Z1. Studiu de Caz Mihai Z1. studiu de caz Mihai Ungurenus_935.pdfUngurenus_935
Multiple Correspondence Analysis 1 Supplementary points Z1...
Transcript of Multiple Correspondence Analysis 1 Supplementary points Z1...
Multiple categorical variablesMultiple Correspondence Analysis
Supplementary pointsMissing and “middle” responses
Supplementary points
1
J1
K
1 J 1 KBurt matrix
B
Suppose J substantive categories and Kdemographic groups
1 J 1 K
Z1 Z2
Z1TZ1
Z2TZ1
Z1TZ2
Z2TZ2
Supplementary points
1
J1
K
1 J 1 KBurt matrix
BSuppose Jsubstantive
categories andK demographic
groups
Indicatormatrix ofindividual
respondent(case) data
Z2TZ1
Z1
shows each case as a point, at the average of his/her responses
shows eachdemographic category, at the average of thecases in this category
Z1TZ1
CA of B (or adjusted): standard coordinatesthe same
Data set “women94”Substantive variables: Do you strongly agree/ agree/ neither…nor…/ disagree/ strongly disagree to these statements…A: a working mother can establish a warm relationship with her childB: a pre-school child suffers if his or her mother worksC: when a woman works the family life suffersD: what women really want is a home and kidsE: running a household is just as satisfying as a paid jobF: work is best for a woman’s independenceG: a man’s job is to work; a woman’s job is the householdH: working women should get paid maternity leaveDemographic variablesg: gender (1=male, 2=female)m: marital status (1=married/living as married, 2=widowed, 3=divorced, 4=separated, but married, 5=single, never married)e: education (0=no formal education, 1=lowest education, 2= abovelowest education, 3=higher secondary completed, 4=above higher secondary level, below full university, 5=university degree completeda: age (1=16-25 years, 2= 26-35, 3=36-45, 4=46-55, 5=56-65, 6=66 and older) Sample: Spanish sample (year 2002); N=2471 (including missing values)
ISSP 1994 survey on Family and ChangingGender Roles
A [+] A working mother can establish just as warm and secure a relationship with her children as a mother who does not work
B [–] A pre-school child is likely to suffer if his or her mother works C [–] All in all, family life suffers when the woman has a full-time jobD [–] A job is all right, but what most women really want is a home and children E [?] Being a housewife is just as fulfilling as working for payF [+] Having a job is the best way for a woman to be an independent personG [?] Most women have to work these days to support their familiesH [+] Both the man and woman should contribute to the household incomeI [–] A man’s job is to earn money; a woman’s job is to look after the home and
familyJ [?] It is not good if the man stays at home and cares for the children and the
woman goes out to workK [?] Family life often suffers because men concentrate too much on their work
24 countries (N=33,590)
+ in favour of working women – against women working ? not clear
Data – middle alternative
Common to both surveys is the measurement scale, a 5-point bipolar scale:
strongly agree neither agree disagree stronglyagree nor disagree disagree
We are particularly interested in the middle alternative, how itassociates
– with other middle alternatives
– with other response categories
– with the demographic covariates
Of course, there are also missing values, but since we will analyse the data at the nominal level, a missing value is just an additional category
extreme categories
moderate categories
missing
Background & previous researchPresser & Schuman (1980)
The measurement of a middle position in attitude surveysThe Public Opinion Quarterly.Revised version in Schuman & Presser, Questions & Answers in Attitude Surveys. Experiments on Question Form, Wording, and Context. Sage, 1996
Use contingency tables and 2 tests in 5 split-ballot experiments.
They assess the consequences of offering/omitting “a logical middle position, for example whether one is liberal or conservative could be answered by ‘middle-of-the-road’ ”…. “Although there is a very slight decrease in the proportion of spontaneous ‘don't know’ responses when the middle alternative is offered, almost all the change in the middle position comes from a decline in the polar positions.”
Background & previous research
Andrich, de Jong & Sheridan (1997)
Diagnostic opportunities with the Rasch model for ordered response categories.In: Applications of latent trait and latent class models in the social sciences
Use Rasch modelling.
“…the middle category designated as Neutral, Not Sure or Undecided in the Likert-style response format … should not be treated as an attitude more or less somewhere between a negativeand a positive attitude.”
Background & previous researchGonzález-Romá and Espejo (2003)
Testing the middle response categories «Not sure», «In between» and «?» in polytomous items.Psicotema
Use Bock’s Nominal Model to verify ordering.
They use different wordings for the middle category, and find that the ordering is verified only for the wording “in between”.
Hernández, Espejo and González-Romá (2006)
The functioning of central categories Middle Level and Sometimes in graded response scales: Does the label matter?Psicotema
Background & previous researchO’Muircheartaigh, Krosnick & Helic (2000)Middle alternatives, acquiescence, and the quality of questionnaire data.Working paper posted on web.
Use response counts and structural equation modelling
‘Approximately half the respondents … were asked to select an answer from among the following five alternatives: strongly agree, agree to some extent, neither agree nor disagree, disagree to some extent, strongly disagree. These responses were coded 5, 4, 3, 2, and 1,respectively. The other half of the respondents … were asked to select an answer from a set of four, omitting the “neither agree nor disagree”option. Answers were coded 5, 3.66, 2.33, and 1, respectively.’
…“We also found evidence of acquiescence response bias in answers to the agree/disagree items...”
MethodsWe use correspondence analysis (CA) and several of its variants to visualize and interpret the relative positions of the response categories, at a country level, a respondent level and demographic-subgroup level.
1. Simple CA provides maps of aggregated or count data, for example proportions of question responses for the set of countries.
2. Subset correspondence analysis permits focusing on a particular set of response categories so that we can eliminate missing responses, for example, or restrict our attention to specific categories such as the middle responses.
3. Multiple correspondence analysis (MCA) and its refinement joint correspondence analysis (JCA) provide maps of individual-level data, concentrating on the two-way relationships between the questions. In this respect MCA functions like an exploratory factor analysis for categorical data (on nominal scales). Usually we are not interested in individual case points but in mean points for the demographic and other external variables (e.g., age, level of interest…)
4. Canonical correspondence analysis (CCA) focuses on (or partials out) external variables – thus we can explore variation in the responses that is focused on interest, for example, or eliminate aquiescence effects.
