Transcript Lab week 6

PSY2005: APPLIED RESEARCH METHODS &
ETHICS IN PSYCHOLOGY
1
Lab Class 6: Using a Repeated Measures ANOVA to conduct a
Time Series Analysis
Tutor Led
AIMS & OUTCOMES
Provide an overview of research focusing on drug
treatments over time
 Conduct a repeated measures one-way ANOVA on
time series data
 Explain the key features of a repeated measures
ANOVA
 Complete Workbook 6

2
Tutor Led
TYPES OF TREATMENT
Low threshold: drop-in services, needle exchange, targeted
delivery of health care, outreach services, and drug
consumption rooms
 Detoxification : drugs that block the effects of the to-bewithdrawn drug (naltrexone) may be combined with
anaesthetics
 Pharmacotherapies : substitute drugs (e.g. Methadone)
 Talking therapies: therapeutic communities; structured
prevention programmes (e.g. cognitive behavioural therapy,
motivational interviewing, community reinforcement and
contingency contracting)

3
Stevens, Hallam, and Trace (2006)
Tutor Led
DRUG TREATMENTS IN CURRENT STUDY

Time of Measurement
Time 1: After 1 month
 Time 2: After 6 months
 Time 3: After 1 year


Outcome Measures
Self-monitored logbooks: drugs taken
 Recordings taken over a 28 day period prior to
measurement

4
Tutor Led
EXAMPLE TREATMENT OUTCOME MEASURE
5
Tutor Led
PARTICIPANTS & THERAPISTS

Participants:
Prolific and other Priority Offender status and tested
positive for cocaine or heroin during their arrest.
 Randomly allocated to one of the three groups.


Therapists: Twelve therapists ran the sessions.

All were qualified to degree standard and had a minimum
of three years experience
6
Tutor Led
PROCEDURE
Participants took part as part of a voluntary
rehabilitation procedure
 Participants had either:

90 minute weekly closed (nobody was allowed to join
after the first session) meetings in groups of 4-8 people
with two Counsellors.
 45 minute weekly individual sessions with one Counsellor


Participants attended the sessions for 1 year.
Treatment outcomes were measured at 1 month, 6
months & 12 months.
7
Tutor Led
ABOUT THE EXPERIMENT!
Today’s ingredients
 Hypotheses:

H0: That there is no significant difference in self-reported drug
use across the three periods of drug treatment
 H1: That there is a significant difference in self-reported drug
use across the three periods of drug treatment


Teasing apart the repeated measures design

Independent variable: Length of drug treatment


3 Levels: 1 month, 6 months, 1 year
Dependent variable

Self-reported drug use
8
Tutor Led
SPSS DATA FILES





Open Psy2005 folder
Open Week 6
Drag Lab Week 6 PPT file to desktop
Drag on to desktop and click on ‘drug treatments2.sav’
Fundamental principle
Each participant has their own row
 Each different bit of data must go in a separate column / variable


Data view vs. Variable View

Change via ‘tabs’ at bottom of window

or keyboard combination ⌘T
Data view for viewing / editing data
 Variable view for details of variables

Tutor Led
Type of session:
Group Vs Individual therapy
Situation. More on this later!
Self-reported drug use
at time: 6 months
Self-reported drug
use at time: 1 month
Self-reported drug
use at time: 1 Year
Type of therapy:
12 Step
CBT with MI
10
Standard care.
CREATING A BAR CHART FOR THE
TIME SERIES ANALYSIS
11
CREATING A BAR CHART FOR THE
TIME SERIES ANALYSIS
12
CREATING A BAR CHART FOR THE
TIME SERIES ANALYSIS

13
INCLUDING ERROR BARS
What is a 95% confidence
interval? An inferential
statistic through which a
range of scores is
calculated
with a confidence (95%)
that a population value lies
within it.
14
THAT’S IT!
15
A BAR CHART FOR THE TIME SERIES ANALYSIS
Comment: Shows a
Reduction in drug use
From Time 1 to Time 2
But less of a reduction
From Time 2 to Time 3
16
Tutor Led
HOW CAN WE ANALYZE THIS?


We could carry out a series of t-tests
 1 month Vs 6 months
 1 month Vs 1 year
 6 months Vs 1 year
Is there a problem with this?
 Type 1 error: rejecting the null hypothesis when it is true
 What is the standard probability of doing this? 5%
 Probability for each t-test of not making a type 1 error is
95%
 For three tests (.95 x .95 x .95 = .857)
 Therefore the risk of making a type 1 error across 3 tests is
17
14.3%
REPEATED MEASURES ONE-WAY
ANALYSIS OF VARIANCE

Tutor Led
Testing the Null Hypothesis
Aim of the t-test: find out whether two samples have the same mean:
 Ho: X1 = X2
 H1: X1 ≠ X2
 Aim of an ANOVA: to test whether more than two samples have the
same mean
Where:
 Ho: X1 = X2 = X3
1. 1 month
 H1: X1 ≠ X2 ≠ X3
2. 6 months
 or H1: X1 = X2 ≠ X3
3. 1 year
 or H1: X1 ≠ X2 = X3
 or H1: X1 = X3 ≠ X2


