<...>
Yes, I remember the recent discussion on the matter in your quote. The Dashboard lets you see Merit Distribution by subsection (i.e. local language or second level within Bitcoin, Alt. