Confirmatory Factor Analysis
description
Transcript of Confirmatory Factor Analysis
![Page 1: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/1.jpg)
Confirmatory Factor Analysis
Psych 818DeShon
![Page 2: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/2.jpg)
Construct Validity: MTMM
● Assessed via convergent and divergent evidence
● Convergent– Measures of the same construct should correlate highly
● Divergent– Measures of different constructs should not correlate
highly
![Page 3: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/3.jpg)
MTMM Example
● Byrne & Goffin (1993) ● 158 11th graders● 4 traits measured using 4 methods
– 16 indicators
![Page 4: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/4.jpg)
MTMM Example
![Page 5: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/5.jpg)
MTMM Example
![Page 6: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/6.jpg)
MTMM Example
![Page 7: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/7.jpg)
MTMM Example
● Based on the comparison of nested models– Freely correlated traits and methods– No traits – freely correlated methods– Perfectly correlated traits – freely correlated methods– Freely correlated traits – perfectly correlated methods
![Page 8: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/8.jpg)
Correlated traits and methods
![Page 9: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/9.jpg)
No traits / Correlated Methods
![Page 10: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/10.jpg)
Perfectly Correlated Traits / Freely Correlated Methods
![Page 11: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/11.jpg)
Freely Correlated Traits / Perfectly Correlated methods
![Page 12: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/12.jpg)
MTMM Example: Model Summary
![Page 13: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/13.jpg)
Convergent Validity - Traits
![Page 14: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/14.jpg)
Discriminant Validity - Traits
![Page 15: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/15.jpg)
Discriminant Validity - Methods
![Page 16: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/16.jpg)
MTMM Summary
● Comparisons of specific CFA models that are consistent with the MTMM logic provide practical and statistical evidence of convergent and divergent validity.
● Model comparisons used to examine the function of both traits and methods
● This model sometimes fails to converge or has inadmissable results– Needs at least 3 traits and methods to be identified
● If so, use a correlated uniqueness approach– Kenny (1976), Marsh (1989)
![Page 17: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/17.jpg)
Hierarchical CFA
● Just as latent variables might explain correlation among items, second order latent variables might explain correlation among factors
![Page 18: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/18.jpg)
Hierarchical CFAQuality of life for adolescents: Assessing measurement properties using structural equation modellingMeuleners, Lee, Binns & Lower (2003). Quality of Life Research, 12, 283–290.
![Page 19: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/19.jpg)
2nd Order CFA
![Page 20: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/20.jpg)
Hierarchical CFADepression(CES-D )
Positive Affect Negative AffectSomatic
Symptoms
Happy Enjoy Bothered Blues Depressed Sad Mind Effort Sleep
.795a
.882a .810a
Model Fit Statistics: N= 868, χ2(26)= 68.690, p<.001, SRMR=.055, IFI= .976a Second-order loadings were set equal for empirical identification.All loadings significant at p < .001.
![Page 21: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/21.jpg)
Residualized 2nd Order CFA
![Page 22: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/22.jpg)
Hierachical CFA
● Scaling the latent variables– 1st order latent variables are scaled by constraining one
path to a manifest variable to 1.0– 2nd order latent variables are scaled by setting their
variances to 10.0 (standardized)
![Page 23: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/23.jpg)
Measurement Invariance
● A precondition of comparing groups on a construct (e.g., efficacy, intelligence, personality) is that the measure of the construct functions the same across groups.
● If the measure doesn't function equivalently across groups, you're done!– No group comparisons are meaningful
● Also relevant to longitudinal studies– The measure must “mean” the same thing over time
![Page 24: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/24.jpg)
Measurement Invariance
● The dominant approach to this issue is Meredith's (1993) model of factorial invariance
● Basically compare the equality of CFAs across groups– Aka – simultaneous factor analysis in several
populations (SIFASP)
![Page 25: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/25.jpg)
Types of Measurement Invariance
● Meredith (1964; 1993)
● Factorial InvarianceConfigural invariance
Weak factorial invariance
Strong factorial invariance
Strict factorial invariance
![Page 26: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/26.jpg)
Invariance logic● Evaluate a single factor model across
populations● Questions:
– Is the basic pattern of parameter estimates the same across groups?
– Are the factor loadings equal across groups?– Are the indicator intercepts the same across groups?– Are the error variances equal across groups?– Are the latent means and variances equal across
groups?● This last issue may not mean anything about invariance
![Page 27: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/27.jpg)
Multi-Sample CFA with Structured Means● Must estimate the intercepts for the regressions
of the manifest variables to the latent variable(s)
● Byrne (2004). Testing measurement invariance in AMOS. SEM, 11, 272-300.
![Page 28: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/28.jpg)
1st Step
● Invariance of population covariance matrices– If you find this, then it is reasonable to assume that the underlying model is the same
![Page 29: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/29.jpg)
Configural Invariance
● Relaxed constraints requiring only the same number and pattern of factor loadings– factor-variable regressions need not be the same
across groups (Horn, McArdle, & Mason, 1983).
● Evidence for qualitative similarity across groups but does not permit quantitative comparisons.
![Page 30: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/30.jpg)
Weak Factorial or "Metric" Invariance● Equality constraints on the factor-variable
regressions across groups while ensuring the factor variances and covariances are free to vary
● minimally necessary condition for establishing invariant measurement across groups (Horn, 1991; Horn, McArdle, & Mason, 1983; Meredith, 1964; 1993)
● The factor loadings are proportionally equivalent to corresponding loadings in other groups since the factor variances and covariances in each group must be free to vary across groups
![Page 31: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/31.jpg)
Strong Factorial Invariance ● Strong factorial invariance requires the additional
constraint of invariant mean intercepts across groups– All mean differences in the variables are
expressed at the factor level– Factor mean differences across groups are
expressed as differences relative to an arbitrary group (where factor means are set to zero or some arbitrary value; Sorbom, 1974; see also Horn & McArdle, 1992).
● Group differences in unique means indicates the presence of additive bias
![Page 32: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/32.jpg)
Strict Factorial Invariance
● Additional constraint of invariant error variances as well as unique means and factor loadings.
● Test of whether there are differences in amount of specific/error variance.
● This model forces the combined specific and random error components of each variable to be equivalent across groups such that differences in variance across groups are permitted only at the latent variable level.
![Page 33: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/33.jpg)
Strict Factorial Invariance
● Under strict factorial invariance the measurement models do not differ across the groups and can, in principle, be collapsed into one group.
![Page 34: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/34.jpg)
Evaluating Factorial Invariance● Equal covariance matrices?● Configural
– Same pattern of factor loadings● Weak (“Metric”)Factorial Invariance
– Factor loadings equal● Strong Factorial Invariance
– Factor loadings equal– Mean intercepts equal
● Strict Factorial Invariance– Factor loadings equal– Mean intercepts equal– Unique variances equal
![Page 35: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/35.jpg)
Example
![Page 36: Confirmatory Factor Analysis](https://reader034.fdocuments.in/reader034/viewer/2022042603/568145f4550346895db2f7af/html5/thumbnails/36.jpg)
What to do?
● If strict factorial invariance, or at least strong invariance, does not hold, comparisons across groups cannot be made.
● Drop items?● Covariates?