Psych 218 – Fourth SPSS Tutorial

Repeated Measures Analysis and
Factorial Design ANOA
Part !" Repeated Measures ANOA #R$%p designs&
Study Background:
You are interested in whether females with a past history of receiving abuse prefer male dating partners
that may be potentially abusive. You decide to assess preferences in dating partners by having female
participants read personal ads and select those individuals they would be interested in getting to know
better. Before you can even begin your study, you need to construct and validate the stimuli to be used in
the study. More specifically, you’ve decided that you want three types of personal ads to be used as
stimuli for your study!"# ads that describe males with a high potential for engaging in abusive
behaviors, !$# ads that describe males with a moderate potential for engaging in abusive behaviors, and
!%# ads that describe males with a low potential for engaging in abusive behaviors. &n order to develop the
personal ads, you have male participants in another study !'( ""$# describe themselves in the form of a
personal ad. You then have three coders rate these ads on potential for engaging in abusive behaviors.
Based on the three coders’ ratings, you select ) ads that are high abusive, ) ads that are moderate abusive,
and ) ads that are low abusive. You then want to validate your classification of these ads !before you use
them as stimuli in your study#, and administer these "$ ads to a sample of females !'($)#. 'ote that
these females were blind to the a priori classification of the ads and to the purpose of the study. You are
interested in whether the ratings of the ads is consistent with your classification of the ads. You identify
this as a *B+% design !*andom Block ,esign with three levels# and plan to run a repeated measures
-'./- to e0amine whether the ads you selected as stimuli indeed differ in abusiveness.
'otice that each row represents a sub1ect, and that each sub1ect has three dependent variables. -ll three
variables !psyhigh, psymod, and psynolo# are 2&34&' S5B6783S factors, which means that every
participant was e0posed and asked to rate all three types of personal ads. 'otice that there are no
B73277' S5B6783S factors.
-gain, you are mainly interested in whether the three types of personal ads differ in ratings of potential
for abusiveness. &f they differ in ways that are consistent with your classification, then you can use these
ads in your study assessing females’ preferences in dating partners.
3he dependent variables are ratings of
potential for abusiveness:
9psyhigh9 refers to ratings of potential
for engaging in abusive behaviors for
personal ads classified a priori as high
abuse.
9psymod9 refers to ratings of potential
for engaging in abusive behaviors for
personal ads classified a priori as
moderate abuse.
9psynolo9 refers to ratings of potential
for engaging in abusive behaviors for
personal ads classified a priori as low
abuse.
3o download the data file, click here. -s
usual, you will need to save this file to a
location on your computer and open it up.
Before you begin analy:ing the data, you decide to look at the data in bo0plot form by selecting:
'RAP(S ) $O*P+OT ) S!MP+, #su--aries .or separate /aria0les&
You define your variables !psyhigh, psymod, psynolo#.
24 24 24 N =
PSYNOLO PSYMOD PSYHIGH
4
3
2
1
0
-1
24
12
4ow do you go about e0amining differences between the ads on ratings of abusiveness;
2ell, for all of these effects, you want to know if they are statistically significant<to go about doing this,
you will run a new command in S=SS:
-'->Y?7 @ A7'7*-> >&'7-* M.,7> @ *7=7-37, M7-S5*7S
Your bo0 plot output should look like this.
&t looks like the three types of ads differ in
mean ratings of Bpotential for
abusivenessC. .f course, the only way to
draw this conclusion is through statistical
tests.
You will get the following dialog bo0:
2hat is your 2ithin+Sub1ect Dactor;
&n this scenario, it is the level of
abusiveness for the personal ads
!high, moderate, or low#.
You can give this factor a name !e.g.,
adEabuse# by replacing Bfactor"C in
the te0tbo0 ne0t to 2ithin+Sub1ect
Dactor 'ame:
You now need to specify the number
of levels for your 2ithin+Sub1ect
Dactor. &n this scenario, there are
three levels. -ds are high, moderate,
or low on potential for abuse.
'ow, click on B-ddC to add the
factor to your model.
