System changes that may affect results

Reproducible results have been central to the development of microdata.no. Built-in data versioning is among the measures to achieve this. Comprehensive test batteries that ensure that updates to underlying libraries and components do not result in changes to analysis results is another important measure.

In certain cases, however, it is not possible to avoid changes that affect results/analyses. Changes in confidentiality mechanisms are one type of system change that can lead to new executions of older scripts yielding slightly different results than before.

We document here system changes that may affect results.

2024-08-19

Change: Barchart is made more consistent and extended

To this date, the barchart command contained a few logical inconsistencies.

Barchart-consistency is now improved. For certain older scripts that included barchart, the change may cause different outputs or an error message.

In most cases it should be straightforward to make marginal adjustments to the command, and to create better and more sensible barcharts than the original script could produce.

2024-06-18

Change: Winsorisation during collapse to non-pseudonomized variable

If the target unit type of collapse() (specified in the by() clause) is a unit type given by a regular non-pseudonymized variable, e.g. municipality, county, type of business, or level of education, calculations made through collapse() will be subject to the same winsorisation as for descriptive statistics carried out on standard pseudonymised unit types such as e.g. personal ID and family ID.

More information about winsorisation here.