Question B: A pre-school child is likely to suffer if his or her mother works
B-agree B-MX B-disagreeAU 0.497 0.149 0.354DW 0.684 0.141 0.175DE 0.325 0.186 0.490GB 0.377 0.215 0.408NI 0.382 0.187 0.431US 0.406 0.142 0.452A 0.720 0.106 0.174H 0.727 0.160 0.113I 0.671 0.152 0.177IR 0.472 0.120 0.407NL 0.439 0.225 0.336N 0.359 0.218 0.424S 0.272 0.267 0.461CZ 0.487 0.197 0.315SL 0.584 0.177 0.238PL 0.664 0.115 0.220BG 0.682 0.178 0.140RU 0.717 0.130 0.153NZ 0.498 0.182 0.320CD 0.306 0.183 0.511RP 0.564 0.144 0.292IL 0.434 0.193 0.373J 0.373 0.238 0.389E 0.516 0.137 0.346ave 0.514 0.171 0.316 Ternary coordinates: Euclidean distance
B: A pre-school child is likely to suffer if his or her mother works
Question B: A pre-school child is likely to suffer if his or her mother works
B-agree B-MX B-disagreeAU 0.497 0.149 0.354DW 0.684 0.141 0.175DE 0.325 0.186 0.490GB 0.377 0.215 0.408NI 0.382 0.187 0.431US 0.406 0.142 0.452A 0.720 0.106 0.174H 0.727 0.160 0.113I 0.671 0.152 0.177IR 0.472 0.120 0.407NL 0.439 0.225 0.336N 0.359 0.218 0.424S 0.272 0.267 0.461CZ 0.487 0.197 0.315SL 0.584 0.177 0.238PL 0.664 0.115 0.220BG 0.682 0.178 0.140RU 0.717 0.130 0.153NZ 0.498 0.182 0.320CD 0.306 0.183 0.511RP 0.564 0.144 0.292IL 0.434 0.193 0.373J 0.373 0.238 0.389E 0.516 0.137 0.346ave 0.514 0.171 0.316 Stretched ternary coordinates: chi2 distance
B: A pre-school child is likely to suffer if his or her mother works Animation
achieved bysaving 101
frames in R of the ternary
diagram as themetric
smoothlychanges from
equal weightedEuclidean
distance to differentially
weighted chi-square
distance, thenframes savedinto a GIF file
Question B: A pre-school child is likely to suffer if his or her mother works
From Euclidean to chi2 distance(regular to irregular simplex)
B-agree B-M B-disagree B-XAU 0.497 0.140 0.354 0.009DW 0.684 0.104 0.175 0.037DE 0.325 0.150 0.490 0.036GB 0.377 0.176 0.408 0.040NI 0.382 0.151 0.431 0.036US 0.406 0.121 0.452 0.021A 0.720 0.088 0.174 0.018H 0.727 0.146 0.113 0.014I 0.671 0.138 0.177 0.015IR 0.472 0.082 0.407 0.038NL 0.439 0.197 0.336 0.028N 0.359 0.179 0.424 0.039S 0.272 0.220 0.461 0.047CZ 0.487 0.178 0.315 0.020SL 0.584 0.144 0.238 0.033PL 0.664 0.068 0.220 0.047BG 0.682 0.067 0.140 0.110RU 0.717 0.086 0.153 0.044NZ 0.498 0.157 0.320 0.026CD 0.306 0.160 0.511 0.022RP 0.564 0.143 0.292 0.001IL 0.434 0.174 0.373 0.019J 0.373 0.199 0.389 0.039E 0.516 0.090 0.346 0.047ave 0.514 0.138 0.316 0.033
Question B: A pre-school child is likely to suffer if his or her mother works
From 3 to 4 categories…
Middle response (M) and missing
response (X)separated
Rotation in three
dimensions of the country
profiles withina tetrahedron,
starting and ending withmiddle (M)
and missing(X) categories lined up (i.e.,
the two-dimensional
map seenpreviously.
Chi2 distanceevens out thecontributions
by thecategories
Question B: A pre-school child is likely to suffer if his or her mother works
irregular simplex in higher dimensions
principal axes of CA
E
JILRP
CD NZRU
BG
PL
SLCZ
S N
NL
IR
I HAUS
NIGBDEDW
AU
B-disagree
B-M
B-agree
-1.5
-1
-0.5
0
0.5
1
1.5
2
-1.5 -1 -0.5 0 0.5 1
0.0875 (80.4 %)
0.0142 (13.0 %)
B-X
Two-dimensionalCA map, with“asymmetricscaling”, i.e.
• rows (countries) in principal coordinates as theprojections of theprofiles
• columns(responsecategories) in standardcoordinates as theprojections of thecorners of thesimplex
93.4% inertiaexplained
1994: response proportions, 24 x 66 table
e.g., Spain14.1% strongly agree to Qu.A
(average across countries: 24.9%)7.8% strongly agree to Qu.K
(average across countries: 13.6%)
Question A“A working mother can establish just as warm and secure a relationship with her
children as a mother who does not work”
A1 A2 AM A4 A5 AXAU 0.183 0.349 0.092 0.282 0.085 0.009DW 0.369 0.356 0.040 0.158 0.040 0.037DE 0.627 0.275 0.017 0.050 0.008 0.022GB 0.176 0.443 0.120 0.188 0.043 0.030NI 0.155 0.468 0.085 0.201 0.054 0.037US 0.289 0.409 0.050 0.189 0.048 0.015A 0.524 0.230 0.033 0.150 0.041 0.021H 0.327 0.197 0.194 0.161 0.107 0.013I 0.215 0.409 0.100 0.179 0.094 0.003IR 0.183 0.428 0.057 0.215 0.100 0.017NL 0.208 0.490 0.098 0.155 0.026 0.023N 0.118 0.415 0.129 0.254 0.052 0.032S 0.222 0.432 0.131 0.152 0.034 0.030CZ 0.212 0.261 0.096 0.263 0.157 0.012SL 0.169 0.430 0.077 0.268 0.038 0.018PL 0.173 0.319 0.065 0.327 0.074 0.041BG 0.293 0.218 0.071 0.139 0.207 0.072RU 0.250 0.406 0.069 0.185 0.039 0.050NZ 0.134 0.402 0.093 0.285 0.059 0.028CD 0.308 0.412 0.069 0.160 0.035 0.015RP 0.054 0.568 0.145 0.213 0.018 0.001IL 0.232 0.429 0.092 0.166 0.064 0.016J 0.515 0.158 0.147 0.073 0.077 0.030E 0.141 0.406 0.036 0.337 0.053 0.027
K1 K2 KM K4 K5 KX0.127 0.608 0.138 0.112 0.013 0.0030.117 0.488 0.148 0.142 0.028 0.0770.099 0.427 0.180 0.180 0.042 0.0720.058 0.544 0.180 0.172 0.021 0.0250.068 0.495 0.181 0.212 0.005 0.0400.084 0.480 0.198 0.168 0.028 0.0410.259 0.418 0.126 0.127 0.039 0.0320.384 0.349 0.171 0.055 0.025 0.0170.117 0.542 0.154 0.145 0.028 0.0140.135 0.596 0.071 0.132 0.033 0.0320.049 0.531 0.217 0.155 0.012 0.0360.090 0.570 0.179 0.107 0.012 0.0430.067 0.352 0.286 0.189 0.041 0.0650.196 0.386 0.220 0.140 0.040 0.0190.098 0.523 0.181 0.130 0.019 0.0480.098 0.519 0.122 0.147 0.023 0.0910.477 0.253 0.083 0.045 0.036 0.1050.111 0.295 0.243 0.224 0.050 0.0770.107 0.606 0.138 0.119 0.011 0.0180.099 0.476 0.210 0.163 0.031 0.0200.046 0.468 0.213 0.251 0.021 0.0020.200 0.494 0.148 0.103 0.023 0.0320.245 0.264 0.138 0.081 0.235 0.0370.078 0.510 0.088 0.226 0.021 0.076
······
Question K“Family life often suffers because men concentrate too much on their work”······
“Symmetric” CA map of response proportions1 strongly
agree2 agreeM middle4 disagree5 strongly
disagreeX missing
23-dimensional
Total inertia= 0.2453
60.9% inertiaexplained in 2-d CA map
KX
K5
K4
KM
K2
K1
JX
J5
J4
JM
J2
J1IX
I5
I4IM
I2
I1
HX
H5
H4
HM
H2
H1GX
G5
G4
GM
G2
G1
FX
F5
F4
FM
F2
F1EX
E5
E4
EM
E2
E1DX
D5
D4 DM
D2
D1CX
C5
C4 CM
C2
C1
CX
B5
B4
BM
B2B1
AX
A5
A4
AM
A2
A1
E
J
IL
RP
CD
NZ
RU
BG
PL
SL
CZ
SNNL
IR
I
H
AUS
NI
GB
DE
DWAU
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
0.1021 (41.6%)
0.0472 (19.3%)
extreme responses
moderate& middle
responses
traditional
liberal
traditional
liberal
Qu.I“Man’s job is to earnmoney; woman’s jobis to lookafter home and family”
Decomposition of inertia across categories
Inertia contributions
12M45X
Missings (X)
Middles (M)
Category Inertia1 0.09212 0.0400M 0.01974 0.03945 0.0418X 0.0123
total 0.2453
middles & missings account for 0.03200 of the total inertia of 0.2453, i.e. , 13.0%; “extreme” responses account for 0.1339, i.e. 54.6%.
