bayesian factor analysis for mixed ordinal and continuous ......make everything continuous (standard...
TRANSCRIPT
IntuitionModel
Example
Bayesian Factor Analysis for Mixed Ordinal andContinuous Responses
Kevin M. Quinn (2004)
Political Analysis
April 19, 2017
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Outline
1 Intuition
2 Model
3 ExampleSurveyCode
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Factor Analysis
Motivation for factor analysis
Explain patterns influence the covariance of observed variablesthrough underlying latent factors
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Factor Analysis
Motivation for factor analysis
Explain patterns influence the covariance of observed variablesthrough underlying latent factors
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Factor Analysis
Depression
Insomnia Loss of Appetite
Factor effects
Figure 1: Factor Effect on Observables
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Factor Analysis
Two methods traditions being integrated:
Normal theory factor analysis (continuous variables)
Item Response Theory (Ordinal/Likert-type variables)
x∗ = φλ′ + E (1)
Problem: posterior density of X would have to come fromdifferent types of distribution
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Factor Analysis
Two methods traditions being integrated:
Normal theory factor analysis (continuous variables)
Item Response Theory (Ordinal/Likert-type variables)
x∗ = φλ′ + E (1)
Problem: posterior density of X would have to come fromdifferent types of distribution
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Factor Analysis
Two methods traditions being integrated:
Normal theory factor analysis (continuous variables)
Item Response Theory (Ordinal/Likert-type variables)
x∗ = φλ′ + E (1)
Problem: posterior density of X would have to come fromdifferent types of distribution
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Factor Analysis
Two methods traditions being integrated:
Normal theory factor analysis (continuous variables)
Item Response Theory (Ordinal/Likert-type variables)
x∗ = φλ′ + E (1)
Problem: posterior density of X would have to come fromdifferent types of distribution
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Factor Analysis for Mixed Factors
Depression
hours of sleep appetite quality
Factor effects
Figure 2: Mixed Factors
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Standard Solutions
Make everything continuous (standard normal theory)
Discretize everything (IRT)
Discard one set of the variables
Do something else that isn’t model based
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Standard Solutions
Make everything continuous (standard normal theory)
Discretize everything (IRT)
Discard one set of the variables
Do something else that isn’t model based
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Standard Solutions
Make everything continuous (standard normal theory)
Discretize everything (IRT)
Discard one set of the variables
Do something else that isn’t model based
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Standard Solutions
Make everything continuous (standard normal theory)
Discretize everything (IRT)
Discard one set of the variables
Do something else that isn’t model based
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Proposed Solution
Generalize the IRT and standard models
Should look familiar:
X∗ = φλ′ + E (2)
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Proposed Solution
Generalize the IRT and standard models
Should look familiar:
X∗ = φλ′ + E (2)
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Posterior Density for Mixed FA
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
Posterior Density for Mixed FA
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
SurveyCode
Data and Context
Survey of SOC204 students (n=130)
Mixture of survey questions designed to be easily influencedby political content
Each precept wrote a question, so there’s a mixture of ordinal,continuous, and open ended responses
So we might assume at least one underlying factor: liberalness(conservativeness)
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
SurveyCode
Data and Context
Survey of SOC204 students (n=130)
Mixture of survey questions designed to be easily influencedby political content
Each precept wrote a question, so there’s a mixture of ordinal,continuous, and open ended responses
So we might assume at least one underlying factor: liberalness(conservativeness)
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
SurveyCode
Data and Context
Survey of SOC204 students (n=130)
Mixture of survey questions designed to be easily influencedby political content
Each precept wrote a question, so there’s a mixture of ordinal,continuous, and open ended responses
So we might assume at least one underlying factor: liberalness(conservativeness)
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
SurveyCode
Data and Context
Survey of SOC204 students (n=130)
Mixture of survey questions designed to be easily influencedby political content
Each precept wrote a question, so there’s a mixture of ordinal,continuous, and open ended responses
So we might assume at least one underlying factor: liberalness(conservativeness)
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
SurveyCode
Code Pointers
Variables have to be numeric or ordered factors
All factor levels have to be represented
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
SurveyCode
Code Pointers
Variables have to be numeric or ordered factors
All factor levels have to be represented
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
SurveyCode
Code
mcmc.out = MCMCmixfactanal(variables, factors = 1,data = dataframe, store.scores = TRUE)
Kevin M. Quinn (2004) Bayesian Factor Analysis
IntuitionModel
Example
SurveyCode
Results
λ1 λ2Trump Approval - -0.09Islam Threat 0.71 0.15Trump Score - -0.83Econ Score - -0.84Foreign Score - -0.80Social Score - -0.89Env Score - -0.80Min Wage 2.73 0.72Crime Issue 0.40 0.05Gun Rights - -0.17Immigration - 0.41Health - 0.55Outsourcing - -0.03Tax Reform - -0.03Dakota Access - -0.02Ed Access - 0.35Military Involvement - 0.03Climate Role 0.12 -0.25Facebook Use 2.92 -0.06Facebook Posts 0.47 0.11Johnson Approval 2.92 -0.16
Kevin M. Quinn (2004) Bayesian Factor Analysis