A government document can't (shouldn't) be rasterized because it would loose all the tagging information making it non-accessible in a 508 sense.
It is however trivial for even a non-technically inclined person to remove the text content through Acrobat's content editing pane.
Just as you suggest, things like this seem to be an example of someone in a large organization not quite understanding how the "magic" behind the tools they use work and winding up with results they didn't expect.
From what I have seen in working with the government and the web I think that anigbrowl's comment is spot on. Many people use tools that are designed for the web or the desktop in the same way that they used to use tools designed for print. In this case just slapping a black line over it works in print, and it does not occur to many people that don't fully understand the technology they use that it might not be directly analogous.
A significant portion of my day job involves working with people in government that have a mental model of content distribution that was constructed during the era of print. Its a fun challenge to help people adjust those models to make things more efficient / more informative / less dangerous.
Perhaps they want to keep the formatting, and don't want pages to reflow after they've already been layed out? The Unicode character "█" works well for this purpose.
I remember a similar story from a few years ago where redacted names where discovered from the length of the redacted characters. Turned out it wasn't too difficult to brute force names and see which ones matched the exact size of the redaction.
It is however trivial for even a non-technically inclined person to remove the text content through Acrobat's content editing pane.
Just as you suggest, things like this seem to be an example of someone in a large organization not quite understanding how the "magic" behind the tools they use work and winding up with results they didn't expect.