in late 2003, jost and colleagues received death threats for publishing a paper that said that being...
TRANSCRIPT
In late 2003, Jost and colleagues received death threats for publishing a paper that said that being politically conservative served a number of ‘existential functions’, not all of which were positive…
Both SDO and RWA were significant predictors…
Right-Wing Authoritarianism
Social Dominance Orientation
Here’s my take on this…
Consider (again) the correlations between Social Dominance, Authoritarianism, and Political Conservatism:
Both SDO and RWA are significantly correlated with Conservatism, BUT they are also correlated with each other.
Correlations
1 .453** .653**
. .000 .000
406 406 406
.453** 1 .632**
.000 . .000
406 406 406
.653** .632** 1
.000 .000 .
406 406 406
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
SDOSCALE
RWASCALE
S6O_SHOR
SDOSCALE RWASCALE S6O_SHOR
Correlation is significant at the 0.01 level (2-tailed).**.
Controlling for.. RWASCALE
S6O_SHOR SDOSCALE
S6O_SHOR 1.0000 .5310
( 0) ( 403)
P= . P= .000
SDOSCALE .5310 1.0000
( 403) ( 0)
P= .000 P= .
Controlling for.. SDOSCALE
S6O_SHOR RWASCALE
S6O_SHOR 1.0000 .4987
( 0) ( 403)
P= . P= .000
RWASCALE .4987 1.0000
( 403) ( 0)
P= .000 P= .
So, the question is whether the variation that these two variables share with conservatism is unique (e.g., SDO and RWA explain different ‘parts’ of
conservatism), or shared (e.g., they both explain the same bit)… Partial correlations can help:
.65 .53 .63 .50
Controlling for.. S6O_SHOR
RWASCALE SDOSCALE
RWASCALE 1.0000 .0678
( 0) ( 403)
P= . P= .173
SDOSCALE .0678 1.0000
( 40) ( 0)
P= .173 P= .
And we can look at the relationship between RWA and SDO once we partial out Conservatism (the bit the two share with conservatism):
.45 .07
So… what does this tell us?
The major part of variance that SDO (and RWA) share with conservatism, is unique to SDO (and RWA) – the part of Conservatism that they explain does not overlap (much).
The major part of the the variation that SDO and RWA share is also shared in common with Conservatism.
Given the results of these analyses, what would you expect to find if you regressed Conservatism onto SDO and RWA?
Model Summary
.754a .569 .567 .47007Model1
R R SquareAdjustedR Square
Std. Error ofthe Estimate
Predictors: (Constant), SDOSCALE, RWASCALEa.
ANOVAb
117.653 2 58.827 266.230 .000a
89.048 403 .221
206.701 405
Regression
Residual
Total
Model1
Sum ofSquares df Mean Square F Sig.
Predictors: (Constant), SDOSCALE, RWASCALEa.
Dependent Variable: S6O_SHORb.
Coefficientsa
1.702 .078 21.704 .000
.281 .024 .424 11.551 .000
.350 .028 .461 12.579 .000
(Constant)
RWASCALE
SDOSCALE
Model1
B Std. Error
UnstandardizedCoefficients
Beta
StandardizedCoefficients
t Sig.
Dependent Variable: S6O_SHORa.
EXCELLENT!…
…as long as SDO, RWA, and conservatism are NOT the same thing!
Remember – Jost et al., say that antiegalitarianism and resistance to change are defining characteristics of conservatism, not the psychological basis of it.
If you EFA the SDO and RWA scales they seem to reduce fairly neatly into two, clearly distinguishable, factors, but…
When you include the conservatism items it gets REALLY messy.
What I need is a technique that will allow me to compare how good a one-factor model (SDO, RWA, Conservatism as all representing the same ‘latent’ construct) with a three factor (SDO, RWA and Conservatism as separate but correlated latent constructs).
