Sie sind auf Seite 1von 31

Digitised by

COMMONWEALTH OF AUSTRALIA

Copyright Act 1968

Notice for paragraph 135ZXA (a) of the Copyright Act 1968

WARNING

This material has been reproduced and communicated to you by or


on behalf of Monash University under Part VB of the Copyright Act
1968 (the Act).

The material in this communication is subject to copyright under the


Act. Any further reproduction or communication of this material by
you may be the subject of copyright protection under the Act.

Any un-associated work has been removed.

Do not remove this notice.


~-~ CENGAGE
1- Learning

SPSS Statistics Version 22: A practical guide 2014 Cengage Learning Australia Pty limited
3rd Edition
Peter Allen Copyright Notice
Kellle Bennett This Work is copyright. No part of this Work may be reproduced, stored In a
Brody Heritage retrieval system, or transmitted in any form or by any means without prior
written permission of the Publisher. Except as permitted under the
Publishing manager: Dorothy Chiu Copyright Act 1968, for example any fair dealing for the purposes of private
Senior Publishing editor: Fiona Hammond study, research, criticism or review, subject to certain limitations. These
Publishing assistant: lydia Crisp limitations include: Restricting the copying to a maximum of one chapter or
Senior project editor: Nathan Katz 10% of this book, whichever is greater; providing an appropriate notice and
Cover design: Danielle Maccarone warning with the copies of the Work disseminated; taking all reasonable
steps to limit access to these copies to people authorised to receive these
Any URls contained in this publication were checked for currency during copies; ensuring you hold the appropriate licences issued by the
the production process. Note, however, that the publisher cannot vouch Copyright Agency limited ("CAL"), supply a remuneration notice to CAL and
for the ongoing currency of URls. pay any required fees. For details of CAL licences and remuneration notices
please contact CAL at Level15, 233 Castlereagh Street, Sydney NSW 2000,
Tel: (02) 9394 7600, Fax: (02) 9394 7601
Acknowledgements Email: info@copyrlght.com.au
IBM, the IBM logo, ibm.com, and SPSS are trademarks or registered Website: www.copyright.com.au
trademarks of lnt:ernatlonal Busl- Michl- Corporltlon, registered in
many jurisdictions worldwide. Other product and service names might For product information and technology assistance,
be trademarks of IBM or other companies. A current list of IBM in Australia call1300 790 853;
trademarks is available on the Web at "IBM Copyright and trademark in New Zealand call 0800 449 725
information" at www.ibm.com/legal/copytrade.shtml.
For permission to use material from this text or product, please email
Permission to quote from, transmit electronically or reprint IBM material aust.permlsslons@cengage.com
is limited to the purpose and quantities originally requested and must not
be construed as a blanket license to use the material for other purposes or National Library of Australia Cataloguing-In-Publication Data
to reprint other IBM copyrighted material. Author: Allen, Peter, author.
Title: SPSS statistics version 22: a practical guide I Peter Allen, Kellie
IBM reserves the right to withdraw permission to reproduce copyrighted Bennett and Brody Heritage.
material whenever, in its discretion, it feels that the privilege of Edition: 3rd edition.
reproducing its material is being used in a way detrimental to its interest ISBN: 9780170348973 (paperback)
or the above instructions are not being followed properly to protect its Notes: Includes index.
copyright. Subjects: SPSS (Computer file) Handbooks, manuals, etc. Social sciences
Statistical methodsComputer programs. Social sciences Statistical
No permission is granted to use trademarks of lnt:ernatlonal Busi- methodsData processingComputer programs. Social sciences--Research
Michi-Corporltlon and its affiliates apart from the incidental Statistical methods--Computer programs.
appearance of such trademarks in the titles, text, and illustrations of the Other Authors/Contributors: Bennett, Kellie, author. Heritage, Brody,
named publications. Any proposed use of trademarks apart from such author.
incidental appearance requires separate approval in writing and ordinarily Dewey Number: 519.5028553
cannot be given. The use of any IBM trademark should not be of a manner
which might cause confusion of origin or appear to endorse non IBM
products. Cengage Learning Australia
level 7, 80 Dorcas Street
THIS PERMISSION IS PROVIDED wnHOUT WARRANTY OF ANY KIND, South Melbourne, Victoria Australia 3205
EITHER EXPRESSED OR IMPLIED, INCLUDING BUT NOT UMITED TO
IMPLIED WARRANTlES OF MERCHANTABILITY AND FITNESS FOR A Cengage Learning New Zealand
PARTlCULAR PURPOSE. Unit 4B Rosedale Office Park
331 Rosedale Road, Albany, North Shore 0632, NZ

For learning solutions, visit cengage.com.au

Printed in China by RR Donnelley Asia Printing Solutions limited.


1 2 3 4 56 718 1716 1514
il~lfllillil~ll~l
004212743

Chapter 5: Independent Samples t Test .AKA:


Independent groups
t test; Between groups
Chapter Overview t test; Between subjects
t test.
5.1. Purpose of the Independent Samples t Test .................................................................. 47
5.2. Questions We Could Answer Using the Independent Samples t Test ................................. 47
5.3. Illustrated Example One ............................................................................................. 48
5.3.1. Setting Up the SPSS Statistics Data File ............................................................... 49
5.3.2. Analysing the Data ............................................................................................ 50
5.3.2.1. Assumptions ............................................................................................. 50
5.3.2.2. SPSS Statistics Procedure (Part 1: Normality) ............................................... SO
5.3.2.3. SPSS Statistics Output (Part 1: Normality) .................................................... 52
5.3.2.4. SPSS Statistics Procedure (Part 2: Homogeneity of Variance and the t Test) ..... 53
5.3.2.5. SPSS Statistics Output (Part 2: Homogeneity of Variance and the t Test) .......... 54
5.3.3. Follow-Up Analyses ............................................................................................ 54
5.3.3.1. Effect Size ................................................................................................ 54
5.3.4. APA Style Results Write-Up ................................................................................. 56
5.3.5. Summary ......................................................................................................... 56
5.4. Illustrated Example Two ............................................................................................. 57
5.4.1. SPSS Statistics Output (Part 1: Normality) ........................................................... 58
5.4.2. SPSS Statistics Output (Part 2: Homogeneity of Variance and the t Test) ................. 59
5.4.3. Follow-Up Analyses ............................................................................................ 59
5.4.3.1. Effect Size ................................................................................................ 59
5.4.4. APA Style Results Write-Up ................................................................................. 60
5.5. Independent Samples t Test Checklist .......................................................................... 60

5.1. Purpose of the Independent Samples t Test


To test for a statistically significant difference between two independent
sample means.

5.2. Questions We Could Answer Using the Independent


Samples tTest
1. Is there a difference between the amount of unpaid overtime worked by
male and female Occupational Therapists (OTs)?

In this example, we're asking whether two independent (or separate) groups
- male OTs and female OTs - work differing amounts of unpaid overtime.
Note how each participant is a member of just one group (i.e., you can be a
male OT or a female OT, but not both). This is an essential characteristic of
the independent samples t test.

Gender is our independent variable (IV), which has two levels: male and
female. Unpaid overtime is our dependent variable (DV), which we could
measure in hours per week.

Similar questions we could answer with an independent samples t test


include:

2. Are nursing home residents with pets happier than those without?

3. Do rats injected with a growth hormone run faster than those injected
with a placebo?

4. Do students who listen to classical music while studying achieve higher


grades than those who listen to rock music?
48 SPSS Statistics: A Practical Guide

5.3. Illustrated Example One


As a class project, Clare and Arash partially replicated Loftus and Palmer's
(1974) classic experiment on the effects of asking leading questions on
memory reconstruction. They showed video footage of a car accident to 30
participants, then asked each of them to answer one of the following
questions: (1) "about how fast were the cars going when they hit each
other?" or (2) "about how fast were the cars going when they smashed into
each other?"

Their research investigates whether the participants who were asked the "hit"
question report, on average, faster or slower speed estimates than those
asked the "smashed" question.

Clare and Arash's data are reproduced in Table 5.1.

Table 5.1

0 Data:
Speed Estimates (in km/h) Given by Participants (N = 30) in Response to
This is data file Either the "Hit" Question, or the "Smashed" Question
data_S_l.sav on the
companion website.

Hit Condition (Group 1) Smashed Condition (Group 2)

Participant ID Est. Speed Participant ID Est. Speed

1 39 16 41
2 33 17 36
3 32 18 49
4 37 19 50
5 35 20 39
6 35 21 38
7 34 22 39
8 33 23 42
9 34 24 41
10 31 25 40
11 38 26 40
12 36 27 45
13 30 28 36
14 34 29 42
15 30 30 47
Chapter 5: Independent Samples t Test 49

5.3.1. Setting Up the SPSS Statistics Data File


~Link:
The UIIMia entered here are what you will
Setting up a data
see during the analyses and In the output. file Is eKplalned In
chapter 1.

1. ID: Participant Identification number. This variable Is ror


administrative purposes, and Is not used In any statistical analyses.

2. Group: The IV. This variable Is used to spectry the experimental


group (I.e., "hit" or smashed") each participant - or case - belongs to.
As It Is a nominal variable, It requires Value Ulbel.

3. Speed: The DV. The participants' speed estimates, In km/h.

Value Labela

Select the Value cell ror the eKperlmental group variable, then click
to open the Value UIIMI dialogue . Use the values 1 and 2 to
Indicate which group each case belongs to.

(I) Tip:
You can toggle
between displaying
values or vlllue /libels
In the Data VIew by
clicking the Value
Ulbel
button:

In the Data VIew, type In your data. Work


In rows, where each row represents a case.

For eJ~ample, participant 10 Is In group 1


(the 'hit" group), and estimated that the
cars were travelling at 31 km/h when they
hit each other.

Participant 20 Is In group 2 (the smashed"


group), and estimated that the cars were
travelling at 39 km/h when they smashed
Into each other.
SO SPSS Statistics: A Practical Guide

~ Syntllx: 5.3.2. Analysing the Data


Run these analyses
with syntax_5_1.sps
on the companion 5.3.2.1. Assumptions
website.
The following criteria should be met before conducting an independent
samples t test. Assumptions 1 and 2 are methodological, and should have
been addressed before and during data collection. Assumptions 3 and 4 can
be tested with SPSS Statistics.
tlllil' AKA: 1. Scale of Measurement. The DV should be interval or ratio data. Some
Both Interval and
ratio data are authors (e.g., Clark-Carter, 2004) indicate that ordinal data are also
referred to as Scale
data In SPSS acceptable, provided the scale has at least seven possible values.
Statistics.

2. Independence. Each participant should participate only once in the


<D Tip: research, and should not influence the participation of others.
Often, we use
Inferential statistics
to estimate 3. Normality. Each group of scores should be approximately normally
population distributed.
parameters from
sample data. Such
generalisations can
only be made 4. Homogeneity of Variance. There should be an approximately equal
confidently if our amount of variability in each set of scores.
sample data has
been randomly
drawn/selected from
the population of The assumption of normality can be assessed in numerous ways (see, for
Interest. example, chapter 4 ). In this example, we will use the Shapiro-Wilk test, and a
If you've used visual inspection of the histograms. The homogeneity of variance assumption
conven lence
sampling (as Clare is tested as part of the t test.
and Arash appear to
have done),
generalisations
should be made
cautiously. 5.3.2.2. SPSS Statistics Procedure (Part 1: Normality)

Tllj[tu
Co!!JIIn .,,_

~-Uri.- Model
Glnlrlllpd lil. . lllocllla
ll!ld lllocllla
.8*-
~-
.eP.......
QOinllll ~f'loll._
!!egrHSion
~
Neunl Nel!!!orb In the Analyze menu, select
~ Descriptive Statistics then
QIIMnlloll RN!don Explore

~
!!!Oft..-..cT....
F~

lnMII

..._,lmpulalon
~Samplu

~~~~-
gu.lllr Corn!
ROC<:wp_
Chapter 5: Independent Samples t Test 51

([) Tip:
Move a variable by
highlighting It,
In l!xplore, move the DV (speed then clicking an
esttm11tes) Into the Depencltlnt Ullt, and arrow button:
the IV (experlmentlll group) Into the
factor Ullt.

Olck the Statllltlca, Plots and/or


Optlona buttons for additional features.

In l!xplore: Statllltlca, Duc:rlptlvea


are selected by default.

a Explore: Statistics

I in !i tf .
[I 1!'11~---~
..
1!'1~

--
1!'1~

Olck Continue to close this dialogue.

sl-------------.,
In l!xplore: Plots, selecting Hllltogrem and
Normality plots with teta will provide 6 }-----------~
In l!xplore: Option the default option
sufficient Information for assessing the
for handling missing values Is l!xclucltl
normality of each distribution of scores.
ca llatwlae. This means that a case Is
not included In the analyses If It Is missing
data on either the IV or the DV.

I ~~~~~~~~--- r~.---
I
1 Oils I,.._.. ~
' 011*
l
l;\ii!lilll!!.--1
.....
.. l.Mt. .lAIIIItli
I! oe-r.......
---
--- !
O~Pip;
Olck Continue to close this dialogue.
I . . I

!1....-------- --- ~------ J


Olck OK to output the analyses.
Click Continue to close this dialogue.
52 SPSS Statistics: A Practical Guide

5.3.2.3. SPSS Statistics Output (Part 1: Normality)


The Cese Proce... ng Summery shows how many cases were
analysed, and how many were dropped due to missing data.
Explore
In this Instance, there were 15 cases In each experimental
group. No cases were excluded because of missing data.
Experimental Group

~rtmenl Group c
VI lid Mallng Totlll
N Pen::ent N Pen::ent N Percent
The table of Descrtptlvee contains a
Speed Estimate Hit
Smllhed
t51100.0%
15 100.0% :I .Olio
.olio
151100.0%
15 100.0%
range of useful Information, Including
measures of central tendency and
dispersion for each group of scores,

--
along with skewness and kurtosis
statistics.
Ellperimentlll Group Slatalc kt.Error
Speed Estimate HH Moon 34.07 .700
811%~1n- LowrBound 32.117
lor-n
Uppor-nd 35.117
5% Trimmed MMn 34,02 When Skewnese and Kurtosis are
Mldlan 34.00 both zero, the data are normally
llllrtonco 7.352
distributed.
Sill. o.lon 2.712
These skewness and kurtosis figures
Mnlmum 30
are reasonably close to zero, and z,
._.mum 38
and Zk (see section 4.3.2.3) are within
Rongo 8 :1:1.96 for both variables. This Is all

Smashed
-
lntlrqUIIrllli Rllnge

Kurloolo
Moon
811% Contldoncoo I n - Lower Bound
41.87
38.27
4
.187
-.!582
v .550
1.121
1.116
reassuring news!

lor-n
Uppor-nd 44.05
SPSS Statistics provides two Teste of
5% Trimmed MMn 41.112
Normellty. The Sheplro-WIIk test is
M1cU1n 41.00 considered more appropriate for
Vlrt.nce 18.1157 smaller samples. A statistically
SII!.D.-.Iollon 4.320 significant (I.e., Slg < .OS), W statistic
Mnlmum 3e Is Indicative of non-normality.
._.mum 50
Rongo 14 Here, W Is .967 (Sig = .804) for the

-
lnlerquartlo Rongo 5 "hit" data, and .920 (S/g z .193) for
.715 .550
the smashed" data. Thus, we can
conclude the assumption of normal!tv
Kul't:llll -288 1.121
Is not ylolated for either group of
~-

~rlmenllll Group ~mogoro.,..SmlmW Shaplro-WIIk If the Assumption Is Violated


AKA:

I
Stltlslc df Sig. Stattatlc df Slg.
The Sig figures The t test Is considered robust against
SpeedEtlmatl HH .200' .1104
reported by SPSS .110 151 .9G7l 15l small to moderate violations of the
Statistics are usually SmMhed .203 15 .088 .920 15 .193
normality assumption, provided the
referred to as p In LlllllitDn Stgnllcence Cori'IICIIon sample Is reasonably large ( 40+ ), and
journal articles. .TNe a lower bound, ... ..,. algnllcence. group sizes are relatively equal.
Researchers concerned about more
For example, W(15) a severe violations - or non-normality
.97, p- .804. combined with heterogeneity of
variance - may consider data
transformation (see Tabachnlck &.
A visual Inspection of the Hllltotlrems Fldell, 2013), or a non-parametric
further confirms that each group of scores js procedure such as the Mann-Whitney
approximately normally distributed. U test (see chapter 17).

Speed Estimate

Histograms
~ Links:
The full SPSS
Statistics output
Included two Normel
QQ Plot., and two
Detnlnded Normel
Q-Q Plot., which
have been omitted
from the current
example. See chapter
3 for Information
about these graphs. I I
llt>Hd-
Chapter 5: Independent Samples t Test 53

5.3.2.4. SPSS Statistics Procedure (Part 2: Homogeneity of Variance and


the t Test)

~-l..ill-lllodll
a.terlllpdunt.llodlle
lllpd IIOdllll
clnlll'l
~
~
NeWII .........
~
In the Analyze menu, select
QlmMIIGn Radon Comp~~re Means then
Sqle Independent-Sample T Te.t.
~T....

l'clrlalfiiD ..
~
lii!IIIH Rupaue
..
g~~a~~ng VIM /onllfllla-
...... lmpulllon ..
CllmiiiiK 8MIIIIet ..
II! ........
QullllrCanhl ..
ROC 011!1--

2 r-------------------~
In lndep~~ndent-SIImple T <D Tip:
Te.t, move the DV Into the Te.t You can
Verteble() list, and the IV Into the simultaneously
Grouping Verteble box. compare two groups
of participants on
several OVs by moving
them all
Into the Te.t
Vartablea() list.

3 r---------------------~
In Define Group~~, enter the
values used to represent each level
of the IV. Here, 1 "hit" and 2
smashed".

8 Define Groups

v::li::u I

I -~" I 4
I OSJ~..-1 I In lndep~~ndent-Sample T Tet:
Option, the defaults are as follows:

~ ~~~-' 8 Independent-Samples TTest-


----------""l
Click Continue to close this
dialogue.
Qllllldlllcl ...... "-*11: ~ I
.......-- i
[::;:::- I!
Click OK to output the t test.
--- i
-~~-- -- ------ ---- - --------- .. ,,_J

Click Continue to close this dialogue.


54 SPSS Statistics: A Practical Guide

5. 3. 2. 5. SPSS Statistics Output (Part 2: Homogeneity of Variance and the


t Test)
The Group Statistics table provides basic
TTest / descriptive information for each group of
scores.

Elcpertmental Group Here, participants asked the "hit" question


(n = 15) Indicated that, on average, the cars
Speed Esllmalll Hit
were travelling at 34.07 km/h at the time of
Smashed
the accident. The participants in the
"smashed" condition (n = 15) reported a
higher average speed of 41.67 km/h.

The standard deviations indicate a larger


spread of speed estimates In the "smashed"
condition.

The Independent Samples Test table actually contains


two tests:

1. The Levene"& Test for Equality of Variances.


2. The t test for Equality of Means. Slg (2-talled) Is a SPSS Statistics default.
Despite some debate (see Howell, 2002),
there appear to be relatively few
circumstances in which a 1-tailed test is
more appropriate than a 2-tailed test.

1
hllpe'*n18ampiH Tnt
~

Le...ne's Teatb'Equalltyot
V.l'lances t-ttst br Equality of Mllnl
95% Conftdenoe 1nr.e1 oflhe

Equal 'Airlsncea
F
2.-'86
Slg.
.126
I df
28 . . (2. . tlod)
.....
DIW.rence
Sld.Enor
Dlfllr1nce ,..., Dlllrenct

Upper

~.~2,~
Speed Estimate -5.771 .000 -7.800 1.317 -10.298
assumed
Equal.,.rlencee not -5.771 23.547 .000 -7.800 1.317 10.321 -4.879
111umed

The Levene's Test for Equality of The t test for Equal variances assumed Is statistically
Variances Is not significant (F = 2.486, significant at a = .05. We can reject the null hypothesis, and
Stg > .05). !be assuml!tl!lll gf conclude that tbere is a differeo'e betweeo tbe tw!l grQUI!S Qf
b!lm!lgeoelt gf llilrlao~ bas O!lt beeo ~.
Yil!lated..
The following details will be needed for the write-up:
Therefore, Interpret and report the t test
for Equal variances assumed. t = -5.771
df = 28
Sig (2-tailed) = .000
If the Assumption Is VIolated

If Levene's Test Indicates that equal


variances cannot be assumed, a modified
version of the t test (often referred to as
Welch's ttest) can be used.
Mean Difference - -- 7.600 -
In SPSS Statistics, this test Is provided
along with the standard t test, and is The difference between the two means Is 7.600 km/h. The
referred to as the t test for Equal minus sign indicates that group 1 (the "hit" condition) had a
variances not aaumed. lower mean than group 2 (the "smashed" condition).

9S% Confidence Interval of the Difference. Lower =


-10.298; Higher= -4.902

If our Intention Is to estimate population parameters based on


sample data, we can be 95% confident that this Interval
(-10.298 to -4.902) Includes the population mean difference.

5.3.3. Follow-Up Analyses

5.3.3.1. Effect Size


The 6th edition (2010) of the Publication Manual of the American
Psychological Association (APA) notes that:

For the reader to appreciate the magnitude or importance of a study's


findings, it is almost always necessary to include some measure of
effect size in the Results section. (p. 34)
Chapter 5: Independent Samples t Test 55

Although SPSS Statistics does not automatically compute an effect size index
