• Ascolta questa pagina usando ReadSpeaker
  • Condividi
  • Lascia un feedback

The contents related to Banff are shown in the following sections:


Banff is a generalized system for editing and imputing survey data based on the SAS architecture. Banff was developed by Statistics Canada to process numeric and continuous variables. It uses consistency rules (edit) that must be expressed in linear form.

Banff has a modular structure: each module corresponds to a particular sub-function of the general structure of an edit and imputation process of quantitative variables:

  • edit specification;
  • check edits for consistency and redundancy;
  • error localization;
  • detection of outlier values;
  • imputation.

The error localization module uses the Chernikova algorithm based on the minimum change principle or Fellegi-Holt paradigm. For each record that fails at least one edit, the algorithm identifies the minimum number of fields to change (impute) so that the record passes all the rules. In general, the Fellegi-Holt paradigm is considered appropriate to treat stochastic errors.

Banff implements several imputation methods:

  • Deterministic imputation
    It checks if there is one and only one value that, once assigned to the field to impute, allow the record to pass all the edits.
  • Donor imputation
    The nearest neighbour record (according to a specific distance function) to the current failed record is chosen among the potential donors, i.e. units that pass all the edits. All required fields are imputed by transferring the corresponding values from the nearest neighbour record.
    It is important to note that a potential donor will be actually chosen as the donor, if the imputed values are such that the imputed record pass the user-specified post-imputation edits.
  • Estimator Imputation
    Values to be imputed are obtained through modeling or observed data. Examples are mean imputation and regression imputation.

Banff also provides outputs that allow the user to analyze the impact of the editing process on the data (for example, the list of redundant rules or the failure frequency of the erroneous record).


Status: decommissioned
Author: Statistics Canada
GSBPM code: 5.3 Review and validate
5.4 Edit and impute
Keywords: editing for numerical variables, error localization,
minimum change principle, nearest neighbour donor

Software and documentation

To get the Banff software as well as the methodological and technical documentation, please contact Statistics Canada.

Only for Istat staff: contact Francesco Dell’Orco.

Last edit: 19 November 2020