Now that we’ve imported nearly 1,600 serious incidents from the Detention Logs data set, here’s a description of our methodology and process.
Our stating point was the raw dataset provided by the ‘Detention Logs‘ team. As this set contains 65 categories ranging from insect bites, power failures and external protests to deaths, the 7,600-odd incidents were filtered down to the more serious ones that directly involved people in detention.
Unlike the Detention Logs site (and, it seems, most other projects that directly use their data), the Refugee Rights Watch project has its own hierarchy of incident types, so a mapping from DIAC’s taxonomy to our own had to be established. The mapping we used is shown below (with our categories marked in bold):
- Accident: Accident/Injury – Serious
- Mental health: Clinical Depression, Self Harm – Threatened
- Deaths: Death
- Medical emergency: Emergency – medical – offsite
- Self-harm: Self Harm – Actual, Self Harm-Attpted Serious
- Use of force: Use of Force, Use of Restraints, Use of Restraints-Unplnd
- Hunger strike: Voluntary starvation (>24 hrs), Voluntary starvation by Minor
- Rights of the Child: Voluntary starvation by Minor
- Relocation: Transfer btwn facilities, Transfer to APOD
- Deportation imminent: Removal – Aborted
Note that incidents of attempted suicide were manually flagged as DIAC’s data doesn’t have a corresponding category; the manual review was doubly necessary given the inconsistent, and arguably incorrect, classification of hangings, etc., variously as attempted serious self-harm, self-harm actual and threatened self-harm.
In addition to those centres that had a unique mapping from location name to facility, the following mappings were used during our import process:
- Christmas Island IDC: Christmas Island, Construction Camp APOD, Lilac Aqua, North West Point Immigration Facility, Phosphate Hill APOD, Phosphate Hill B Compound
- Leonora APOD: Leonora APOD, Gwalia Lodge, Leonora Lodge
- Northern IDC: Northern IDC, Berrimah Accommodation Facility
As with other projects, data from the Jandakot APOD and Virginia Palms Motel were omitted due to their low incident count.
Importing the Data into the Watch Site
The data import process proper consisted of two steps:
- running a custom Python script to create an XML document with one node per report — in addition to including the remapped categories and locations, incident titles were constructed based upon incident types and locations, and fields such as date, lat–long, the description and source links were included based on the corresponding fields in the ‘Detention Logs’ dataset;
- the resulting XML file was imported into the watch site using the underlying engine’s import functionality (after a bug in the Ushahidi platform was found and fixed).