MethodsWe use correspondence analysis (CA) and several of its variants to visualize and interpret data at a country level as well as respondent and demographic-subgroup level.
1. Simple CA provides maps of aggregated or count data, for example proportions of question responses for the set of countries.
2. Subset correspondence analysis permits focusing on a particular set of response categories so that we can eliminate missing responses, for example, or restrict our attention to specific categories such as the middle responses (Greenacre & Pardo, SMR, 2006)
3. Multiple correspondence analysis (MCA) and its refinement joint correspondence analysis (JCA) provide maps of individual-level data, concentrating on the two-way relationships between the questions. In this respect MCA functions like an exploratory factor analysis for categorical data (on nominal scales). Usually we are not interested in individual case points but in mean points for the demographic and other external variables (e.g., age, level of interest…)
4. Canonical correspondence analysis (CCA) focuses on (or partials out) external variables – thus we can explore variation in the responses that is focused on interest, for example, or eliminate aquiescence effects.
CA of proportions of response categories
KX
K5
K4
KM
K2
K1
JX
J5
J4
JM
J2
J1IX
I5
I4IM
I2
I1
HX
H5
H4
HM
H2
H1GX
G5
G4
GM
G2
G1
FX
F5
F4
FM
F2
F1EX
E5
E4
EM
E2
E1DX
D5
D4 DM
D2
D1CX
C5
C4 CM
C2
C1
CX
B5
B4
BM
B2B1
AX
A5
A4
AM
A2
A1
E
J
IL
RP
CD
NZ
RU
BG
PL
SL
CZ
SNNL
IR
I
H
AUS
NI
GB
DE
DWAU
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
0.1021 (41.6%)
0.0472 (19.3%)
extreme responses
moderate & middle
responses
missings
middles
Animation achieved by changing weightThe animations which link methods are achieved by either reducing the mass of certain points or transferring mass between points.
For example, to show the missings and all moderate responses in this example) we multiply these dummy variables by where starts at 1 (i.e., the regular MCA) and decreases in small steps of 0.01 until 0 (i.e., the subset MCA). At each step a hybrid of a regular MCA and a subset MCA is performed, maintaining the margins of the table constant.
= 1, 0.99, 0.98, 0.97, …………, 0.04, 0.03, 0.02, 0.01, 0
…………
CA tosubset
CAData:
WOMEN WORKING,
stackedfrequencies,
N=33590
Method:Contributionof categoriesnot in subset reduced by
factor , from1 (CA) to
limiting case of 0 (subset CA), always
usingoriginal
masses forcentring andweighting)
Total inertia
Principal inertiasin 2-d soln.
CA tosubset
CAData:
WOMEN WORKING,
stackedfrequencies,
N=33590
Method:Contributionof categoriesnot in subset reduced by
factor , from1 (CA) to
limiting case of 0 (subset CA), always
usingoriginal
masses forcentring andweighting)
Total inertia
Principal inertiasin 2-d soln.
Subset CA of proportions of middles and missings
KMJM
IM
HM
GM
FMEMDM
CMBMAM
KXJX IX
HXGX FXEX DX CXBX
AX
E
J
IL
RP
CDNZ
RU
BGPL
SLCZ
S
NNL
IRI
H
A
US NIGB
DE
DW
AU
-0.2
0
0.2
0.4
-0.6 -0.4 -0.2 0 0.2 0.4 0.6
0.0144 (45.0%)
0.0065 (20.3%)
Of the (small – 13.0% – part of the) inertia of the M- and X-percentages, which is contained in a 22-dimensional space, 65.3% is explained in this map. The fact that all the M’s are together and all the X’s together and separate, does not reflect category associations at an individual level. For example, the proximity of Spain (E) and Russia (RU) means that their percentages of M- and X-responses are similar (less M’s than average, more X’s than average)
MethodsWe use correspondence analysis (CA) and several of its variants to visualize and interpret data at a country level as well as respondent and demographic-subgroup level.
1. Simple CA provides maps of aggregated or count data, for example proportions of question responses for the set of countries.
2. Subset correspondence analysis permits focusing on a particular set of response categories so that we can eliminate missing responses, for example, or restrict our attention to specific categories such as the middle responses.
3. Multiple correspondence analysis (MCA) and its refinement joint correspondence analysis (JCA) provide maps of individual-level data, concentrating on the two-way relationships between the questions. In this respect MCA functions like an exploratory factor analysis for categorical data (on nominal scales). Usually we are not interested in individual case points but in mean points for the demographic and other external variables (e.g., age, level of interest…)
4. Canonical correspondence analysis (CCA) focuses on (or partials out) external variables – thus we can explore variation in the responses that is focused on interest, for example, or eliminate aquiescence effects.
Looking at respondent-level dataTo investigate response behaviour at individual respondent level we pass from CA to multiple correspondence analysis (MCA). The classic definition of MCA is the CA of the data coded in an indicator matrix, i.e., as dummy variables, one variable for each response category
Dimensionality is 66 less the 11 linear restrictions on the columns, i.e. 55
A B C D E F G H I J K1 3 3 5 5 1 1 1 5 5 4
9 2 2 2 2 4 4 4 2 2 2...
Original responses (Q = 11)
A1 A2 AM A4 A5 AX B1 B2 BM ...1 0 0 0 0 0 0 0 1 ...
0 0 0 0 0 1 0 1 0 ......
