I would say that depending on the exact data source and methodology, they are a kind of primary source, but one that like all other primary sources has it's own particular set of biases which the careful scholar must take into consideration. Often modern government data sets are fairly transparent about their collection and collation methodologies, so you can understand how the data was collected, and exactly that is and isn't being measured. For example, I could without hesitation call most of the reports and data sets produced by the U.S. Bureau of Labor Statistics (itself an agency within the U.S. Department of Labor) a type of primary source, complete with manual describing the known set of priorities and biases on the part of the authors.

That said, I would be a little more careful about how I used data from an international organization that aggregates data from multiple countries, because data collection practices can vary greatly between international jurisdictions. Even seemingly socially-neutral data like urbanization rates can be hard to compare across different countries, as each state uses a slightly different way of measuring the difference between urban vs rural areas.