for the t test, one can be easily calculated from the output SPSS Statistics
provides.

For example, Cohen's d is a scale-free measure of the separation between two


group means. It provides a measure of the difference between the two group
means expressed in terms of their common standard deviation. Thus, ad of
0.5 indicates that one-half of a standard deviation separates the two means.

d can be calculated with the following formula:


AKA:
This formulation of d,
which takes unequal
sample sizes into
account, is referred to
as Hedges' g.

Where Mt is the mean of group 1, M2 is the mean of group 2, and 5p is the


pooled standard deviation, calculated as:
<D Tip:
When both groups of
scores are the same
(n1 -l~~+{n2 -I~i size, this formula can
sp= be simplified to:
n1+n2 -2

Where n1 is the size of group 1; n2 is the size of group 2; 5 2 1 is the variance of where s 1 and s 2 are
group 1; and 5 22 is the variance of group 2. the standard
deviations for groups
1 and 2 respectively.

All of these figures are available in (or easily calculated from) the Group
Statistics table outputted with the t test:

M, = 34.07 <D Tip:


If you do not want to
M, = 41.67
calculate d by hand,
there are many online
calculators that will do
Sid. Error the work for you.
Moon
.700
1.116

s2 1 = 2. 712 x 2. 712
= 7.352

s2 2 = 4.320 x 4.320
= 18.667

So,

8.667
sp =

=
28

= ~364.266 = .J13.0095
28
=3.607
56 SPSS Statistics: A Practical Guide

Then,

d = 34.07-41.67
3.607
=-2.11

Cohen (1988) suggested some general conventions for effect sizes in the
CD Tip: social sciences. According to these conventions, an effect size of d = .20 is
Report the absolute
value of d, rather than
a negative value.
considered small, d = .50 is medium, and d = .80 is large.

Cohen (1988) cautions that these general recommendations are more useful
in some circumstances than others. However, the concept of small, medium,
and large effect sizes can be a reasonable starting point if you do not have
more precise information to work from.

5.3.4. APA Style Results Write-Up

Results

An independent samples t test was used to compare the

average speed estimates reported by participants in the "hit" condition


Assumption testing is often
(n = 15) to the average speed estimates reported by those in the not reported in journal articles
(especially when they are not
violated). It Is seen much
"smashed" condition (n = 15). Neither Shapiro-Wilk statistic was more frequently in student
papers.

significant, indicating that the assumption of normality was not

violated. Levene's test was also non-significant, thus equal variances When the t test Is statistically
significant, both the size and
direction of the effect should
can be assumed. The t test was statistically significant, with the "hit" be reported.

In this instance, direction is


Indicated by the mean
group (M= 34.07, SD = 2.71) reporting speed estimates some 7.60 difference (and 95% CI), or
can be Inferred by comparing

-
the two group means.
km/h lower, 95% CI [-10.30, -4.90], than the "smashed" group (M =
Size is indicated by Cohen's d.

41.67, SD = 4.32), t(28) = -5.77,p < .001, two-tailed, d= 2.11.

5.3.5. Summary
In this example, the independent samples t test was statistically significant,
and neither the normality nor homogeneity of variance assumptions were
violated. Real-life research, however, is often not this "neat"!

In the second illustrated example, our t test is non-significant, and the


homogeneity of variance assumption is violated - two issues that researchers
must commonly deal with.
Chapter 5: Independent Samples t Test 57

5.4. Illustrated Example Two


As part of a national assessment program, the students in Mrs Sommers' year
7 class recently completed an intelligence test. Their global IQ scores are
tabulated in Table 5.2.

Mrs Sommers would like to know whether there is a difference between the
average IQ of her male students, and that of her female students.

Table 5.2
CJ Dm:
Gender and IQ Data for Each Member of Mrs Sommers' Class (N = 35) This Is data file
dm_5_2.Hv on the
companiOn website.

Participant ID Gender IQ

1 1 100
2 1 82
3 1 109
4 2 105
5 2 110
6 1 118
7 1 97
8 1 108
9 2 112
10 1 124
11 1 104
12 2 95
13 2 106
14 2 94
15 2 109
16 1 110
17 2 105
18 2 106
19 1 105
20 1 103
21 1 112
22 1 97
23 2 99
24 1 96
25 2 108
26 2 109
27 2 104
28 1 117
29 1 114
30 2 106
31 2 99
32 2 98
33 1 89
34 1 98
35 2 100

Note. Gender Is coded as 1 = male and 2 = female.


58 SPSS Statistics: A Practical Guide

5.4.1. SPSS Statistics Output (Part 1: Normality)


.Syntllx:
Run these analyses
with .yntllx_5_2.e.- Explore The procedure for gene'atlng this output Is the
on the companion Slime as that used In Illustrated Example One.
website.
gender

gender c
Valid Msslng Totol
N Percent N Percent N Percent
IQ Millo 181100.0% .0% 18 1100.0% The Caee Proceulng Summary Indicates that
Fomolo 17 100.0% :I .0% 17 100.0% there are 18 males In the sample, and 17 females.
'- There are no missing data.

Dlsorlpllwe - In the table of Deec:rlptlvee we can see that the


gender S1alolc Sid. Error =
Mean male IQ Is 104.61 (SO 10.639), while the
IQ Millo Moon 104.81 2.!08 Mean female IQ Is 103.82 (SO= 5.376).
115% Conldonc:o .........I Lower Bound 1111.32
lor Moon The Variance, Std. Deviation, Minimum,
U-llound 1011.90
5% Trtmmect ,.." 104.78 Maximum and Range for each group of scores
Mldian 104.80
Indicate that there Is more variability In the male
data than the female data.
Varlonoo 113.183
Std. OW.Ion 10.838
The Skewne and Kurtoels statistics are all
Minimum 82 close to zero, and z, and z. are within ::1:1.96 (see
Mulmum 124 section 4.3.2.3) for both genders, Indicating that
Rongo 42 each group of scores Is reasonably normally
lntl<quorlleRongo 18 distributed.

Female
-
Kui'IDIII
Moon
.228
.Q!Q

103.82
.838
1.038
1.304
11!1% Conldoncoln1oMI LOW8f' Bound 101.08
lor-n
U-llound 108.88
5% Trtmmect MHn 103.82
a..ctlan 108.00
V.rtance 28.904
Std.Oa'tlladon 8.378
Minimum 84
,._mum 112
Rona 18

--
~tlrquartNa Ft.nge 10
A31 .850
Kui'IDIII .871 1.083

Both Shaplro-WIIk tests are statistically non-

gender Kolmogorov-Smlmoyl Shaplro.-WIIk


I significant at a .OS, confirming that the
normality assymptlon Is not ylolated.
Stlllllc elf Slg. SWill tiC elf Slg.
IQ Millo 200' .9881 .996
.0118 I 18 _I 18 I
Fomolo .178 17 .178 .840 17 .320

a. LIHiefora Slgnllcence Conwdlon The HletDgrame also confirm that each


distribution of IQ scores looks reasonably normal.
.Th.. ll 1 towerboundofttetruelgntllclra.

IQ

Histograms
Hlotlogram

,..,104.11
at.Dtw.IO...
N11
/ H1otDgnm

I I
IQ
Chapter 5: Independent Samples t Test 59

Boxplot8 were obtained


by selecting l'ctor level
Explore: F
together In the Explore:
Plot. dialogue.

Each boxplot Is roughly


----
8[1C1Drlllllll .......
0QIIIIIIIIII......_
symmeb1clll, Indicating tlult
each group of scores Is OI!IM
II normally distributed.

The male boxplot covers 11 much wider range of IQ


scores than the female box plot, suggesting that
heterogeneity of varlllnce may be of concern. The
Laven' Tellt (below) will c:lllrlfy this.

- -
5.4.2. SPSS Statistics Output (Part 2: Homogeneity of Variance and
the tTest)

TTest

-
As suspected Levene' Tellt for !quaHty of
a.-.- Vrtance. Is' statistically significant (F 5.689,
gondot' Slid. Error
N - ]s...o........ Slg .023), Indicating that the b!lm ggeneltv llf
IQ ...... 11 104.01 I 10.138 2.1108 variance assumptiOn Is ylglated.
Femllo 17 103.82 1.378 1.304
Consequently, we'll use the t test for !qual
vrtanca not umect.

_,... ....

__ - J.....
lewne.. Tntb'Equelttyof
/

--.. -.....
...,.~oiMNnt
11'4 eonw.nc. ~of e-.

. _ u-.....
-- -
0.....
, . .. ....en.. .._
...
.....
I .... ( 2 - )
IQ
............ 214 .710 .1.. 2.1111

e11umed '.2111 21..411 .m ., o6.02t

The t test for ! Cll ual


__, When the Conftdenc. Inmrval of
vartnctJ not umect Dlfferenc. contains zero, the t test Is
(Welch's t test) Is non- non-significant.
significant (at a .OS). Ibm
Is ng difference between tbe
means llf tbe two groyps llf
ggw.

For the wrtte-up, we require


the following:

t .279
df 25.458
Slg (2-talled) .783

5.4.3. Follow-Up Analyses

5.4.3.1. Effect Size

where: sp =
2
(n1-1 )s 1 +(n2-1 )s i
n1+n2-2
60 SPSS Statistics: A Practical Guide

So,

sp =
1924.281 +462.464
= 33

= 33
= .J72.326 = 8.504

Then,

d = 104.61-103.82
8.504
= 0.093
An effect size of this magnitude is trivial. There is virtually no difference (and
certainly not a "significant" one) between the average IQs of the male and
female students in Mrs Sommers' class.

5.4.4. APA Style Results Write-Up

Results

Preliminary assumption testing indicated that both the boys'

(M= 104.61, SD = 10.64) and girls' (M= 103.82, SD = 5.38) IQ

scores were normally distributed, but that there was substantially more