Dummy variables (J = 66)
MCA of all response categories1 strongly
agree2 agreeM middle4 disagree5 strongly
disagreeX missing
Total inertia= 0.6626
34.5% inertiaexplained
Adjusted value:
59.9% inertiaexplained
KX
K5
K4KMK2
K1
JX
J5
J4 JM
J2
J1
IX
I5
I4
IMI2
I1
HX
H5
H4
HM
H2
H1
GX
G5
G4GM
G2
G1
FX
F5
F4FM
F2
F1
EX
E5
E4 EM
E2
E1
DX
D5
D4 DM
D2
D1
CX
C5
C4 CM
C2
C1
BX
B5
B4BM
B2
B1
AX
A5
A4AMA2
A1
-2.5
-2
-1.5
-1
-0.5
0
0.5
-0.5 0 0.5 1
0.1213 (18.3%)
0.1071 (16.2%)
allmissings
allextremes
allmoderates& middles
Decomposition of inertia across categories in MCA
middles & missings account for 0.2366 of the total inertia of 0.6626, i.e. , 35.7% ; missings account for largest part of inertia
Inertia contributions
12M45X
Missings (X)
Middles (M)
Category Inertia1 0.12972 0.0827M 0.08814 0.09285 0.1207X 0.1485
total 0.6626
MCA tosubset MCAData:
WOMEN WORKING, Burt matrix,
N=33590
Method:Contributionof categoriesnot in subset reduced by
factor , from1 (CA) to
limiting case of 0 (subset CA), always
usingoriginal
masses forcentring andweighting)
Total inertia
Principal inertia in 2-d soln.
middles
Data: WOMEN WORKING,
stackedfrequencies,
N=33590
Using capackage in R.Also showingthe country
points in “symmetric”
scaling so theyare more
spread out in the
visualizationfor ease of
interpretation
Reference: Nenadić, O. & Greenacre, M.J. (2007). Correspondence analysis in R, with two- and three-dimensional graphics: the ca package. Journal of StatisticalSoftware 20(3). URL http://www.jstatsoft.org/v20/i03/.
Rotating the subset MCA solution(excluding missings)
What does a “perfect unidimensional model”look like in an MCA?
In first two dimensions:
Parabola: the “horseshoe”/ “arch”/
“Guttman” effect
In dimensions 1 and 3:
Cubic
In dimensions 1 and 4:
Quartic, etc....
Data that follow a perfect traditional-to-liberal scale were generated, reversing the scales of oppositely worded statements, and randomlyadding 3% missing responses. Here are different MCA maps of the data:
-1.0 -0.5 0.0 0.5 1.0 1.5
-1.0
-0.5
0.0
0.5
1.0
A1
A2
AM
A4
A5
A9
B1
B2
BMB4
B5
B9
C1
C2
CM
C4
C5
C9
D1
D2
DM
D4
D5
D9
E1
E2
EM
E4
E5
E9
F1
F2
FM
F4
F5
F9
G1
G2
GM
G4
G5
G9
H1
H2
HM
H4
H5
H9
I1
I2
IM I4
I5
I9
J1
J2
JM
J4
J5
J9
K1
K2
KM
K4
K5
K9
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
B1
B2
BM
B4
B5
BX
F1
F2
FM
F4
F5
FX
C1
C2
CM
C4
C5
CX
D1
D2
DM
D4
D5
DX
E1
E2
EM
E4
E5
EX
I1
I2
IM
I4
I5
IX
A1
A2
AM
A4
A5
AX
K1
K2
KM
K4
K5
KX
G1
G2
GM
G4
G5
GX
J1
J2
JM
J4
J5
JX
H1
H2
HM
H4
H5
HX
-2 -1 0 1 2
-3-2
-10
1
B1
B2
BM
B4
B5
BX
F1
F2
FM
F4
F5
FX
C1
C2
CM
C4C5 CX
D1
D2
DM
D4
D5
DX
E1
E2
EM
E4
E5
EX
I1
I2
IM
I4
I5
IX
A1
A2
AM
A4
A5
AX
K1
K2
KM
K4
K5
KX
G1
G2
GM
G4 G5GX
J1
J2
JM
J4
J5
JX
H1
H2
HM
H4
H5HX
What does a “perfect unidimensional model”look like in an MCA?
Looking by country: Spanish and WestGerman data
• Previous MCA maps based on all 33590 respondents from 24 countries.
• We have already seen that there are inter-country differences in theiroverall levels of middle and missing responses.
• Our aim is to investigate how respondents use the middle responsesand if there any associations with demographic variables. To avoidinter-country differences in our results we concentrate (separately) on two countries: Spain (N = 2494) and West Germany (N = 2324) –as we shall see, they present contrasting results.
• We will also introduce the following demographic variables into ourstudy:
Gender (2 categories)
Age (6 categories)
Marital status (5 categories)
Education (7 categories)
(Education not available in Spanish 1994 sample)
MCA of Spanish “women working” data (N = 2494)
-2 0 2 4
-10
12
34
Dim 1
Dim
2.
.
.
.
.
..
..
..
.
.
..
.
.
.
...
..
.
..
.
.
.
.
.
.
.
.
. ... .
.
.
.
.
.
.
.
.
.
.
.
.
.
..
..
.
. .
.
..
.
.
.
.
.
..
..
.
.
.
.
.
.
.
.
..
.. .
.
.
.
.....
.
.
.
.
.
.
.
. .. ..
......
.
..
.
.
. .
. ..
.
.
....
.
..
.
. .. .
...
.
.
. ....
.
....
.
.. ...
.
.
.
.
.. . .
.
....
..
.
.....
.
.. ...
..
.. .
.. .
..
.
...
.
.
..
.
..
...
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
..
....
.
.
. ..
.
.
..
.....
.. ..
.
.
.
.
.
.
.
.
..
.
...
. ..
... .
.
.
..
.
...
...
.
...... ..
..
.
.
.
...
.
.
.. ...
.
. .
..
...
.
.
..
...
. .
...
.
.
.
..
.
.
..
.
.
.
.
.
...
..
..
..
.
.
.
.. ..
.. .
..
.
..
..
.
.
.
.
. ..
.. .
.
..
..
.
..
.
.
.
.
....
.... .
.
. .
..
.
.
..
.
.
.
..
.
...
.
.
.
.
..
.
. .
.
.
.
.
.
.
.
.
..
.
.
..
.
. ..
..
..
..
.
..
.
....
.
..
..
. .
..
..
. .
.
.
.
...
.
..
.
..
.....
..
.
. .
.
.
.
.
.
..
.
... .
....
..
..
. .
.. .
.
.
..
.
.
. ..
..
.
.
. ..
.
..
.. ....
.
.... .
..
.
.
.
..
.
...
..
.
.
.
.
..
.
.
.. ... .. .
.
.
.
.
.
.
.
.
..
.
.
...
..
...
.
..
.
.
.
.
..
..
.
..
.
..
.
.
. .. ...
.
.
.. .. .
.
.
..
...
.
..
.
.
..
.
.
... ..
.
.
..
....
.
.. ..
.
.
.
. ..
.
.
.
...
..
. .
.
.
.
. .
.
.
..
..
.....
.
.
.
. ... ..
..
...
.
.
..
.
.....
..
.
..
.
. .
.....
.
.. .
.
..
..
..
.
.
...
. .
.
...
.
.
. .. .
....
..
...
..
.
...
..
.
.
.
.
.
.
..
.. .
.
..
...
.
... .
.
.
.
. ...
..
.
.
.
.
.
.
.
.
..
..
.
..
..
..
..
..
..
..
.
... . .
.
.
.
.....
.
.
.
.
.
.
.
...
.
.
...
..