Your dialog bo0 should now look
like this:
3he ne0t step is to define the 2ithin+
Sub1ect Dactor. &n order to do this
click on B,efineC.
'ow, you need to identify all the
variables that reflect your 2ithin
Sub1ect Dactor !adEabuse#. You do
this by moving the relevant variables
over to the bo0 under 2ithin+
Sub1ects /ariables.
'otice that in our design, we do not
have any Between Sub1ects Dactors
or 8ovariates, so we can simply
leave these bo0es empty.
You are ready to click on .F, and run the analysis. S=SS will print out the following output:
General Linear Model
Within-Subjects Factors
Measure: MEASURE_1
PSYHIGH
PSYMOD
PSYNOLO
AD_ABUSE
1
2
3
Dee!"e!#
$ar%a&'e
Multivariate Tests
b
()10 2*(+24
a
2(000 22(000 (000
(2+0 2*(+24
a
2(000 22(000 (000
2(44, 2*(+24
a
2(000 22(000 (000
2(44, 2*(+24
a
2(000 22(000 (000
P%''a%-s .ra/e
0%'1s- La2&"a
H3#e''%!4-s .ra/e
R35-s Lar4es# R33#
E66e/#
AD_ABUSE
$a'ue 7
H53#8es%
s "6 Err3r "6 S%4(
E9a/# s#a#%s#%/
a(
Des%4!: I!#er/e#
0%#8%! Su&:e/#s Des%4!: AD_ABUSE
&(
Mauchly's Test of Sphericity
b
Measure: MEASURE_1
(+3* 1(4*4 2 (4)+ (+40 1(000 (;00
0%#8%! Su&:e/#s E66e/#
AD_ABUSE
Mau/8'5-s
0
Ar39(
<8%-S=ua
re "6 S%4(
Gree!83u
se-Ge%ss
er
Hu5!8-7e
'"#
L3>er-&3
u!"
Es%'3!
a
.es#s #8e !u'' 853#8es%s #8a# #8e err3r /3?ar%a!/e 2a#r%9 36 #8e 3r#83!3r2a'%@e" #ra!s63r2e" "ee!"e!# ?ar%a&'es %s
r33r#%3!a' #3 a! %"e!#%#5 2a#r%9(
Ma5 &e use" #3 a":us# #8e "e4rees 36 6ree"32 63r #8e a?era4e" #es#s 36 s%4!%6%/a!/e( <3rre/#e" #es#s are
"%s'a5e" %! #8e .es#s 36 0%#8%!-Su&:e/#s E66e/#s #a&'e(
a(
Des%4!: I!#er/e#
0%#8%! Su&:e/#s Des%4!: AD_ABUSE
&(
Dirst bo0 identifies your within Sub1ect
factor !adEabuse#, and all the dependent
variables that reflect it.
2e are now going to skip over bo0 two and
three, and go directly to the fourth bo0
labeled 3ests of 2ithin+Sub1ect 7ffects.
3his is the output that is most relevant to us.
Tests of Within-Subjects Effects
Measure: MEASURE_1
1+(3,3 2 +(*+1 2+(22+ (000
1+(3,3 1(,)+ 10(31; 2+(22+ (000
1+(3,3 2(000 +(*+1 2+(22+ (000
1+(3,3 1(000 1+(3,3 2+(22+ (000
1;(2;2 4* (332
1;(2;2 43(21, (3;3
1;(2;2 4*(000 (332
1;(2;2 23(000 (**3
S8er%/%#5 Assu2e"
Gree!83use-Ge%sser
Hu5!8-7e'"#
L3>er-&3u!"
S8er%/%#5 Assu2e"
Gree!83use-Ge%sser
Hu5!8-7e'"#
L3>er-&3u!"
S3ur/e
AD_ABUSE
Err3rAAD_ABUSEB
.5e III
Su2 36
S=uares "6
Mea!
S=uare 7 S%4(
Tests of Within-Subjects Contrasts
Measure: MEASURE_1
1,(;42 1 1,(;42 ;*(1+1 (000
(,40 1 (,40 2(;22 (12*
)(;+0 23 (330
)(**2 23 (333
AD_ABUSE
L%!ear
Cua"ra#%/
L%!ear
Cua"ra#%/
S3ur/e
AD_ABUSE
Err3rAAD_ABUSEB
.5e III
Su2 36
S=uares "6
Mea!
S=uare 7 S%4(
-gain, we are most interested in the output for 3ests of 2ithin+Sub1ect 7ffects. 2e are
conveniently assuming that the assumption of 4omogeneity .f treatment ,ifference /ariances
!4.3,/#, also referred to as the sphericity assumption, is not violated. 3his simplifies our
task. Aiven that 4.3,/ !sphericity# is not violated, we can see the effect of ad type !our
within+sub1ect factor# on the dependent variable !ratings of abuse# by looking at the F obtained
!circled below in *7,# for -,E-B5S7 when Sphericity is assumed. 2hat can we conclude;
(o1 is F o0tained co-puted .or a R$%p design2 Similar to 8*+p designs, D obtained is the
ratio between the MS for the treatment effect and MS for error. &n our scenario, MS for
treatment effect is G.HG" and MS error is .%%$. 3hus, D obtained ( G.HG" I .%%$ ( $G.$.
*emember that one of the big differences between the two e0perimental designs is the error
ter-. Dor 8*+p designs, the error term !referred to as MS2A# includes stable individual
differences, as well as chance fluctuations, and measurement error. Dor *B+p designs, the error
term !referred to as MS*7S# does not include variability due to stable individual differences.
3his makes the error term smaller than that obtained from 8*+p designs, and thus in general
makes *B+p designs more powerful that 8*+= designs.
Tests of et!een-Subjects Effects
Measure: MEASURE_1
.ra!s63r2e" $ar%a&'e: A?era4e
1+1();3 1 1+1();3 31*(,,4 (000
13(+1, 23 (*0;
S3ur/e
I!#er/e#
Err3r
.5e III
Su2 36
S=uares "6
Mea!
S=uare 7 S%4(
Part !!" 3o-pletely Rando-i4ed Factorial Designs #3RF%p5 designs&
Study Background:
-dult attachment style refers to the characteristic ways in which adults feel, think, and behave within their
romantic relationships. 3wo dimensions, view of oneself !as positive or negative# and view of others !as
positive or negative# are hypothesi:ed to underlie four adult attachment styles: secure, preoccupied,
dismissing, and fearful. *esearch and theory on adult attachment styles suggests that the differences in
these two dimensions are a result of early relationship e0periences with oneJs primary caregiver, such as
oneJs mother. 3he purpose of the study !from which the present data was obtained# was to e0amine
whether individuals with different attachment styles differ in their automatic associations toward their
mother !i.e., feelings that may e0ist outside of conscious awareness about their motherJs supportiveness#.
'otice that there are two independent variables in this study, view of self and view of others. 3hese two
independent variables have two levels each, positive and negative !in Firk’s terms, this is a 8*D+$$#.
3hus, with these two variables, there are four different possible classifications for each participant in the
study:
"# =ositive view of others, positive view of self !attachoth ( ", attachsel ( "# BsecureC
$# =ositive view of others, negative view of self !attachoth ( ", attachsel ( $# BpreoccupiedC
%# 'egative view of others, positive view of self !attachoth ( $, attachsel ( "# BdismissingC
3he independent variables are two
dimensions underlying adult attachment
styles:
"# 9attachoth9 refers to a view of others
as positive, supportive, trustworthy,
dependable, versus negative, cold,
uncaring. !variable labels are ": positive,
and $: negative#
$# 9attachsel9 refers to a view of self as
positive and worthy of love, versus
negative. !variable labels are ": positive,
and $: negative#
3he dependent variable is the &-3, an
implicit measure that assesses automatic
associations toward oneJs mother as
supportive versus re1ecting !larger scores
on momlniat reflect stronger supportive
associations toward oneJs mother#.
3o download the data file, click here. -s
usual, you will need to save this file to a
location on your computer and open it up.