variance in the boys' scores. Consequently, Welch's t test was used


Numbers that cannot
exceed 1 do not need a
to compare the boys' average IQ to that of the girls. The t test was leading zero. So, t = 0.28,
but p = .783.

non-significant, t(25.46) = 0.28,p = .783, two-tailed, d= 0.09, 95%

CI of the mean difference [-5.03, 6.60].

5.5. Independent Samples t Test Checklist


Have you:

., Checked that each group of data is approximately normally distributed?


., Checked for homogeneity of variance?
., Interpreted the appropriate t test (i.e., for equal variances assumed versus equal
variances not assumed}, and taken note of the t value, degrees of freedom, significance
mean difference and confidence interval for your write-up? '
., Calculated a measure of effect size, such as Cohen's d?
., Written up your results in the APA style?
Chapter 6: Paired Samples t Test -AKA:
Repeated measures t
test; Dependent
samples t test; Within
Chapter Overview samples t test; Matched
samples t test;
6.1. Purpose of the Paired Samples tTest ........................................................................... 61 Correlated samples t
6.2. Questions We Could Answer Using the Paired Samples t Test .......................................... 61 test.
6.3. Illustrated Example One ............................................................................................. 62 Additionally, "samples
6.3.1. Setting Up the SPSS Statistics Data File ............................................................... 63 Is sometimes replaced
6.3.2. Analysing the Data ............................................................................................ 63 by groups", "subjects
or "participants".
6.3.2.1. Assumptions ............................................................................................. 63
6.3.2.2. SPSS Statistics Procedure (Part 1: Normality and Normality of Difference Scores)
........................................................................................................................... 64
6.3.2.3. SPSS Statistics Output (Part 1: Normality and Normality of Difference Scores) .. 66
6.3.2.4. SPSS Statistics Procedure (Part 2: Paired Samples t Test) ............................... 67
6.3.2.5. SPSS Statistics Output (Part 2: Paired Samples t Test) ................................... 68
6.3.3. Follow-Up Analyses ............................................................................................ 69
6.3.3.1. Effect Size ................................................................................................ 69
6.3.4. APA Style Results Write-Up ................................................................................. 70
6.3.5. Summary ......................................................................................................... 70
6.4. Illustrated Example Two ............................................................................................. 70
6.4.1. SPSS Statistics Output (Part 1: Normality and Normality of Difference Scores) .......... 71
6.4.2. SPSS Statistics Output (Part 2: Paired Samples tTest) ........................................... 72
6.4.3. Follow-Up Analyses ............................................................................................ 73
6.4.3.1. Effect Size ................................................................................................ 73
6.4.4. APA Style Results Write-Up ................................................................................. 74
6.5. Paired Samples t Test Checklist ................................................................................... 74

6.1. Purpose of the Paired Samples t Test


To test for a statistically significant difference between two related sample
means. Two samples are considered related when:

a. They are both comprised of the same group of individuals, who've


provided data on two separate occasions (e.g., before and after a
treatment).

b. Each individual in one sample is connected or linked with a specific


individual in the other (e.g., husband and wife dyads).

Each of these situations is illustrated in section 6.2.

6.2. Questions We Could Answer Using the


Paired Samples t Test
1. Do people report a higher sense of subjective wellbeing after 15 minutes
of aerobic exercise?

To answer this question, we would need to measure the subjective wellbeing


of each participant twice: before and after they've done 15 minutes of aerobic
exercise. We would then use the t test to compare the mean of the "before"
data with the mean of the "after" data.

2. Is students' short-term memory more accurate when tested in quiet or


noisy surroundings?

Questions 1 and 2 are generally referred to as repeated measures designs.


Question 3 is a little different, as it requires that we recruit participants in
62 SPSS Statistics: A Practical Guide

pairs, but need each to provide us with data on one occasion only. This is
usually referred to as a matched design.

3. Is there a difference between fathers' and mothers' estimates of their


first-born sons' IQ scores?

6.3. Illustrated Example One


A local cycling association encourages its members to wear fluorescent vests
when riding after dark. It argues that doing so makes cyclists more visible to
passing motorists, and thus safer.

To demonstrate this, the association's chairperson hires a driving simulator


that has been specially programmed to drop a virtual cyclist into an evening
driving simulation at random intervals. When a cyclist appears, the "driver"
must respond - as quickly as possible - by pressing a button located on the
simulator steering wheel. The simulator automatically records the time (in
milliseconds) it takes the driver to react to each cyclist.

In a complete testing session (which lasts around 20 minutes) a driver will be


exposed to ten cyclists: five in fluorescent vests (the experimental trials); and
five in non-fluorescent attire (the control trials). These trials are presented in
a random order. After all 10 trials, the driver's average reaction times for both
the experimental and control conditions can be calculated. These averages,
for a group of 15 participants, are reported in Table 6.1.

The chairperson thinks that reaction times will be faster during the
experimental trials (i.e., when the cyclists are wearing fluorescent vests).

Table 6.1
CJ Data:
This is data file
data_6_1.sav on the Average Reaction Times (in Milliseconds) to the Presentation of Virtual Cyclists
companion website.
in Fluorescent and Non-Fluorescent Attire During a Driving Simulation Test
(N = 15)

ID Control (Non-Fiuroescent) Experimental (Fiuroescent)


Trials Average RT (msec) Trials Average RT (msec)

1 345 268
2 540 340
3 430 310
4 470 322
5 420 286
6 364 320
7 388 292
8 392 388
9 378 296
10 362 304
11 420 318
12 446 312
13 452 334
14 434 346
15 498 376
Chapter 6: Paired Samples t Test 63

6.3.1. Setting Up the SPSS Statistics Data File ~Unk:


Setting up a data
tile Is explained In
chapter 1.

(]) np:
The Name you use
will appear at the top
In the Variable VIew, set up two variables: of the variable column
In the Data VIew.
However, the Label
1. Control: The average of the reaction time data collected during
will be used In your
the control (I.e., non-fluorescent) trials.
output.
2. Exp: The average of the reaction time data collected during the
experimental (I.e., fluorescent) trials.

You may want to set up a third variable - ID or Identlflc:atlon


Number - to help you keep track of your participants. This variable
will not be used In any analyses.

In the Data VIew, enter your data.


Work In rows, where each row
represents one participant. For example,
participant 1 (In row 1) averaged 345
msec over the 5 control trials, and 268
msec over the 5 experimental trials.

6.3.2. Analysing the Data esynr:.x:


Run these analyses
with ayntax_l_l.a.-
6.3.2. 1. Assumptions on the companion
website.
Three assumptions should be met before conducting a paired samples t test.
The first is methodological, and is addressed when selecting measures. The
second and third can be tested with SPSS Statistics.

1. Scale of Measurement. Interval or ratio data are required for a paired


samples t test. If your data are ordinal or nominal, you should consider a
non-parametric test instead (see chapter 17).
64 SPSS Statistics: A Practical Guide

2. Normality. Each group of scores should be approximately normally


distributed.

3. Normality of Difference Scores. The differences between pairs of


scores should be approximately normally distributed.

6. 3. 2. 2. SPSS Statistics Procedure (Part 1: Normality and Nonnality of


Difference Scores)

Vlllalllu-.