…Confirmatory Factor Analysis…
Here, we have the end result of a CFA that forces all the items (‘packeted’ together) into a one-factor solution:
LSRP 2004chisq= 827.801df= 90p= .000gfi= .588cfi= .595rmr= .172srmr= \srmrrmsea= .176
.08
S6
s6_pkt5c
.48
e11
1.031
s6_pkt4c
.44
e101.651
s6_pkt3c
.36
e9 1.081
s6_pkt2c
.33
e8.69
1
s6_pkt1c
.40
e7
1.00
1 rwapkt1b
.46
e11
rwapkt2b
.49
e21
rwapkt3b
.38
e31
rwapkt4b
.38
e41
rwapkt5b
.23
e51
sdopkt5
.86
e16
1
sdopkt4
1.04
e15
1
sdopkt3
1.28
e14
1
sdopkt2
1.19
e13
1
sdopkt1
.85
e12
1
3.08
2.89
3.04
2.45
2.73
1.111.24 1.55 1.301.34
And the results when we allow packets of items to ‘load’ onto their respective latent variable:
LSRP 2004chisq= 182.292df= 87p= .000gfi= .912cfi= .948rmr= .044srmr= \srmrrmsea= .064
.16
S6
s6_pkt5c
.34
e11
1.181
s6_pkt4c
.41
e101.211
s6_pkt3c
.32
e9 .921
s6_pkt2c
.25
e8.86
1
s6_pkt1c
.32
e7
1.00
1
.80
RWA
.21
rwapkt1b
.40
e1
1.00
1
rwapkt2b
.44
e2.94
1
rwapkt3b
.33
e3.981
rwapkt4b
.35
e4.79 1
rwapkt5b
.22
e5
.861
.72
SDO
sdopkt5
.28
e16
1.00
1
sdopkt4
.54
e15
.94
1
sdopkt3
.54
e14
1.14
1
sdopkt2
.74
e13
.89
1
sdopkt1
.52
e12
.77
1
.22 .21
One-Factor model
Χ2(90)=827.80, p<.001GFI=.588CFI=.595RMR=.172RMSEA=.176
Three factor model: Χ2(87)=182.92, p<.001GFI=.912CFI=.948RMR=.044RMSEA=.064
So, the question is whether the variation that these two variables share with conservatism is unique (e.g., SDO and RWA explain different ‘parts’ of
conservatism), or shared (e.g., they both explain the same bit)… Partial correlations can help:
The difference is Χ2(3)= (827.80–182.92) = 644.88, which is significant.
This means that the data are a better fit for a model in which SDO, RWA, and conservatism are separate constructs, than if they are all parts of the same thing.
sdoscale
rwascale
s6o_shor
e3
1
e5
1
e4
1 sdo play with gen sampchisq= .000df= 0p= \pgfi= 1.000cfi= 1.000rmr= .000rmsea= \rmsea
sdoscale
rwascale
s6o_shor
.88
e3
1 .22
e5
1
1.15
e4
1
.35
.28
.46
If I want to have a look at the relationship between Social Dominance, Authoritarianism, and Political Conservatism, I can do that using SEM:
First, specify the paths… Then run the model
Coefficientsa
1.702 .078 21.704 .000
.281 .024 .424 11.551 .000
.350 .028 .461 12.579 .000
(Constant)
RWASCALE
SDOSCALE
Model1
B Std. Error
UnstandardizedCoefficients
Beta
StandardizedCoefficients
t Sig.
Dependent Variable: S6O_SHORa.
Do those unstandardised path coefficients (.35 for SDO to Conservatism and .28 for RWA to conservatism) look familiar?
They should – they’re identical to the unstandardised regression coefficients we got from regressing conservatism scores onto SDO and RWA.
In fact, for these examples, SEM does exactly the same thing as regression…But with SEM we can do SO much more!
Openness to Changevs
Conservation
Self-Transcendence vs
Self-Enhancement
DOMINANCE(Social Dominance
Orientation)
Conservatism
SUBMISSION(Right-Wing
Authoritarianism)
Social Values → Beliefs about Subordination → Ideology
Y’see – I have a theory… that political conservatism is founded on two sets of ‘deeper’ beliefs – about subordination (submission or RWA, and Domination or SDO) which are in turn founded on even more basic values – guiding principles in people’s
lives.
And here it is – and it works, okayish… The fit indices are good, and the RMSEA is borderline satisfactory. If there’s one thing that looks dodgy it’s the
non-significant path from openness to change to conservatism.
GT samp 1chisq= 8.398df= 3p= .038gfi= .988cfi= .981rmr= .351rmsea= .082
.00
rslfvale
.00
ropennes
.14
sdoscale
.24
rwa_shor
.45
s6o_shor
e3
e5
e4
-.43
-.37
.40
.23
-.28
e1
e2
-.06
.23
When I eliminate that path, the model improves…
GT samp 1chisq= 9.751df= 4p= .045gfi= .986cfi= .979rmr= .358rmsea= .073
.00
rslfvale
.00
ropennes
.14
sdoscale
.24
rwa_shor
.45
s6o_shor
e3
e5
e4
-.43
-.37
.40
.26
-.28
e1
e2
.23