Some key developments in data analysis

1
Some key developments in data analysis

• Michael Babyak, PhD

2
(No Transcript)
3
(No Transcript)
4
Areas of development

• Discarding flawed techniques
• New types of models
• Treatment of missing data
• Simulation and empirical tests
• Validation

5
Techniques largely discredited or highly suspect

• Categorization of continuous variables without
  good reason
• Automated variable selection without validation
• Overfitted or cherry-picked models

6
New types of models

• Regression family
• Clustered data
• Factor analysis family

7
Generalized Linear Model
Normal
Binary/Binomial
Count, heavy skew, Lots of zeros
Poisson, ZIP, Negbin, gamma
General Linear Model/ Linear Regression
Logistic Regression
ANOVA/t-test ANCOVA
Transformed
Chi-square
Can be applied to clustered (e.g, repeated
measures data)
8
Factor Analytic Family
Structural Equation Models
Partial Least Squares
Latent Variables (Common Factor Analysis)
Multiple regression
Principal Components
9
You Use Latent Variables Every Day

• A Single Measurement is an indicator of an
  underlying phenomenon, e.g. mercury rising in a
  sphygmomanometer measures the underlying
  construct of blood pressure.
• How do you improve the reliability of blood
  pressure measurement? Measure more than once,
  perhaps even in different setting (e.g.
  ambulatory monitoring).
• A Psychometric Scale is also a collection of
  indicators of an underlying process, attempting
  to triangulate on an underlying construct by
  multiple items (indicators).
• A Latent Variable is a collection of indicators
  with the unshared/unreliable part of the
  indicators removedwhats the problem?

10
Missing Data

• Imputation or related approaches are almost
  ALWAYS better than deleting incomplete cases
• Multiple Imputation
• Full Information Maximum Likelihood

11
Out of Missing Data Work

• Propensity Scoring
• Matches individuals on multiple dimensions to
  improve baseline balance
• Complier Average Causal Effect (CACE)
• Generates a guess at the effect of a treatment
  among all potential compliers, including those in
  the control arm

12
Simulation Example
Y .4 X error
bs1
bs2
bsk-1
bsk
bs3
bs4
.
Evaluate
13
True ModelY .4x1 e
14
Validation

• Split-half better than nothing, but often too
  conservative
• Bootstrap
• Repeated splitting

15
Some Premises

• Statistics is a cumulative, evolving field
• Newer is not necessarily better, but should be
  entertained as regards the scientific question at
  hand
• Keeping up is hard to do
• Theres no substitute for thinking about the
  problem

16

• http//www.duke.edu/mababyak
• michael.babyak _at_ duke.edu
• http//symptomresearch.nih.gov/chapter_8/