designs
There are different ANOVAs for different kinds of studies!
– In this Chapter 11, we will cover the One-Way ANOVA, or …
One IV independent design ANOVA
One IV correlated design ANOVA

– (In Chapter 12, we will look at experiments with more than


one IV)
Statistical Analysis
 1). Analyzing Multiple-Group Designs
The ANOVA – Analysis of Variance
– The One-Way ANOVA is a statistical test used to analyze
data from an experimental design with one independent
variable that has three or more groups (levels)
The completely randomized ANOVA is appropriate for
research that uses independent designs
The repeated measures ANOVA is appropriate for
research on correlated designs
Statistical Analysis
 2). Planning Your Experiment
Recall the experiment we discussed in Chapter 10 (the same
experiment discussed in your textbook) looking at the amount
of time it takes for a salesclerk to approach a shopper dressed
in either sloppy or dressy clothes.
– Appropriately, we ran t-Tests for this study in Chapter 10, as
we compared only two means
Statistical Analysis
 2). Planning Your Experiment
Chapter 11 focuses on a three group design, so we add a third
IV level: Sloppy clothes, Dressy clothes, and Casual clothes.
– Keep in mind we could add other levels (naked shoppers,
clown shoppers, etc.)
Statistical Analysis
 3). Rationale of the ANOVA
Remember our formula for variability from Chapter 10

– When looking at Between Group Variability in the two-group


design, we ask one basic question …
Does Condition A differ from Condition B
Statistical Analysis
 3). Rationale of the ANOVA
Remember our formula for variability from Chapter 10

– When thinking about Between Group Variability in multiple-


group designs (three or more), you ask three questions:
Does Condition A differ from Condition B?
Does Condition A differ from Condition C?
Does Condition B differ from Condition C?
Statistical Analysis
 3). Rationale of the ANOVA
Remember our formula for variability from Chapter 10

Consider data from Table 11:2 in your Smith and Davis text …
Clothing & Salesmen
Clothing Style
Dressy Sloppy Casual
37 50 39
38 46 38
44 62 47
47 52 44
49 74 50
49 69 48
54 77 70
69 76 55
M = 48.38 M = 63.25 M = 48.88
Clothing & Salesmen
Clothing Style
Dressy Sloppy Casual
37 50 39
38 Between 46 You want 38
44 Groups 62 differences 47
47 variability 52 between the 44
49 74 dressy and 50
49 69 the sloppy 48
54 77 conditions 70
69 76 55
M = 48.38 M = 63.25 M = 48.88
Clothing & Salesmen
Clothing Style
Dressy Sloppy Casual
37 50 39
38 Between 46 You want 38
44 Groups 62 differences 47
47 variability 52 between the 44
49 74 sloppy and 50
49 69 the casual 48
54 77 conditions 70
69 76 55
M = 48.38 M = 63.25 M = 48.88
Clothing & Salesmen
Clothing Style
Dressy Sloppy Casual
37 50 39
38 Between 46 You want 38
44 Groups 62 differences 47
47 variability 52 between the 44
49 74 dressy and 50
49 69 the casual 48
54 77 conditions 70
69 76 55
M = 48.38 M = 63.25 M = 48.88
Statistical Analysis
 3). Rationale of the ANOVA
While we want high Between Group Variability, Error Variability
should be minimized. This “bad” error variability …
– may be due to characteristics of the participant
– may be due to measurement error
– may be due to recording errors

Having control in the study and trying to reduce the influence of


confounds / extraneous variables will help, but it is impossible
to eliminate all error variability.
Statistical Analysis
 3). Rationale of the ANOVA
Remember our formula for variability from Chapter 10

– As we move into multiple-group designs, we must consider


that error variability is also present between different groups
Clothing & Salesmen
Clothing Style
Dressy Sloppy Casual
37 50 39
38 46 IV variability 38
44 62 and error 47
47 52 variability 44
differ
49 74 50
between
49 69 48
conditions
54 77 70
69 76 55
M = 48.38 M = 63.25 M = 48.88
Statistical Analysis
 3). Rationale of the ANOVA
Remember our formula for variability from Chapter 10

At the same time, we also need to recognize that error occurs


within each group as well. That is, even though all participants
in a condition may be exposed to the same treatment, natural
fluctuations exist even within that condition
– Consider our table again. Notice how scores differ within the
dressy condition (ranging from 37 to 69 seconds to help)
Clothing & Salesmen
Clothing Style
Dressy Sloppy Casual
37 50 39
38 46 38
44 Within 62 Error or 47
47 groups 52 Subject 44
49 variability 74 differences 50
49 69 48
54 77 70
69 76 55
M = 48.38 M = 63.25 M = 48.88
Clothing & Salesmen
Clothing Style
Dressy Sloppy Casual
37 50 39
38 46 38
44 Within 62 Error or 47
47 groups 52 Subject 44
49 variability 74 differences 50
49 69 48
54 77 70
69 76 55
M = 48.38 M = 63.25 M = 48.88
Clothing & Salesmen
Clothing Style
Dressy Sloppy Casual
37 50 39
38 46 38
44 Within 62 Error or 47
47 groups 52 Subject 44
49 variability 74 differences 50
49 69 48
54 77 70
69 76 55
M = 48.38 M = 63.25 M = 48.88
Statistical Analysis
 3). Rationale of the ANOVA
To sum up, we are dealing with several sources of variability in
any data set, some due to error within groups and some due to
error and good IV variability between groups
– Participant characteristics (both within & between error)
– Measurement / recording error (both within & between error)
– Variability of independent variables (between variability)
The ANOVA “partitions” this variability by dividing the treatment
effect by the error. If the treatment effect is large (and the error
low), your F statistic will be large (and significant!)
Statistical Analysis
 3). Rationale of the ANOVA
Here is the basic Statistic formula we’ve used recently

Now, below is the formula for the F test, which is similar to our
usual formula, just a bit more precise

Between-Groups Variability + Error Variability


F=
Error (Within-Groups) Variability
Statistical Analysis
 3). Rationale of the ANOVA
The F-ratio statistic is the statistic computed in an ANOVA (our
obtained value), which we compare to critical values of F
– Similar to the t-Test, you can find the critical value for an F
test (the value the obtained value F must be above to find it
significant) by looking at an F table (Table A-2 in your book)

IF significant, an ANOVA F requires further follow-up analyses


(post hoc tests, like Tukey) to see if A differs from B, if A differs
from C, or if B differs from C.
Part Three

Interpretation: Making Sense of Your Statistics


Interpretation
 Understanding the ANOVA is only half the battle. You also have to
be able to interpret your results in SPSS. In this section, we will
look at how to interpret statistical output, focusing on …
1). The One-Way ANOVA for Independent Samples
2). Translating Statistics into Words (Independent)
3). The One-Way ANOVA for Correlated Samples
4). Translating Statistics into Words (Correlated)
5). Comparing Independent and Correlated Designs
Interpretation
 1). One-Way ANOVA for Independent Samples
Once again, recall the clothing study for a multi-group design,
which asks the question,
– “Do salespeople take longer to respond to customers who
are dressed in sloppy, casual, or dressy clothes?”

Let’s consider the data from Table 11-2)


– Note: I strongly urge you to read section 11.3 in your Smith
and Davis book to better understand the following slides …
Cond Time (Sec)
1 37.00
Table 11-2, page 250 1
1
38.00
44.00
1 47.00
1 49.00
Condition (IV) 1 49.00
1 54.00
1 = Dressy 1 69.00
2 = Sloppy 2
2
50.00
46.00
3 = Casual 2
2
62.00
52.00
2 74.00
2 69.00
Time to help (in seconds – our DV) for 2 77.00
eight participants per condition. Put it 2
3
76.00
39.00
into SPSS as usual. Since we did this 3
3
38.00
47.00
before, I will go through it quickly … 3 44.00
3 50.00
3 48.00
3 70.00
3 55.00
Interpretation
 1). One-Way ANOVA for Independent Samples
Put the data into SPSS as we have done before, making sure
to label everything we need

