Release 9/11: short release post on jaredknowles.com #50
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: Civilytics/uscogdata#50
Reference in New Issue
Block a user
Depends on #47 (needs a working install line to point at).
Short release post on jaredknowles.com. Precedent:
content/posts/mertools-1-0-released/— same front matter shape (type,slug,lede,tags,license,citationblock), and note the raw-HTML dropcap convention, since goldmark does not parse markdown inside a raw HTML block.What the post has to say
The pitch is the corpus, not the package. Every dollar US state, county, municipal and township governments reported raising and spending, FY1967–FY2024 — 46,148,034 rows, 190.6 MB, in one queryable place, free.
The honest hook is that the Census of Governments is the only nationwide source for this and is genuinely hard to use: item codes change meaning across vintages, identifiers were renumbered in 2017, and an absent value means "published zero" in one era and "not reported" in the next. The package handles each of those and tells you when it has.
Worth including: the three-line quickstart that runs with no configuration, and one real finding from the data rather than a feature list.
Deliberately NOT in this post
Done when
Draft ready for review:
content/posts/uscogdata-released/index.mdOn branch
feat/uscogdata-r-package-cardin the jaredknowles.com repo (same branch as#51, so they can be reviewed together). Not published — merging is what publishes.
Follows the
mertools-1-0-releasedprecedent: same front-matter shape, and the raw-HTMLdropcap with hand-written HTML inside it, since goldmark will not parse markdown in a raw
HTML block.
The finding is real, and it changed while I was checking it
The issue asked for one real finding rather than a feature list. I computed it against the
published corpus, and the first version was wrong in an instructive way.
A ten-category basket said police went from 40.1% to 45.1% of city spending — emphatic,
and nearly meaningless, because the denominator was a basket I picked. Against all city
current operations:
Police operations grew 2.46x in real terms over 45 years; all city operations grew
2.39x. The share is roughly where it started. What actually moved is per resident:
median city police operations went $109 → $225 in real 2024 dollars.
Both statements are true and which one leads is a choice — which is the argument for the
data being easy to check. That is the post's spine.
Checks run before writing it: census years only (every other year is a sample);
identical under both the
primaryanddirectexpenditure concepts; zero coverage-gapsuggestions from provenance.
The post then names the four breaks the package surfaced unprompted for that query —
including the FY2022 capital-outlay merge, which turned out not to apply because the
series is operations only. That non-applying case illustrates what the tool does better
than a feature list would.
Held back, per this issue
until #52 is designed.
One thing to check on review
The scope paragraph says special districts and school districts being out of scope is
"a real gap, and for school finance questions it is a disqualifying one." That is honest
and I think it belongs, but it is the sharpest self-criticism in the post — worth a
deliberate yes/no from you rather than sliding through.
Renders clean: dropcap intact, table and both code blocks fine, no stray markdown.