~~~--~~~i~~. 88COCII"*' o..nntv.talllea_


~!!!!!!llt--~+-J ~R~.
~-~i--~+-JII 1111u11 !!ftllnO...
17'!!!'!!!!'!!'4f----:~l(:~~

2 1--------.
Type a name
<D Tip: for the new
variable under
Define other aspects
of the new variable by Terget
clicking the Type Verteble.
Lllbel button. We've used dlff
as shorthand
for d lfference
scores.

Build the
Numeric
Exprenlon
needed to
calculate the
difference
scores using
the list of
available
variables, the
arrow button
and the
keypad.

Specifically, select, Corlrai(Nan.I'Uo)(a~.;;;i, click.; click.;


select, ExperlrnerUI (FU"O) (e ...; then click

Your numeric expression should now match the one pictured Click OK to create the
above. new variable.
Chapter 6: Paired Samples t Test 65

T....
caa.-..... <D Tip:
fieMnlu.e.lllloclll Although obscured by
Genlrllpd ...... ..,....
the Analyze menu In

IIIJ!Id IIIOCIIII ~ this Illustration, there


Is now a third variable
~ ~Piall-. In the data file, called
dlff.
~ g.a f'lall...
The first participant's

.............
L!llllnelr

C2llllt
score on dltr Is 77. This
means that his/her
average control trial
reaction time of 345
msec was 77 msec
slower than hiS/her
average experimental
~T. . .
trial reaction time of
268 msec.
F--.o
gaM~!

..................._
lll!flllle RlltflanM

Mllll!lllle .........
~SimiiiH
Ill---
Qulllreonnl
ROC~.-
6 7
In l!xplo,.., move the control, Select Both to display all the options
experimental and difference scores available In l!xplo..., or either Smtllltlc:a or
variables Into the De.,.nclent Ullt. Plots for a more limited range.

Here, we've limited our choices to just Plots.

Some HllltOgrama will give us an


adequate sense of whether or not the
assumptions of normality and normality of
difference scores are violated.

I~ lliilillllilil ~~
1!1 ...........
~Link:

oDIJlf....., ..,....
, o..-
There are many ways
to assess normality.
Several of these are

...................
', ~ Illustrated In chapter 4.

I
I
e-.
~~-~~~----~
.'flllllt.......
~~---'
ey.naiiintllll

[ _____~~--
Click Continue to exit this dialogue.

Click OK to output the


histograms.
66 SPSS Statistics: A Practical Guide

6.3.2.3. SPSS Statistics Output (Part 1: Normality and Normality of


Difference Scores)

Explore The Ceee Proceeelng Summ11ry


shows how many cases were analysed,
Coon and whether any were excluded due to
1/olld Mating Tollll missing data.
N Percent N Pwrcont N Percent
Control (NOf>.fluro) 15 100.0% 0 .0% 15 100.0%
EJpertmenlill (Fiuro) 15 100.~ 0 .0% 15 100.0%
dHI 15 100.0% 0 .0% 15 100.0%

Control (Non-Fiuro)
On a Hletogrem, the vertical axis
represents the frequency of each band
w..rta:a..e (or range) of values, which are ordered
lld.Div.D.D5
Ntl from smallest to largest along the
()) np: horizontal axis.
Double-click a graph Looking at the first histogram, one
to open up the Olert participant had an average reaction
Editor. In this editor time between 300 and 350 msec
you can superimpose during the Control trials; five had
a normal curve on top averages between 350 and 400 msec;
of a histogram, as well and so on.
as change Its
appearance in Ideally, a histogram should look
numerous ways. Turn reasonably "normal". That Is, It should
to chapter 3 for a resemble an "Inverted u with greatest
closer look at the frequency of cases clustered around
SPSS Statistics Chert the mean, and progressively fewer
Editor. cases towards the tails.

Even though these histograms all have


slightly different appearances they are
all relatively normal, thus~
the normalitY and normality of
difference scores assumotlons.
Experimental (Fiuro)

If the Auumptlons ere VIolated

Small to moderate violations of the


f normality assumptions are of little
concern In samples of 30+ pairs.
l Severe violations may prompt the
researcher to consider transformations
(see Tabachnlck & Adell, 2013) or a
non-parametric alternative such as the
Wilcoxon signed rank test (see chapter
17).

.,_...,..,... (Fluro)