Variable View

Condition Labels

Data View
Interpretation
 1). One-Way ANOVA for Independent Samples
To run a One-Way ANOVA, go into the “Analyze” menu, scroll
down to “Compare Means” and then click “One-Way ANOVA”
– Move the IV to “Factor” and the DV to “Dependent List”
– Click “Post Hoc” select “Tukey”
– Click “Options” for “descriptives”
Interpretation
 1). One-Way ANOVA for Independent Samples
The first table will be your descriptive statistics (which describe
the data but do not let us know if they are significant).
– The means and standard deviations below should look very
familiar to you by now!
Interpretation
 1). One-Way ANOVA for Independent Samples
Descriptive Statistics (Means, Standard Deviation)
– Dressy Clothing: M = 48.38 sec, SD = 10.11
– Sloppy Clothing: M = 63.25 sec, SD = 12.54
– Casual Clothing: M = 48.88 sec, SD = 10.20

The means seem to differ between conditions, but our real


question is “Do they differ significantly?”
– To answer this question, we must look at the inferential
statistics (One-Way ANOVA)
Interpretation
 1). One-Way ANOVA for Independent Samples
The ANOVA source table refers to a table that contains the
results of an ANOVA. Source refers to the source of different
types of variation (within variation and between variation)

– The “Sum of Squares” is the amount of variability in the DV


attributable to each source (within + between). It’s tough to
calculate by hand, so we will go with the SPSS numbers!
Interpretation
 1). One-Way ANOVA for Independent Samples
The ANOVA source table refers to a table that contains the
results of an ANOVA. Source refers to the source of different
types of variation (within variation and between variation)

– The between-groups df is the # of groups – 1, or 3 – 1 = 2


– Within groups df is # participants – # groups, or 24 – 3 = 21
– Total df is between groups + within groups, or 2 + 21 = 23
Interpretation
 1). One-Way ANOVA for Independent Samples
The ANOVA source table refers to a table that contains the
results of an ANOVA. Source refers to the source of different
types of variation (within variation and between variation)

– But we can ignore the “Total df” in our write up


Interpretation
 1). One-Way ANOVA for Independent Samples
The ANOVA source table refers to a table that contains the
results of an ANOVA. Source refers to the source of different
types of variation (within variation and between variation)

– The “Mean Square” is averaged variability for each source


(divide Sum of squares by the df, or 1141.750 / 2 = 570.875
and 2546.25 / 21 = 121.25). This is essentially the variance
Interpretation
 1). One-Way ANOVA for Independent Samples
The ANOVA source table refers to a table that contains the
results of an ANOVA. Source refers to the source of different
types of variation (within variation and between variation)

– The F ratio is determined by dividing our variation sources


Mean Square Between Groups 570.875
F= = = 4.708
Mean Square Within Groups 121.250
Interpretation
 1). One-Way ANOVA for Independent Samples
The ANOVA source table refers to a table that contains the
results of an ANOVA. Source refers to the source of different
types of variation (within variation and between variation)

– Finally, we have our significance level. In this case, p < .05


Interpretation
 1). One-Way ANOVA for Independent Samples
The ANOVA source table refers to a table that contains the
results of an ANOVA. Source refers to the source of different
types of variation (within variation and between variation)

– So, are we all done? Not yet! We have three means, so we


need to figure out if mean A differs from B, if A differs from
C, or if B differs from C. Time for some post-hoc tests!
Interpretation
 1). One-Way ANOVA for Independent Samples
Post hoc tests refer to statistical comparisons made between
groups means after finding a significant F ratio. As you saw
when selecting a post hoc test, there are several you can use
– Scheffe: This tends to be a very conservative test, and thus
the IV has to be very strong in order to be significant
– Tukey is more liberal than Scheffe, so finding significance is
easier with Tukey than with Scheffe
– LSD is even more liberal than Tukey
Consider these three tests (and try others on your own!) …
Interpretation
 1). One-Way ANOVA for Independent Samples

For the Dressy vs.