A one-way repeated measures ANOVA tells us whether
the treatments had a different effect on the dependent
variable across the three time periods
18
Type I and Type II Errors
•Let us suppose that a person is judged because he or she has commited a crime.
•If we approach this case as a hypothesis contrast:
•H0: the person is inoccent wilst the contrary is not proved.
•H1: the person is guilty.
•In order to not accept the H0 we have to find evidence against the H0 and supporting the
H1.
•But still, when we are going to make a decision, we might make the following mistakes
reality
Inoccent
Inoccent
Correct
Guilty
Error
verdict
Less severe
Guilty
Error
Correct
19
Very severe
Metodología de la Investigación y Estadística II-UMA
Tutor Led
WHAT DOES POWER MEAN?
Decision
Fail to Accept
Null
State of
nature
20
Retain Null
Null is
true
Type 1 Error:
Correct Decision:
Fail to Accept Null Retain Null when
when true
true
Null is
False
Correct decision:
Fail to accept null
when false
Type 2 Error:
Accept Null when
false
Power is the probability of correctly rejecting a false null
hypothesis. Things that effect Power: Effect size, sample size,
significance criterion and the amount of variability in the data set
WHY DOES A RM ANOVA HAVE MORE
POWER THAN AN IG ANOVA?
Tutor Led
Total Variability
WithinParticipant
Effect of
Experiment
BetweenParticipant
Error: Variation
not explained by
the Experiment
We control for
Individual differences
21
Tutor Led
Sphericity


Homogeneity of variance and covariance: The variances are
equal and the variance of the differences between the conditions
are equal
The Mauchly’s Sphericity Test
Don’t worry if this is hurting your head SPSS conducts a test to
assess whether we are breaking the rules
 If the test is significant we have broken the rules and we need to
apply a correction (E.g. The Greenhouse-Geiser Correction)
 For more information read Chapter 13 of Andy Field’s book

22
CONDUCTING A ONE-WAY RM
ANOVA IN SPSS
23
CONDUCTING A ONE-WAY RM ANOVA
IN SPSS
Insert these

24
CONDUCTING A ONE-WAY RM ANOVA
IN SPSS

25
CONDUCTING A ONE-WAY RM ANOVA
IN SPSS

26
CONDUCTING A ONE-WAY RM ANOVA
IN SPSS

27
CONDUCTING A ONE-WAY RM ANOVA
IN SPSS
28
CONDUCTING A ONE-WAY RM ANOVA
IN SPSS
29
Tutor Led
THE OUTPUT: DESCRIPTIVE STATISTICS

Observations
The first box informs us that we entered the correct variables
 The second provides us with descriptive statistics suggesting
that the biggest decrease in drug use occurred between
drugs use 1 month and drugs use 6 months

30
Tutor Led
THE OUTPUT: MAUCHLY’S SPHERICITY TEST

Observations
This test shows that we have not violated the sphericity assumption
(p>0.05); our data set shows homogeneity of the variances of the
differences.
 In essence these outputs direct us to the correct row in the ANOVA. We will
be looking at the top row (Sphericity Assumed)
 The word Epsilon is used in statistics as a measure of error. If we did not
meet the sphericity assumption we could select one of these measures of
error to guide us in the following ANOVA table.
31

THE OUTPUT: THE MAIN ANOVA

Observations:
We are interested in the F-Ratio highlighted. It is a ratio of average
variability explained (Systematic variance: 249.862) to average
variability unexplained (Unsystematic Variance: 3.528).
 The F-ratio has a probability distribution which can be used to
determine significance levels
 The F-ratio is written thus: (F(2,282)=70.819,MSe=3.528 p<0.001)

32
THE OUTPUT: POST-HOC BONFERRONI TESTS

These test the differences between the treatment outcome
times whilst controlling for family wise error (remember ttests). Remember this is a very conservative test. These tell
us that there are significant differences between time 1 (1
month) and time 2 (6 months) and time 1 (1 month) and
time 3 (1 year). There are no differences between time 2 (6
months) and time 3 (1 year).
33
CONCLUSIONS



The bar chart indicated a difference between the three times that
the outcomes were measured
The Mauchly’s Test showed us that we did not violate an
important rule for carrying out a repeated measures ANOVA
The ANOVA informed us that we can fail to accept:


And accept:


H0: That there is no significant difference in self-reported drug use
across three periods of drug treatment
H1: That there is a significant difference in self-reported drug use across
three periods of drug treatment
The Post-hoc tests tell us that:
There are significant differences between time 1 (1 month) and time 2 (6
months) and time 1 (1 month) and time 3 (1 year). There is no difference
between time 2 (6 months) and time 3 (1 year).
34
 These findings show that in this study the participants reported
significantly more drug use at time 1 than at time 2 or time 3.

COMPLETE WORK & SAVE YOUR FILES
Data Set: ‘drug treatment2.sav’
 Output: ‘week6workbook.spv’
 Cut and paste the graphs in to Workbook (week 6)
 Upload to Unihub

35
36