.
..
.
...
.
.
.. . .
..
.
.
.
.
..
.
.
.
..
.
.
.
..
.
. .
.. .
.
. .
..
.
.
.
..
.
.
..
.
.
..
.
..
.
..
.
.
.
.
.
..
.
.
...
.
.. .
.
.
.
.
.
.... .
.
....
...
...
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.
.
.
.. ..
.
. .
.
.
..
.
..
.
. ....
..
.
. ..
..
..
.
.
.
. .
. .
.
.
..
. ..
.
..
. .
.
.... .
.
..
..
.
.
.
..
.
... .....
.
.
.
.
...
.
..
.
..
.
.. ..
...
.
...
.
.
..
..
.. ...
.
.
.
.
..
.
.
..
.
. ...
.
... .
.
.
.
..
...
.
. .
.
.
.
.
.
..
..
.
.
..
.
...
.
.
.. .... .
.
. .
..
.... .
..
.
.
.
.
..
..
...
.
...........
.
..
.
....
.
.
...
. .....
..
.
.
..
.
.
. .
.
...
.
. .
.
... .
.
.
.
..
.
.
..
. .. .
.
...
. ..
.
....
.
..
.
.
.
.
.
.
.
..
.
.
.
.
..
. ...
.
.
. ....
.
......
.
.
...
..
..
.
... .
..
...
..
. ..
.
.
.. .... .
.
.. .
.
..
.
..
.
. .
.
....
..
..... ...
.
.
.
...
.
..
..
. ...
..
..
.
.
...
..
. .
... .
.
..
. .
.
.
...
..
...
. ...
.
.
.
. .
....
..
.
.
.
.
..
..
.
.
.
.
.. .
.
....
..
..
..
.
.
.
.
..
. .
.
.
.
.
.. ..
..
..
.
.
..
.
.. .. ...
..
.
..
..
.
.
.
.
..
.
...
..
.
.
.
....
.
..
.
.
.. ..
.
..
. ...
..
.
.
..
.
.
.
.
.
.
.
.
..
.
.
.
.
.
..
.. ..
..
......
. .
.
..
.
.
.
..
.
. ... .. .
.
..
.
.
.
.
. .
.
.
.
..
.
..
.
.
.
..
...
.
.
..
.
.
..
...
.
.
...
...
.
. .
.
. ...
... ....
... ...
.
.
.
.
..
.
..
..
.
...
..
.
..
.
.
...
.
.
.
..
.
.
.
....
.
.
.
.
.
. .
.
.....
.
. .
.
.
. ..
. ..
..
.. .
.
....
.
...
.
...
.
.
.
..
.
.
..
..
.
..
.
.
..
..
.
...
.
...
. ..
...
.
.
.
.
.. ..
..
.
.
.
..
.
.
.
.
...
.
. .
.
..
.
.
.
.
.
..
.
. .
.
.. .
.
.
.
. ..
..
.
. ....
.
.
..
.
...
..
..
.
. ...
.
...
...
.
..
.
.
.
..
.
.
.
.
..
.
. .
.
..
.
.
.
..
.
.
......
.
.
.
.
.
...
.. ..
.
..
.
.
..
.
.
.
.
.
..
..
.
. .
.
..
.
..
.
..
.
.
.
...
.
....
.
.
.
.
.
...
.
...
.
...
....
...
.
..
.
.
.
..
.
.
.
.
.
..
.
.
.
. ..
..
.
. ..
..
.
. .
.
.
.
.
.
.
.
.
.
.
.
....
...... ....
.
..
.
..
.
. .
.
.
.
....
.
..
.
.
... ..
. ..
.
..
.
..
.
.
...
. ...
.
......
.
..
..
.
..
. ..
..
.
....
.
.
.
..
..
.
.
... ....
...
.
..
.
...... .
..
.
.. .
.
.
..
..
.
..
. .
.
.. .
.
.
.
.
. .
.
..
...
.
..
...
..
...
.
.
.
.
..
. .
.
.
.
..
..
..
.
. .
.
...
..... .
.
..
.
. .
.
.
..
....... ...
..
.
..
. ...
...
..
.
.....
..
.
.
...
...
....
..
.
.
..
..
.
.
.
.. . .
... .
.
.
.
.
.
.
.
.
.
.
.
.. ..
.
.
.
..
.
.
..
.
..
.
.
....
.
.
.
.
..
.
.. .
.
....
. .
..
..
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
....
.
.
.
. ...
.
.
..
.
.
.
.
.
..
.. .
..
.
.
.. ..
.
.
.
.
..
.
..
.
...
. ...
.
.. .
.
.
.
.
.
....
.
..
.
..
.
... .
.
.
..
..
.
.
.
..
...
.
.
......
.
.
.
..
.
.
.
.
.... .
.
.
.
.
.
...
.
.
..
...
..
..
.
.
.
.
.
.
... .
.
..
.
..
...
.
...
.
... .
..
.. .
.
.
..
.
...
.
. .
.
.
..... .
.
.
.
..
... .
..
.
A1
A2AMA4
A5
AX
B1
B2BMB4
B5BX
C1
C2CM
C4
C5CX
D1
D2DMD4
D5DX
E1
E2EME4
E5
EX
F1
F2FMF4
F5FX
G1
G2
GMG4
G5
GX
H1
H2
HMH4
H5
HX
I1
I2IMI4
I5
IX
J1
J2JMJ4
J5
JX
K1
K2KM
K4
K5KX
Missing categories
Extreme categories (stronglyagree and strongly disagree)
Moderate and middlecategories
Subset MCA of Spanish “women
working” data(N = 2494)
missing responsesexcluded from subset
middles amongst themoderate responses in
this two-dimensionalview
-4 -3 -2 -1 0 1
-20
24
.
.. .
......
..
.....
. ...
.
..
.
..
.
.
..
.
..
..
.
..
.
.. .. .
.
.. .
.
...
.
..
.
..
.
. .
.
....
.
...
.
..
..
..
.
.
.
..
.. .
.
.
.
.
.
..
.
.
...
.
.
.
.
..
. ...
..
. ...
.
. .. ..
.
.
.
.. .
.
.
.
. .
...
.
...
.
....
..
.. ..
.
....
..
.
..
.
..
.
.
. .. .
.
.
...
.. .
.
...
.
...
. ..
.
..
..
.
..
.
.
..
..
.
.
.....
.
.
.
..
..
.. ..
.. .
.
.
... .
.
..
.
.
..
.
.
.
.
.....
..
..
. .. ..
..
..
..
.
.. .
.
.. .
...
.
....
... ...
.
.... ...
.
. .. ..
...
.
.
..
.
..
....
.
.
.
.
.
..
.
...
.
.
.
..
.
.
.
.
...
.
..
.
..
.
..
.
.. .
.
..
.
.
.
.
....
.
.
..
.
..
.
.
. .
.
. ..
.
. .
.
.
..
.. ..
.
.
.
.
...
.. ...
...
.
. .. ... .. .
.
..
.
.
..
.
..
.
.
.
.
.
.. .
..
.
...
..
.
.
..
.
..
.
.
.
.
..
.
.
.
.
.
..
...
.
.
...
...
.
..
.
. .. .
.
..
.. .
..