Sloppy comparison,
significance is .034 for
Tukey, .044 for Scheffe
(barely significant), and
.013 for LSD. Although
all are significant, let’s
focus on Tukey …
Interpretation
 1). One-Way ANOVA for Independent Samples
According to Tukey, Sloppy differs from Casual
Interpretation
 1). One-Way ANOVA for Independent Samples
According to Tukey, Sloppy differs from Casual, Dressy differs
from Sloppy
Interpretation
 1). One-Way ANOVA for Independent Samples
According to Tukey, Sloppy differs from Casual, Dressy differs
from Sloppy, but Dressy does NOT differ from Casual
Interpretation
 2). Translating Statistics into Words (Independent)
Now we need to write it up. Again, this should be familiar!

“The effect of different clothing on salesclerks’ response time was


significant F(2, 21) = 4.71, p = .02. Tukey tests indicated that
clerks waiting on customers dressed in sloppy clothes (M = 63.25,
SD = 11.73) responded slower than clerks waiting on customers
in both dressy clothes (M = 48.38, SD = 9.46) and casual clothes
(M = 48.88, SD = 9.55). However, the response times of clerks
waiting on customers in dressy and casual clothes did not differ
from each other.”
Interpretation
 3). One-Way ANOVA for Correlated Samples
The One-Way ANOVA for Correlated Samples
– A correlated one-way ANOVA test is like a correlated t-Test
but with three (or more!) matched participants (matched or
natural set) or repeated measures (more than one score)
Like the correlated t-Test, correlated One-Way ANOVAs
are powerful, controlling for participant characteristics
I want to focus on the repeated measures One-Way
ANOVA, as it makes sense for our clothing study
Interpretation
 3). One-Way ANOVA for Correlated Samples
The One-Way ANOVA for Correlated Samples
– Again, recall our research question …
“Do salespeople take longer to respond to customers
dressed in sloppy clothes than those dressed in casual
clothes or dressy clothes?”
– Let’s reconsider this data for a repeated measures format.
Imagine that researchers now approach the SAME
salesclerk three times, but each approach they dress in
a different style of clothing (dressy, sloppy, casual)
Table 11-2, page 246
P T1 T2 T3
Condition (IV) 1 37 50 39
T1 = Dressy 2 38 46 38
T2 = Sloppy 3 44 62 47
T3 = Casual 4 47 52 44
5 49 74 50
6 49 69 48
Time to help (in seconds) for 7 54 77 70
eight participants per condition 8 69 76 55
(DV)
Notice we have just as many
Input these values into an scores (24 scores) but fewer
SPSS file to follow along! participants (only 8)
Interpretation
 3). One-Way ANOVA for Correlated Samples
In SPSS, our file should look like this …

Variable View

Data View
Interpretation
 3). One-Way ANOVA for Correlated Samples
The One-Way ANOVA for Correlated Samples
– To run the ANOVA, go into Analyze, go to “General Linear
Model”, and click on “Repeated Measures”. In the dialogue
box, type in a variable name (for me, “Time” seemed like a
good name), and then type in 3 Levels (for Dressy, Sloppy,
and Casual). Then click “Add” and then “Define”

– SPSS calls a repeated measures design “within subjects”


Let’s see what this looks like in SPSS …
Interpretation
 3). One-Way ANOVA for Correlated Samples
The One-Way ANOVA for Correlated Samples
Interpretation
 3). One-Way ANOVA for Correlated Samples
Now, move your three variables (DressyTime, SloppyTime,
and CasualTime) over to “Within-Subjects Variables” (Time)
– Click options
Select Descriptives
Also, Compare Means
for “Time” (use LSD)
– Note we have Dressy as (1)
Sloppy as (2), Casual as (3)
This will be important!
Interpretation
 3). One-Way ANOVA for Correlated Samples
Our first table includes descriptive statistics