)# 'egative view of others, negative view of self !attachoth ( $, attachsel ( $# BfearfulC
You can look at the data in bo0plot form by selecting
'RAP(S ) $O*P+OT ) 3+6ST,R,D #su--aries .or groups o. cases&
and defining your variables !B&-3 effect,C or momlniat is your variable of interest, and you can define
either of your &/s as the category a0is, and the other as the Bdefine clusters byC variable#. *egardless of
how you place the two &/s, you will be able to see the four distinct groups mentioned above.
Your bo0plot output is below:
33 22 22 33 N =
?%e> 36 3#8ers
!e4a#%?e ?%e> 36 3#8 3s%#%?e ?%e> 36 3#8
I
A
.

e
6
6
e
/
#
D

A
2
3
2
E
r
e
:
e
/
#
%
!
4
B
-
A
2
3
2
E
s
u

3
r
#
%
?
e
B

'
3
4
1(0
(,
(*
(4
(2
0(0
-(2
?%e> 36 se'6
3s%#%?e ?%e> 36 se'
6
!e4a#%?e ?%e> 36 se'
6
You may also wonder if there is an interaction between view of self and view of others in determining the
way in which one sees one’s mother !as supportive or not#. 3his Kuestion can be phrased in several ways:
"# ,oes the effect of the views of self depend on the effect of the views of others in determining the
dependent variable, view of mother;
$# -re differences in views of mother, as determined by views of self, different depending on how
one views others;
&t appears that, in general, a positive
view of others is related to a view of
the mother as supportive !as compared
to a negative view of others#.
3hough it’s somewhat more difficult to
see here, view of self does not appear
to be as influential in terms of its effect
on views of mother !compare the red
bo0plots with the green ones#.
&f you like, you can switch the two &/s
to get a different view of the data.
3hese are 1ust two e0amples of ways in which one can ask Kuestions about an interaction.
2ell, for all of these effects, you want to know if they are statistically significant<to go about doing this,
you will run a new command in S=SS:
-'->Y?7 @ A7'7*-> >&'7-* M.,7> @ 5'&/-*&-37
!You may be wondering what univariate meansL it refers to the fact that there is one dependent variable in
the study#
You will get the following dialog bo0:
3he BmodelC option in S=SS allows you to specify e0actly which factors you are interested in testing.
3he default is Bfull factorial,C which will test the effects of each of the independent variables, as well as
their interaction.
You can also select the type of sums of sKuares you want !the default is 3ype &&&#. 2hat does this mean;
2ell, it refers to how you would account for sums of sKuares if you had uneKual sample si:es in your
treatment groups. 4ere, you have eKual sample si:es !MM in each level of each condition#, so don’t worry
about this for now.
3he dependent variable has been
entered !momlniat#.
4ow do you enter the two &/s;
&n the Aeneral >inear Model
command in S=SS, these are
considered D&N7, D-83.*S.
Select the two &/s and click them
over into the Bfi0ed factor!s#C bo0.
'e0t, select M.,7>.
8lick on Bcontinue,C then select that you want BoptionsC for the A>M command. You will get the
following:
'ow, you are ready to run your analysis. 2hen you run the test, you will get the following output:
'ote: 3he tutorial
originally asked that you
deselect the Binclude
intercept in modelC
function. 3he output below
corresponds to the intercept
being e0cluded.
3o see what the output
would be if the intercept
were included, go to the
end of the tutorial !look for
the asterisk#.