.
..
.
.
.
.
.
..
.
..
.
. ...
.
..
.
.
.
.. ..
.
.
.....
.
.
. .. .
.
.
.
.
.
..
...
.
.
... ..
..
....
. .
.
. ..
.
.
.. ...
.
.
.
. .
.
. ...
... ..
.
. ..
.
.
. .
..
.
.. .
.
.
.
.
.
.
....
.
.
.
.
.
.
.. ..
..
.... .
. ...
....
..
..
..
.
.
.
...
.
..
..
..
.
.
.
.
.
.
.
..
.
.
.. .
.
.
.
..
...
.
..
.
.. ....
..
..
...
.
..
.
.
.
. ..
..
. .
.
...
..
.
. .
.
.
..
.
..
...
.
.....
.
..
.
.
. .
...
. ...
.
..
. ..
..
.
.
.
.
.. .. .
. ...
.
.
.
.
. ..
.
..
...
..
. .
.
..
. .
.
.
.
...
.
. . ...
.
.....
.
..
. .
.. . .
.
.
.
..
.
..
.
.
.
..
.
.. .
..
..
.
.
.
.
..
..
. ....
.. .
.
.
.
.
.. .
...
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
. ..
..
.
.
..
. .
.
..
.
.
. .
..
.. .
.
.
.
.
. ...
. ..
.
.
..
...
.
..
.
. ..
.
.. .
.
.
...
... .
.
.
.
..
.
.
.
.
.
.
.
..
..
..
.
..
..
.
.
...
...
.. . .
..
.
.
.
..
. . .
.
.
.
.
.
.
..
. ..
.
... . ..
.
.
.
.
.
.
. ..
.
..
.
..
..
.
.
...
..
.
.
.
.
..
.
.
.
..
.... .
.
......
.. ...
.
.
.
.
.
..
.
.
....
.
.
.
.. .
.
.
..
.
.
..
.
.
.
.
.
..
.
.
.
. .
. .
.
...
... .
.. . ..
..
. ..
.
.....
...
.
..
.. .
.
...
..
.
.
.
.
. ..
..
.
..
.
..
..
..
.
.
. .
...
..
...
.
.
..
.
.
.
.. .
. ...
.
...
.
.
.
. ........
.
. ..
.
.
.
..
..
.
.
...
.
..
..
.
..
.
.
. .
.
. ..
.
.
.
.
.
..
.
.
. ..
.
.
.
.
.....
.....
.
.
..
....
. ..
.
.
.
.
.
.
...
..
.
.
..
.
..
.
.
....
..
..
.
..
..
.
.
.
.. .
...
.
.
.
.
..
....
.
.
.
..
..
. . .
.
..
..
..
.
.
.
.
.
..
..
.
....
. ..
.
.
.. .
..
..
. .
..
..
.
..
..
....
. ..
.
...
. ..
.
..
.
.
.
..
.. .
.. .
..
.
.
.
.
. .
.
.. .
.
. ..
.
.
.
. ..
... .
. ...
.
...
.
.
........
...
...
.
. ..
.
..
.
.
.
.
.
...
..
..
.
. .
.
...
...
.. .
.
.
...
.
.. . ..
.
....
.
.
... .... .
. ..
..
..
.
..
. .
..
..
....
.
..
. .
.
.
.
. ...
.
.
.
.
....
..
...
.
. ...
...
..
.
...
.
.
.
.
.
.
...
.
..
.
.
.
..
.
.
...
. .
... ..
.
...
.
.
..
.
..
.
...
.
. ..
..
..
.
..
. .
..
.. .
..
. ...
...
..
..
.
.
..
.
.
.
...
..
...
.
...
.
.....
.
... .
.
.
...
.
.
.
.
.
..
......
.
..
..
.
.
.
.
.
.
.
..
.
.
.
.
...
.
.. .
.
.
.
..
..
.
.
.
.
.
..
..
..
.
...
.
.
..
.....
.
..
.
.
.
..
.
.
.
...
.
.
..
.
.
.
.
.
.
..
.
. ..
.
.
..
..
.
.
.
..
.
. .
.
.
.
.
.
.
.
.
.
..
.
.
.
.
....
.
.
...
.
.
.. .
.
.. .
.
...
... ..
..
..
.
.. ..
.
.
. .
.
..
. .
..
..
..
.
.
.
.
... .
.
.
. .
.
.
.
.....
.
. .. .
.
.
.
.
.. ..
.
.
.
..
.
..
.
..
.
.
..
..
.
..
...
...
..
.
..
..
.
..
..
..
..
.....
.
..
.
.
.
..
.
..
..
..
.
.
....
.. .
..
.
. ..
.
.
.
.
.
.
.
..
.
..
.
.
.
.
..
.. .
.
.
....
.....
.
.
.. ..
.
.
...
..
.
.
.
. ..
.
.
..
.. .
..
.
.
.
..
.
.
.
..
.
.
.
.
.
.
. ....
. .
.
.
..
.
..
.
.
.
..
.
.....
.
.
.... .
..
..
.
... ...
.
..
.
..
.. ....
. ...... ..
.
.
..
.
.
.
.
.
..
.
.
.
. .
.
.
..
.. .
.
.
....
.
.
...
.
..
.
.
.
..
.
.
.
.
.
.. .
.
..
.
.
.
.. .
.
.
.
...
..
.
.
.
.
.
. .
..
..
..
.
.
..
..
.
... ..
.
.
.
.
.
.. . .
....
. .
.
...
...
. .
. ..
.
.
.
... .
.
.
. ..
..... ..
..
.
.
...
. .
.
. ...
.
.
.
...
.
.
.
.
..
.
.
.
...
.
.
.
.
..
.
.. .
.
....
.
.
..
....
.
.
.
. .. .
.
.
.
.
..
.
.
.
.
.
.
.
..
..
...
.
.
. .
..
.
.
.
..
..
.
..
. .
..
.
.
...
..
....
..
...
...
.
.
..
..
.
..
...
.
. .
..
..
.
....
.
.
.
.
..
.
...
.
...
.
...
..
..
.
..
..
.
.
.
... ..
.. .
.
.
. .
..
.
.
..
.. ..
. ..
.
.
...
....
.
.
. .
.
. ..
.
..
..
. .
.
..
.
.
.
..
.
.
.
...
.
.. .
.
..
.
....
.
. .
.
.
.
.
. ..
.
.
.
.
A1
A2
AM
A4
A5B1
B2
BM
B4
B5
C1
C2
CM
C4
C5
D1
D2
DM
D4
D5
E1
E2
EME4
E5
F1
F2FM
F4
F5
G1G2GM
G4
G5
H1
H2
HM
H4
H5
I1
I2
IM
I4
I5
J1
J2
JMJ4
J5
K1
K2
KM
K4
K5
AXBXCXDXEXFXGXHXIXJXKX
traditional
liberal
middles
moderate& middle
strong Rotating thesubset MCA
solution (subset excludingmissings)Data: WOMEN
WORKING, Spanish data, subset MCA,
N=2494
Method:Rotation about
2nd axis, recorded using
plot3d function in capackage for R
KX
K5K4KM
K2
K1
JX
J5J4
JM J2
J1
IX
I5 I4
IMI2
I1
HX
H5
H4HM
H2
H1
GX
G5
G4GM
G2
G1
FX
F5
F4FM
F2
F1
EX
E5E4
EME2
E1
DX
D5D4
DM D2
D1
CX
C5C4 CM
C2
C1
BX
B5B4BM
B2
B1
AX
A5
A4AM
A2
A1
-3
-2
-1
0
1
2
3
4
5
6
-2 -1 0 1 2 3 4
0.318 (33.1%)
0.288 (24.9%)
(B5) Stronglydisagree that pre-school child willsuffer becausemother works
LIBERAL
MCA of W.GERMANDATA
N=2324
TRADITIONAL
Missingcategories
(H5) Stronglydisagree that
man & womenshould both
contribute to household
income
-3
-2
-1
0
1
2
3
4
5
6
-2 -1 0 1 2 3 4
0.318 (33.1%)
0.288 (24.9%)
MCA tosubset MCA
Data: WOMEN WORKING,
W.German sample, N=2324
Method:Contribution of
categories not in subset reduced by
factor , from 1 (CA) to limiting case of 0 (subset CA), always
using original masses for centring
and weighting)
Total inertia
Principal inertia in 2-d soln.
SubsetMCA of W German“womenworking”data(N = 2324)missingresponsesexcluded fromsubset
K5
K4
KMK2
K1
J5
J4 JM
J2
J1
I5
I4
IMI2
I1
H5
H4HMH2
H1 G5
G4GMG2
G1
F5
F4FMF2
F1
E5
E4EM
E2
E1
D5
D4DM
D2
D1
C5
C4
CM
C2
C1
B5
B4
BM
B2
B1
A5
A4AM
A2
A1
-2
-1
0
1
2
3
4
5
-2 -1 0 1 2 3 4
traditional
liberal
“strongly disagree” that family lifeoften suffers because menconcentrate too much on their work
Data: WOMEN WORKING,
W.German data, subset MCA,
N=2324
Method:Rotation about
2nd axis, recorded usingrgl.snapshotin rgl packageand plot3d.cain ca package
for R
Rotating thesubset MCA solution(excludingmissings)
Data: WOMEN WORKING,
W.German data, subset MCA,
N=2324
Method:Rotation about
1st axis, recorded usingrgl.snapshotin rgl packageand plot3d.cain ca package
for R
Rotating thesubset MCA solution(excludingmissings)
Response sets in Spanish sampleLooking more closely at the individual Spanish data, we discover the following response sets (figures for W.Germany for comparison):
(WG)“strongly agree” to all questions – 4 respondents (2)“agree” to all questions – 20 respondents (6)“neither/nor” to all questions – 6 respondents (0)“disagree” to all questions – 2 respondents (0)“strongly disagree” to all questions – 0 (0)missing values for all questions – 18 respondents (6)
50 out of the 2494 respondents, i.e. 2% (½%)
The “middle” and “missing” response sets accentuate the association within these categoriesThe “strongly agree” and “agree” response sets (categories 1 and 2) to questions which have reverse orientations will tend to bring the opposite poles closer than they would otherwise have been.We now remove all these response sets, all the features previously seen are still there, just their ordering on the principal axes changes: e.g., the group of “missings” is now on the 3rd dimension and the “middles” on the 5th.
MCA ofSpanish data
withoutresponse
sets
Data:WOMEN
WORKING, Spanishsample, N=2444
(without 50 response
sets)
Dimensions 1 and 2
KX
K5
K4
KM
K2
K1
JX
J5
J4
JM
J2
J1
IX
I5
I4
IM
I2
I1HX
H5
H4
HM
H2
H1
GXG5
G4
GMG2
G1
FX
F5
F4
FMF2
F1
EX
E5
E4EM
E2
E1
DX
D5
D4
DM
D2
D1
CX
C5
C4
CM
C2
C1
BX
B5
B4
BM
B2
B1
AX A5
A4
AM
A2
A1
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
-1 -0.5 0 0.5 1 1.5 2 2.5
0.102 (41.0%)
0.059 (23.8%)
MCA ofSpanish data
withoutresponse
sets
Data:WOMEN
WORKING, Spanishsample, N=2444
(without 50 response
sets)
Dimensions1, 3 and 5
Focusing the display on the number of middle responses
11 questions (A,B,C,D,E,F,G,H,I,J,K) from ISSP survey
Pass the weight smoothly from therespondents to the group centroids:
Xn
respon-dents
m0 m1 ... m6+
Z
A1 A2 AM A4 A5 B1 B2 BM B4 B5 ... K1 K2 KM K4 K5
0 0 1 0 0 0 1 0 0 0 ... 0 0 0 0 1 1 0 0 0 0 0 0 0 1 0 ... 0 1 0 0 0
0 1 ... 01 0 ... 0
Classifica-tion in terms of number of middles
XTZ
1/n1/n.......
1/n
r
0
0
1w )1(
)(1
n
g groups
# m
10
r1r2.rg
Count of middles
MCA MCA-Discriminant analysis (middle groups): Spain
MCA MCA-Discriminant
analysis(middlegroups):
W. Germany
MethodsWe use correspondence analysis (CA) and several of its variants to visualize and interpret data at a country level as well as respondent and demographic-subgroup level.
1. Simple CA provides maps of aggregated or count data, for example proportions of question responses for the set of countries.
2. Subset correspondence analysis permits focusing on a particular set of response categories so that we can eliminate missing responses, for example, or restrict our attention to specific categories such as the middle responses.
3. Multiple correspondence analysis (MCA) and its refinement joint correspondence analysis (JCA) provide maps of individual-level data, concentrating on the two-way relationships between the questions. In this respect MCA functions like an exploratory factor analysis for categorical data (on nominal scales). Usually we are not interested in individual case points but in mean points for the demographic and other external variables (e.g., age, level of interest…)
4. Canonical correspondence analysis (CCA) focuses on (or partials out) external variables – thus we can explore variation in the responses that is focused on interest, for example, or eliminate aquiescence effects.
Partialling out acquiescence effects
O’Muircheartaigh, Krosnick & Helic (2000)
‘We estimated a model in which all items were allowed to load on the same latent factor representing attitude toward science, plus a second latent factor intended to represent acquiescence. All items were constrained to load equally on this latter factor, an assumption required to identify the model. This is reasonable, because acquiescence is defined as a tendency to agree with any item regardless of its content, so it should account for the same amount of variance in responses to all the items. The acquiescence factor was constrained to be uncorrelated with the factor representing attitudes toward science, another assumption required in order to identify the model.’
Identifying or partialling out “acquiescence effects”11 questions (A,B,C,D,E,F,G,H,I,J,K) from
ISSP Family and Changing Role survey II (1994).
X
1 2 M 4 5 X
Z
A1 A2 AM A4 A5 AX B1 B2 BM B4 B5 BX ... K1 K2 KM K4 K5 KX
0 0 1 0 0 0 0 1 0 0 0 0 ... 0 0 0 0 1 01 0 0 0 0 0 0 0 0 0 1 0 ... 0 0 0 0 0 1
2 3 1 3 1 13 1 0 2 2 3
counts of numberof responses in each category
The counts in the matrix X are variables which quantify tendenciesto use the same response category. CCA can restrict the MCA solution to be linearly related to any subset of these externalvariables. This is a projection in the MCA space. We can alsopartial out acquiescence effects by looking in the space orthogonalto the restricted space (this is an alternative strategy to subset CA).