dlff

f
l

dill
Chapter 6: Paired Samples t Test 67

6.3.2.4. SPSS Statistics Procedure (Part 2: Paired Samples t Test)

g..IIIJIINrllodll
GeMrllpciUIIWIIDdlll
lll!!tCIIIDdlll
~

.......
BtgiiNIGII

.............
~
~Ridlldlall
Sqle
~.........-T-
~~~
.......
...........
llil!llllle RleiiCNe
,....._
................
COmllll---
lis ........
9AIIr CGI*1III
.ROCCUI!L

2 r-----------------------,
In the Palred-Samplea T Tellt ]~~=---~==========~~
1
The default (]) np:
dialogue select the pair of variables to Paired Samplea T 8 Paired-Samples T Test Options To select multiple
be compared and click the arrow Teat: Option will variables, hold down
button to move them Into the first row
of the Paired Varlablea list.
be fine In most
situations.
~--~.::::~- the Shift: key on your
keyboard, then dick
each with your mouse.
If you want to make additional Click Continue to
comparisons, repeat this process with close this window. One click will select a
additional pairs of variables. variable; a second
click will deselect lt.

Paired-Samples T Test

Click OK to
output the t test.
68 SPSS Statistics: A Practical Guide

6.3.2.5. SPSS Statistics Output (Part 2: Paired Samples t Test)

TTest
The Paired Sample StatlatiCII table Includes

I some basic descriptive data, Including the means


and standard deviations for each group of scores:

Control (Non-Fiuro): M = 422.60, SD = 53.835


Experimental (Fiuro): M = 320.80, SD = 32.371
Pair 1 Control (Non..fluro)
It certainly appears as though reaction times were
Elilpartmanllll (Fluro)
faster during the experimental trials, but Is this
effect statistically significant?

The paired samples t test will answer this question.

If we repeated this research many hundreds of times (with samples of N = 15 all drawn from the same
population), and calculated a pair of means for each replication, we would find that most of these means varied
from those presented In the Paired Sample Statlatlca table. The Std. Error Mean Is an estimate of this
variability (expressed In standard deviation units).

A larger standard error of the mean Indicates more variability In the sampling population.

The Paired Sample Correlation table reports a

~Link:
I Pearson's correlation coefficient, which Is a
commonly used Index of the strength of the linear
association between the two groups of scores.

Here, the correlation coefficient is .499, indicating


that partlcloants who responded faster to the
Correlation is covered Pair 1 Control (Non-Auro) & fluorescent cyclists tended to also respond faster
81pertmantlll (Auro)
In chapter 12. to the control cyclists. Although "large" by Cohen's
( 1988) conventions, this correlation failed to reach
statistical significance (at a = .05), primarily due
to the small sample size. This highlights the perils
of conducting underpowered research.

Typically, you would not report this correlation


when writing up a paired samples t test.

Palnd 81mplla Teat


,.,. Paired Differences

_.. ~
v 95% Confidence lnlltMI of the
Dltfarence

Std. Da'tjjation
Std. Error
Moan Lower I Upper l df Slg. (21BIIed)
Palr1
Con"rf'~)
E>q>erl (Fiuro)
101.800 48.987 12.132 75.780 I 127.820 8.391 14 .000

(
In the Paired Difference section of the Paired The t test Is statistically significant at a = .05.
Sample Teat table there are several useful pieces of Reaction times were significantly faster when the
(D Tip: Information: cyclists were wearing fluorescent vests.
The difference
Mean. The difference between the two sample means For your write-up, make note of:
between the two
(422.60 - 320.80 = 101.800).
means is simply t x
Std. Error Mean.
t = 8.391
Std. Deviation. The standard deviation of the difference df = 14
scores. Here, the difference scores deviate from the 5/g (2-talled) = .000
8.391 X 12.132 =
mean of 101.800 msec by an average of 46.987 msec.
101.800
Note. SPSS Statistics rounds to three decimal
95~ Confidence Interval of the Difference. We can places. Sig (2-talled) = .000 simply means that the
be 95% confident that this Interval (75. 780 to 127 .820) likelihood of observing a t value of 2: 8.391 If the
contains the true (population) mean difference. null hypothesis Is true (i.e., if the population mean
difference Is zero) Is less than .0005.
Chapter 6: Paired Samples t Test 69

6.3.3. Follow-Up Analyses

6.3.3.1. Effect Size


Cohen's d can be used to assess the size of the difference between two
related sample means. It can be calculated with the following formula:

Where M1 and M2 are the two sample means and 5p is the pooled standard (D Tip:
deviation, calculated as: There are many
websltes that you can
use to calculate
Cohen's d. Just make
sure you use one for a
paired sample t-
test.

Where 51 and 52 are the two sample standard deviations. All of these figures
can be derived from the Paired Samples Statistics table:

Pair 1 Control (Non-Auro)


~rtmental (Fluro)

So,

53.835 + 32.371
s =-------
p 2
=43.103

Then,

d = 422.60-320.80
43.103 (D Tip:
Jacob Cohen ( 1988)
=2.36 suggests that a d of
.20 can be considered
small, ad of .50 is
medium, and a d of
.80 Is large.
This is a very large effect. Participants in this study responded substantially
He also stresses that
faster to the fluorescent cyclists than they did to the control (non-fluorescent) these values were
subjectively derived
cyclists. with the Intent that a
medium effect should
be "vlslb le to the
naked eye of a careful
observer, a small
effect should be
"noticeably smaller
than medium but not
so small as to be
trivial", and a large
effect should be "the
same distance above
medium as small [is]
below It" (Cohen,
1992, p. 156).
70 SPSS Statistics: A Practical Guide

6.3.4. APA Style Results Write-Up

A comprehensive results
Results section should provide:

A description of each test


A paired samples t test with an a of .05 was used to compare used, and Its purpose.
Descriptive statistics for
each group of scores.
mean reaction times (in milliseconds) to virtual cyclists wearing either Each test's outcome.
The size and direction of
each effect.
fluorescent (M= 320.80, SD = 32.37) or non-fluorescent (M= 422.60,

SD = 53.84) vests in an evening driving simulation. On average, the

participants reacted 101.80 msec, 95% CI [75.78, 127.82], faster


Typically, assumptions only
get mentioned in published
during the fluorescent trials than they did during the non-fluorescent research reports when they
are violated.
trials. This difference was statistically significant, t(14) = 8.39, However, a greater level of
detail Is usually expected In
(D Tip: student papers. This may
As Sig/p can never be p < .001, and large, d= 2.36. Include descriptions of:
zero, report p < .001,
rather than p = .000. How each assumption
It was concluded that the assumptions of normality and was tested, and whether
or not any were violated.
(If applicable) how any
normality of difference scores were not violated after outputting and violations were dealt
with.

visually inspecting the relevant histograms.

6.3.5. Summary
The effect observed in the first illustrated example was large and significant.
The effect in example two is not so clear.

6.4. Illustrated Exam pie Two


A psychologist wishes to assess the efficacy of a new cognitive-behavioural
treatment for generalised anxiety. Before beginning the treatment with a new
client, she asks him/her to complete the Beck Anxiety Inventory (BAI). The
pre-treatment BAI scores for 12 clients are listed in the first column of Table
6.2. Scores on this measure can range from 0 through to 63.

At the end of the eight-week treatment program, the psychologist asked each
client to again complete the BAl. The post-treatment BAI scores for the 10
clients who completed the full eight-week program are listed in the second
column of Table 6.2. Post-treatment BAI data are not available for two clients
who dropped out of the program after two and five weeks respectively.
Chapter 6: Paired Samples t Test 71

Table 6.2
CJ Dbl:
Beck Anxiety Inventory (BAI) Scores Before and After an Eight-Week This Is data file
dlit11_1_2.AY on the