Select that you want
the descriptive statistics
for the data
et!een-Subjects Factors
3s%#%?e
?%e> 36
3#8er
;;
!e4a#%?e
?%e> 36
3#8ers
;;
3s%#%?e
?%e> 36 se'6
;;
!e4a#%?e
?%e> 36 se'6
;;
1(00
2(00
?%e> 36
3#8ers
1(00
2(00
?%e> 36
se'6
$a'ue La&e' N
"escriptive Statistics
Dee!"e!# $ar%a&'e: IA. e66e/#D A232Ere:e/#%!4B-A232Esu3r#%?eB '34
(3,2; (1+0+ 33
(3;;0 (1,*) 22
(3)1; (1,,0 ;;
(2;*) (1)++ 22
(2;4, (12)* 33
(2;;* (14+1 ;;
(3321 (1+;1 ;;
(2+4+ (1*02 ;;
(313; (1),) 110
?%e> 36 se'6
3s%#%?e ?%e> 36 se'6
!e4a#%?e ?%e> 36 se'6
.3#a'
3s%#%?e ?%e> 36 se'6
!e4a#%?e ?%e> 36 se'6
.3#a'
3s%#%?e ?%e> 36 se'6
!e4a#%?e ?%e> 36 se'6
.3#a'
?%e> 36 3#8ers
3s%#%?e ?%e> 36 3#8er
!e4a#%?e ?%e> 36 3#8ers
.3#a'
Mea! S#"( De?%a#%3! N
No17 this is the ta0le o. interest" Notice that you get an F /alue and a p%/alue .or each o. your
independent /aria0les7 as 1ell as .or the interaction" 8hich e..ect#s& are signi.icant in this study2

Tests of et!een-Subjects Effects
Dee!"e!# $ar%a&'e: IA. e66e/#D A232Ere:e/#%!4B-A232Esu3r#%?eB '34
11(1+2
a
4 2()+, +;(*)) (000
(33) 1 (33) 11(;24 (001
;(*),E-03 1 ;(*),E-03 (1+4 (**0
4(3;;E-03 1 4(3;;E-03 (14+ ()00
3(100 10* 2(+24E-02
14(2+2 110
S3ur/e
M3"e'
A..A<O.H
A..A<SEL
A..A<O.H F A..A<SEL
Err3r
.3#a'
.5e III Su2
36 S=uares "6 Mea! S=uare 7 S%4(
R S=uare" = (),3 AA":us#e" R S=uare" = ());B
a(
3he first bo0 lists the number of
sub1ects you have in each level of each
independent variable.
3he second bo0, as usual, provides means
and standard deviations.
*The .ollo1ing output corresponds to 1hat you 1ould see i. you selected to include
the intercept in the -odel9 #the ta0le 0elo1 is the only part that 1ould 0e
di..erent&
Tests of et!een-Subjects Effects
Dee!"e!# $ar%a&'e: IA. e66e/#D A232Ere:e/#%!4B-A232Esu3r#%?eB '34
(3,0
a
3 (12) 4(32* (00*
10(2+; 1 10(2+; 3;2(03+ (000
(33) 1 (33) 11(;24 (001
;(*),E-03 1 ;(*),E-03 (1+4 (**0
4(3;;E-03 1 4(3;;E-03 (14+ ()00
3(100 10* 2(+24E-02
14(2+2 110
3(4)+ 10+
S3ur/e
<3rre/#e" M3"e'
I!#er/e#
A..A<O.H
A..A<SEL
A..A<O.H F A..A<SEL
Err3r
.3#a'
<3rre/#e" .3#a'
.5e III Su2
36 S=uares "6 Mea! S=uare 7 S%4(
R S=uare" = (10+ AA":us#e" R S=uare" = (0,4B
a(
'otice that your effects of your two independent variables, and of the interaction, remain the same. Your
error term is also the same. 2hat is different is that the corrected model does not include the intercept.
Your total SSbetween is that which is listed in your Bcorrected modelC row, .%OP. &t is the total for the
two main effects and the interaction. Your Bcorrected totalC is the total sums of sKuares in the e0periment.
So, regardless of whether you include the intercept in the model, you have the following results:
"# 3here is a significant main effect of the view of others on views of the mother, with positive views
of others resulting in more positive views of the mother, D !", "PH# ( "".M$), p ( .PP".
$# 3here is not a significant main effect of views of the self on views of the mother, D !", "PH# ( .
"G), p ( .HHP.
%# 3here is not a significant interaction of views of the self and views of other on how one regards
one’s mother on the &-3, D !", "PH# ( .")G, p ( .QPP.
&f you wanted to complete a summary table for your -'./-, it would be as follows:
Source df SS MS F p
Between 3 .380

View of Others 1 .337 .337 11.524
.001
View of Sef 1 .00! .00! .1"4
.!!0
#nteraction 1 .004 .004 .14"
.700
$ithin 10! 3.1 .02"

%ota 10" 3.47"