MCA
Data:WOMEN
WORKING, Spanishsample, N=2494
-3 -2 -1 0 1
-3-2
-10
1
CA
2A1
A2AMA4
A5
AX
B1
B2BMB4
B5
BX
C1
C2
CMC4
C5CX
D1
D2DMD4
D5
DX
E1
E2EM
E4
E5
EX
F1
F2FMF4
F5
FX
G1
G2
GMG4
G5
GX
H1
H2
HMH4
H5
HX
I1
I2
IM
I4
I5
IX
J1
J2JMJ4
J5
JX
K1
K2
KMK4
K5KX
missings
CCAcounts of
missings as external variable which
restricts thesolution
Data:WOMEN
WORKING, Spanishsample, N=2494
missings
0 1 2 3 4
-2-1
01
CCA1
CA
1
A1
A2AMA4
A5
AX
B1
B2
BMB4
B5
BX
C1
C2
CMC4
C5
CX
D1
D2
DMD4
D5
DX
E1
E2
EME4
E5
EX
F1
F2
FMF4
F5
FX
G1
G2
GM
G4
G5
GX
H1
H2
HMH4
H5
HX
I1
I2
IMI4
I5
IX
J1
J2
JMJ4
J5
JX
K1
K2
KM
K4
K5
KX
Partial CCAsolution
orthogonal toone that
restricts thesolution to be
linearlyrelated tomissingcounts
Data:WOMEN
WORKING, Spanishsample, N=2494
missings
-3 -2 -1 0 1
-2-1
01
CA1
CA
2
A1
A2
AM
A4
A5
AX
B1
B2
BM
B4
B5
BX
C1
C2
CM
C4
C5
CX
D1
D2
DM
D4
D5
DX
E1
E2
EME4
E5
EX
F1
F2FM
F4
F5
FX
G1G2
GM
G4
G5
GX
H1
H2
HM
H4
H5
HX
I1
I2
IM
I4
I5
IX
J1
J2
JMJ4
J5
JX
K1
K2
KM
K4
K5
KX
CCAcounts ofmissings
and middlesas external
variable which
restricts thesolution
Data:WOMEN
WORKING, Spanishsample, N=2494
missings
0 1 2 3 4
-2-1
01
CCA1
CC
A2
A1A2
AM
A4A5
AX
B1B2
BM
B4B5
BX
C1C2
CM
C4C5CX
D1D2
DM
D4D5
DX
E1E2
EM
E4E5EX
F1F2
FM
F4F5
FX
G1G2
GM
G4G5
GX
H1H2
HM
H4
H5
HXI1I2
IM
I4I5 IX
J1
J2
JM
J4J5
JXK1K2
KM
K4K5
KX
middles
Data:WOMEN
WORKING, Spanishsample, N=2494
CCA solution
partiallingout the
middles andmissings
andrestricted to
theextremes
andmoderates
Partial CCAsolution orthogonal to one that restricts the solution to be
linearly related to all acquiescence effectsData: WOMEN WORKING,
Spanish sample, N=2494
A1
A2
AM
A4
A5
AX
B1
B2
BMB4
B5
BX
C1
C2CM
C4
C5
CX
D1
D2DM
D4
D5
DXE1
E2EME4
E5EX
F1
F2
FM
F4
F5
FXG1
G2
GM
G4
G5
GXH1
H2HM
H4
H5
HX
I1
I2
IMI4
I5
IX
J1
J2
JMJ4
J5JXK1K2KM
K4
K5
KX
CCA of Spanish “women working” data:categories
-4 -2 0 2
-10
12
34
A1
A2
AM
A4
A5
AX
B1
B2
BMB4
B5
BX
C1
C2
CM
C4
C5
CX
D1
D2DM
D4
D5
DXE1
E2EME4
E5EX
F1
F2
FM
F4
F5
FX
G1
G2
GM
G4
G5
GXH1
H2
HM
H4
H5
HX
I1
I2
IM
I4
I5
IX
J1
J2
JMJ4
J5
JXK1K2KM
K4
K5
KX
conservative, extremeresponses
conservative
conservative, moderate responses
liberal, extremeresponses
liberal
liberal, moderate responses
all middles and missings at centre:
CCA of Spanish “women working”:cases
-8 -6 -4 -2 0 2 4
-20
24
6
+
+
+
+
+
+
+
+
+ +
+
++
++
+
+
+
+
+ ++
+
+++ +
+
++
+
+
+
+
+
++
+
++
+
+ +
++
+
+
+
+
+
++
++
+
+
++
+++++
+
+
+
CCA of W.German “women working”:categories
-4 -3 -2 -1 0 1 2
-10
12
34
A1
A2
AM
A4
A5
AXB1
B2
BMB4
B5
BXC1
C2
CM
C4
C5
CX
D1
D2 DM
D4
D5DX
E1
E2
EME4
E5
EX
F1
F2
FMF4
F5
FXG1
G2
GM
G4
G5
GX
H1
H2
HM
H4
H5
HX
I1
I2
IM
I4
I5IX
J1
J2
JM
J4
J5JX
K1K2
KMK4
K5
KXconservative
conservative, moderate responses
liberal, extremeresponses
liberal
liberal, moderate responses
conservative, extremeresponses
CCA of W.German “women working”:cases
-6 -4 -2 0 2 4
-20
24
6
Demographic categories
• Gender : g1 (male), g2 (female)
• Age 6 groups: a1 (up to 25), a2 (26-35), a3 (36-45)a4 (46-55), a5 (56-65), a6 (66 and over)
• Marital status 5 groups: m1 (married), m2 (widowed), m3 (divorced), m4 (separated), m5 (single)
• Education 7 groups: e1 (none), e2 (incomplete primary), e3 (primary), e4 (incomplete secondary), e5 (secondary), e6 (incomplete tertiary), e7 (tertiary)
(Education not available for Spanish sample in 1994)
Subset MCA of middle categories –“women working”, Spain
0 1 2 3
AM
BMCM
DM
EM FMGM
HM
IMJM
KM
exactly thosewordednegativelytowardswomen
0 1 2 3
-1
0
1
2
exactly thosewordedpositivelytowardswomen
Demographic averages on subset MCA map ofmiddle responses – “women working”, Spain
-0.10 -0.05 0.00 0.05 0.10 0.15
-0.1
0-0
.05
0.00
0.05
0.10
g1
g2a1
a2
a3a4 a5
a6
m1
m2
m3
m4
m5
widowed
separated
56-65yrs
divorced 66+yrs
moremiddles
statementspositive aboutwomen working
lessmiddles
statementsnegative
aboutwomen
working
16-25 yrs
M
F
Checking the MCA results on the data:calculating averages per respondent
Average per Respondent (Spain'94)
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1 2 3 4 5 6
Age Group
MiddlesMissings
(M4)separated 0.654
(M2) widowed 0.308 (not significant)
(significant P<0.0001)
AGE GROUPSMARITAL GROUPS
(not significant) widowed &
separated groupssmall samples