Topic (vii): New and emerging methods Discussion

1
Topic (vii) New and emerging methodsDiscussion

• Discussants
• Rudi Seljak / SORS
• Daniel Kilchmann / SFSO

2
Summary of papers (1)

• WP.40 Partial (donor) imputation with
  adjustments (Netherlands, Norway)
• Discussion of different distance metrics to
  perform minimal adjustments to imputed data in
  order to guarantee consistency.
• WP.41 A forward search algorithm for
  compositional data (Italy)
• Discussion of a procedure for detecting outliers
  in compositional data.

3
Summary of papers (2)

• WP.42 Automatic editing with soft edits
  (Netherlands)
• New formulation of the error localisation
  problem including hard and soft edits and its
  solution with a branch-and-bound algorithm.
• WP.43 On imputation of binary variables in
  registers (Sweden)
• Discussion of random imputation versus
  deterministic imputation in the case of imputing
  binary variables with respect to accuracy leading
  to a recommendation which method to use.

4
Summary of papers (3)

• WP.44 Imputation and editing of income from the
  administrative file in the Census (Israel)
• Procedure adopted to overcome the deficiencies
  of Census and administrative data in the case of
  imputing income.
• WP.45 An R package for selective editing based
  on a latent class model (Italy)
• Discussion of the use of the R package SeleMix
  where a method taking explicitly the probability
  of being in error and the error impact into
  account is implemented.

5

• Paper-Specific Questions

6
Paper-Specific Questions

• WP.40 Partial (donor) imputation with
  adjustments (Netherlands)
• What happens if the only reported value is 0?
• Is it possible to include domains (e.g. salary
  boundaries per education level)?
• WP.41 A forward search algorithm for
  compositional data (Italy)
• Efficiency of the method (true, false) x
  (detected, undetected)?
• Does the method work if there are lots of zeros
  in the data?

7
Paper-Specific Questions

• WP.42 Automatic editing with soft edits
  (Netherlands)
• Is there already a strategy or idea how to define
  the cost of failed soft edit rules?
• Advances in the implementation?
• WP.43 On imputation of binary variables in
  registers (Sweden)
• Why not use multiple imputation instead of
  repeated random imputation?
• The risk of bias and the loss of variance for
  deterministic imputation and the loss of
  efficiency for random imputation increases for
  small sub-populations. Limitation for imputation
  in general?

8
Paper-Specific Questions

• WP.44 Imputation and editing of income from the
  administrative file in the Census (Israel)
• Is it possible to validate the strategy adopted
  by the insurance salary data base 2009?
• WP.45 An R package for selective editing based
  on a latent class model (Italy)
• When will the R package be available on the Cran
  library?