– Our data, of course, are similar to what we found before, but


note that we have only 8 participants per condition
– The means look different, but do they differ significantly?
Interpretation
 3). One-Way ANOVA for Correlated Samples
There are a few other tables in the data display, including a
Multivariate Tests table and Mauchly’s test of Sphericity
– We’ll ignore these in Methods Two. We only have one DV,
so the multivariate test is not relevant. Also, we don’t want a
significant Sphericity test, as significance would violate One-
Way ANOVA assumptions. Since our data does not violate
Sphericity assumptions, we’ll ignore this table as well!
What we are interested in is the “Tests of Within Subjects
Effects” …
Interpretation
 3). One-Way ANOVA for Correlated Samples
According to this ANOVA Table, we have significance in our
data set, F(2, 14) = 19.71, p < .001

We assume Sphericity (as the prior slide noted), so use the


first row. Note the df (for Time and for Error), F, and Sig
Interpretation
 3). One-Way ANOVA for Correlated Samples
Next, look at the “Pairwise Comparison” table (“post hoc” test)

This tells us that 1 differs from 2 (1 is dressy, 2 is sloppy) and 2


differs from 3 (2 is sloppy, 3 is casual). But 1 = 3
Interpretation
 4). Translating Statistics into Words (Correlated)
Now we need to write it up. Again, this should be familiar!

“The effect of three different clothing styles on clerks’ response


times was significant, F(2, 14) = 19.71, p < .001. Post hoc tests
showed that clerks took longer to respond to customers dressed
in sloppy clothes (M = 63.25, SD = 12.54) than to those in dressy
clothes (M = 48.38, SD = 10.11) or those in casual clothes (M =
48.88, SD = 10.20). Response times did not differ between the
clerks waiting on customers in dressy or casual clothing.”
Interpretation
 4). Translating Statistics into Words (Correlated)
Now we need to write it up. Again, this should be familiar!

“The effect of three different clothing styles on clerks’ response


times was significant, F(2, 14) = 19.71, p < .001. Post hoc tests
showed that clerks took longer to respond to customers dressed
in sloppy clothes (M = 63.25, SD = 12.54) than to those in dressy
clothes (M = 48.38, SD = 10.11) or those in casual clothes (M =
48.88, SD = 10.20). Response times did not differ between the
clerks waiting on customers in dressy or casual clothing.”
Please note the book standard deviations are wrong! (Page 259)
Interpretation
 5). Comparing Independent and Correlated Designs
As we finish this section, keep in mind the differences between
independent and correlated tests. For correlated tests …
– 1). there are fewer degrees of freedom (fewer participants)
– 2). the F value tends to be larger
– 3). the probability chance (p value) is closer to zero
– 4). the proportion of variance accounted for is larger
– 5). post hoc comparisons (p values) are closer to zero
Correlated tests control for variance, leading to some of these
differences. Yup, lots of advantages in correlated designs!
Part Four

An Eye Toward The Future


An Eye Toward The Future
 An Eye Toward The Future
In Chapter 11, we focused on one IV with more than two levels.
But what if you need more than one IV to fully understand your
topic?

In Chapter 12, we will look at more complex designs, focusing


on studies that have 2 or more IVs. But we will get to that in a
few weeks

For now …
An Eye Toward The Future
 An Eye Toward The Future
For now I want to backtrack just a bit and look at Chapter 17 in
Salkind (What to do when you’re not normal)
– In our next lecture, we’ll discuss the Chi Square, a type of
statistic we can use when we have dependent variables not
based on the mean score
– This chapter will be very important as you work on the Chi
Square in your results section of your paper.
An Eye Toward The Future
 Finally, it is VERY, VERY, VERY important for you to read your lab
presentation immediately. Since many of your papers are based
on content covered in the lab, you need to know about that content
sooner rather than later

 So, here is your reminder to read that lab presentation immediately

Vous aimerez peut-être aussi