(Feedback from ONS, 2 October, 2017)

Describe issue or need

Statistical Methods need abstracting.

1) Some statistical methods are used as examples of subprocesses, and that is interpreted as those methods only used in those subprocesses. For example, aggregation based on index numbers and seasonal adjustment are part of 6.1 Prepare Draft Outputs. However, aggregation based on index numbers can be applied as part of processing, for example, when high frequency time series are being used such as daily web-scraped indices. Same is true with Seasonal Adjustment.

2) Abstracting methods should link up to a basic classification of methods. Some methods can be used during Design (for example, to estimate the validity of production methods), while others could be used both to support processing and analysis activities.

Phases impacted

Design, Collect, Process, Analyse

Suggested Improvement

1) Provide clarity around methods being applied with different purposes in different subprocesses.

Examples:

  • Outliering methods could be used as part of estimation / calculate aggregates, or repurposed as part of cleaning as anomaly detection methods.
  • Sampling methods being used as part of design / analysis processes in order to simulate or decide on what sampling methods to use in collection / processing.

2) Revise language across subprocesses so that they do not mandate what methods they apply (including, making the language more generic).