Multivariate Data Exploration with Stata:

1
Multivariate Data Exploration with Stata

• Evaluation and Wish List

Stephen Soldz Boston Graduate School of
Psychoanalysis ssoldz_at_bgsp.edu
2
Principal Components Analysis Purpose Data
exploration and data reduction

• Available in Stata
• Base ado (pca)
• Built-in (factor, pcf)
• score will produce component scores

• Issues/Limitations
• pca just a wrapper for (now undocumented) pc
  option to factor, which user cannot access and
  modify
• Confusing documentation on difference between pca
  and factor, pcf (i.e., scaling of eigenvectors)
• Does not directly allow pca of correlation/
  covariance matrix must use corr2data,
  introducing error
• Does not allow rotate to protect user seems
  patronizing and uncharacteristic of Stata

3
Exploratory Factor Analysis Purpose Data
exploration and data reduction

• Available in Stata
• Built-in factor allows principal factors (with
  and without iteration of communalities), maximum
  likelihood
• Built-in rotate allows varimax (with and without
  Horst correction) and promax

• Issues/Limitations
• factor, pfi (prinipal factors with iteration)
  does not allow specification of number of times
  to iterate this directly conflicts with Gorsuch
  (1983) recommendation that communalities be
  iterated only 3-4 times
• As factor built-in, users cannot modify or build
  on it rotate options very limited (only varimax
  and promax) and users cannot modify, though they
  could access eigenvectors (matix_get) and write
  their own

4
Exploratory Factor Analysis, Continued

• Issues/Limitations
• rotate not well documented, so not clear if one
  could, e.g., rotate canonical correlations as
  suggested by Cliff Krus (1976).

• Available in Stata
• Built-in factor allows principal factors (with
  and without iteration of communalities), maximum
  likelihood
• Built-in rotate allows varimax (with and without
  Horst correction) and promax

5
Correspondence Analysis Purpose Data
exploration and reduction of categorical data

• Available in Stata
• User-written coranal (correspondence analysis)
• User-written mca (multiple correspondence
  analysis)

• Issues/Limitations
• Graphics broken in Stata 8
• Statalist question as to whether mca is producing
  correct output
• Few variations implemented

6
Optimal Scaling Purpose Data exploration,
reduction, and transformation

• Available in Stata
• None (that Im aware of)

• Issues/Limitations

7
Multidimensional Scaling Purpose Data
exploration

• Available in Stata
• None (that Im aware of)

• Issues/Limitations

8
Conclusion

• Stata is weak inmultivariate exploratory data
  analysis procedures.
• Many existing procedures are inflexible and not
  extensible, or user-contributed and not currently
  maintained.
• Stata lags behind SPSS, SAS, S-Plus, and R in
  this area.