Treatment Program for Generalised Anxiety Disorder (N = 12) companion website.

Client ID Pre-Treatment BAI Post-Treatment BAI


Score Score

1 55 53
2 36
3 39 37
4 53 49
5 46 40
6 52
7 47 39
8 45 44
9 34 38
10 41 43
11 33 29
12 50 46

6.4.1. SPSS Statistics Output (Part 1: Normality and Normality of


Difference Scores) !! Synblx:
This Explore output was generated Run these analyses
with the same procedures as those used with synblx_l_2.pa
Explore In Illustrated Example One. on the companion
website.
The diff (difference scores) variable was
c created In Compute Vrtble (In the
lllllld M.. lng TaWil Trnsfonn menu) by subtracting the
N Pen:ont N Pereent N Percent post-treatment BAI variable from the
p,.Trwatment BAI 10 83.3% 2 18.7% 12 100.0% pre-treatment BAI variable.
Poll-TNalment BAI 83.3% 2 18.7% 12 100.0%
'10
dill 10 83.3% 2 18.7% 12 100.0%

\ When the Exclude cee llstwle


strategy for handling missing values Is
selected In Explore: Options, only
cases that have data for every variable
in both the Dependent Llt and the
Fctor List are used In the analyses.
Pre-Treatment BAI Explore: Options

-------,

As dlents 2 and 6 are missing data on


some of the variables used In Explore,
( speclflca lly, the post-treatment BAI
and difference scores variables) they
are not Included In any of the statistics
or graphs.

If we used Exclude ce- p11lrwlse


Instead, the pre-treatment BA1
histogram would have been based on
data from 12 clients, whereas the post
tre11tment BA1 and difference scores
histograms would have been created
using data from only 10 clients. This
PreTr..tnwnt BAI would make comparisons difficult.
72 SPSS Statistics: A Practical Guide
Assessing the normality of a distribution
of scores using a histogram can take a bit
of practice. Over time, you will develop a
Post-Treatment BAI sense of what "normal enough" looks like,
Hilt'~ ram and be able to apply this mental template
to each histogram you create.
... ~ .......41.1
lld.Diw.8.771
NtO In the meantime, you can easily confirm
your assessments with the Slulplro-WIIk
test. If the S/g value associated with the
,. p..-...;;.. Shaplro-WIIk statistic Is greater than the
specified alpha level (usually .OS), you
can assume normality.

I :r
The Shaplro-Wilk test is a part of the
Normality ploa with wa option In
Explore: Plot8:

II Explore: Plots
' r--

E
Ballltoll
0 fadar ...... loglelll' ~

'"
Oil .. It .. o~.....,
.ftanl
-
dlff
.._Ill~~~~
In this example, the three histograms
look reasonably normal (although with so
few cases, It can be difficult to tell). This
assessment Is supported by the three
Shaplro-WIIk tests, which are all non-
significant (at a .OS).=
Ibe carmalltll acd carmalltl af dlfferecte
S!:Qrt:S iiSSUmlrti!!DS ilrt: D!lt lll!lliitlld.

Kolmogorov-Smlmo.,
~ Links: Stllllltc df Slg. Slg.
Some Normal Q-Q Pre-TretmentBAI .137 10 .200' 10 .750
Ploa and Detrended Poat-Tr..tmentBAI .138 10 .zoo 10 .880
Normal Q-Q Ploa, d111 .1113 10 .200' 10 .788
were generated along
with these normality LMIIefDrl&gntllc8r'ICII Coi'Ndlon
tests. Guidelines for .lllllll: lowerboundollleW.Igntbnc:e.
Interpreting each can
be found In chapter 4.

6.4.2. SPSS Statistics Output (Part 2: Paired Samples t Test)

T-Test The purpose of the Paired Sllmpla t Tat


Is to compare two related sample means. In
this example, we are comparing anxiety data

--entiiN - -
collected before and after an eight-week
treatment program.
Sld.Emw
N Sid.-
-1 44.30 10 7AII8 2.371
PaoH-oniMI nao 10 1.778 2.144
The Paired Sample Teat table Indicates
that the difference between the pre- and
post-treatment means Is D!lt statlstlcailll
Polrod . . . . . . c.rr.- slgclflcact (at a .OS). Make note of:
(D Tip:
You will sometimes
see tests with Stg
IPWr1 --ontMI& N 10 ICorM: I ~1 t = 2.216
df 9
values only slightly Stg (2-talled) .OS4
above the specified
alpha level (usually
.OS) described as Plred Oiffllrence1
approaching 95% Confidence lntBMI of lhe
significance In the Dttllrwnce
research literature. Sid. Error
Moon Sid. 0.-AIon Moon Lower I Upper I df Slg. (21ollod)
Palr1 Pre-Treatment BAI- Poat.
TrutmentBAI
2.500 3.587 1.128 -.052 I 5.052 2.216 8 .054
Chapter 6: Paired Samples t Test 73

6.4.3. Follow-Up Analyses

6. 4. 3. 1. Effect Size

Where M1 and M2 are the two sample means and Sp is the pooled standard
deviation (the average of the two sample standard deviations). So,

d = 44.30- 41.80 2.5

=0.35
7.139 7.139
''---- 8
p
=s,+s 2 = 7.499+6.779
2
=7.139
2

Using Cohen's {1988) conventions as a guide, d = 0.35 is within the small to


medium range.

This presents us with an interesting dilemma. We've observed a non-trivial


effect size (d = 0.35), yet it's statistically non-significant. Why? The answer is
tied to statistical power.

Statistical power refers to the likelihood of detecting an effect, where one ~ Links:
actually exists. (By detecting, we mean finding p < .OS.) Generally speaking, We recommend
Howell (2013b) for
we are more likely to detect larger effects, in larger samples. Conversely, an accessible
Introduction to
small effects in small samples can be particularly difficult to catch. As the statistical power.
current sample is quite small (N = 10), it is possible that the study was
underpowered. That is, it is possible that the psychologist did not give herself
a very good chance of finding a statistically significant effect.

Retrospective power estimates are provided as part of the output for many <D Tip:
SPSS Statistics procedures. When they are not provided, you can use the Retrospective power
analysis can be
power tables included in the appendices of many good textbooks (e.g., controversial. Power is
best considered (and
Howell, 2013b), or one of several power calculators. The calculator that we most useful) before
use- G*Power- is maintained by Faul, Erdfelder, Lang, and Buchner (2007) collecting data, when
it can be used to
and can be freely downloaded from http://www.gpower.hhu.de/en.html calculate an
appropriate sample
slle. You should avoid
Retrospective power analyses with G*Power confirmed that the t test was (mls)uslng It to
explain away non-
likely underpowered. It is recommended that the psychologist replicate this significant results.
study with a much larger sample. She will need around 70 participants to
have a decent (i.e., 80%) chance of observing an effect of around d = 0.35.
74 SPSS Statistics: A Practical Guide

6.4.4. APA Style Results Write-Up

Results

A two-tailed, paired samples t test with an alpha level of In her Method section the
author should note that 12
people began the study,
.05 was used to compare the pre- (M = 44.30, SD = 7.50) and post- but only 10 completed it. If
known, the reasons for this
attrition should also be
reported.
treatment (M= 41.80, SD = 6.78) Beck Anxiety Inventory (BAI)
In her Results section, she
Is correct to report that the
scores of l 0 individuals. On average, participants' post-treatment BAI analyses are based on data
from just 10 individuals.

scores were 2.5 points lower than their pre-treatment scores, 95% CI

CD Tip:
Numbers that cannot
exceed 1 do not
[--0.05, 5.05]. However, this difference was not statistically

-
significant, t(9) = 2.22, p = .054. Cohen's d for this test was 0.35,
In the Results section,
simply report what was
found In the data.

Interpretations, speculation
need a leading zero. which can be described as small to medium. about why the test did not
Therefore, p = .054, support the hypothesis and
but d = 0.35. suggestions for subsequent
Visual inspection of the relevant histograms indicated that research should all be
saved for the Discussion.

neither the normality nor normality of difference scores assumptions

were violated.

6.5. Paired Samples t Test Checklist


Have you:

"' Checked that each group of data is approximately normally distributed?


"' Checked that the difference scores are normally distributed?
"' Interpreted the results of the t test and taken note of the t value, degrees of freedom,
significance, mean difference and confidence interval for your write-up?
"' Calculated a measure of effect size, such as Cohen's d?
"' Reported your results in the APA style?