Abstract:Ex-post harmonization, whereby data collected under one classification standard are reclassified and redistributed under another to facilitate joint analysis is often treated as a simple data preparation task, but it is in fact complex imputation. When a value is split across several target categories, the harmonized values rest on assumptions that should be stated, checked, and carried through to downstream analysis. Unfortunately, these assumptions are often hidden in custom data wrangling scripts and seldom systematically checked. The Crossmaps framework separates the specification and implementation of transformation logic into two new data structures based on a new task abstraction for mapping aggregate statistics from one classification to another. The part-to-whole array holds a total together with its distribution across a set of keys, and the crossmap holds the redistribution logic. Combined they define the block-level operation of a crossmap transform, which imputes values reported in a source classification into a target classification. We give equivalent graph, matrix and edge list encodings, each supporting different inspection, validation and extraction tasks, to improve the transparency and reuse of harmonization efforts. We include a demonstration based a published ex-post harmonised industrial-statistics dataset using the `xmap` package.
From: Cynthia A. Huang [view email]
[v1]
Thu, 20 Jun 2024 10:07:07 UTC (1,169 KB)
[v2]
Wed, 16 Sep 2026 03:38:46 UTC (1,193 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。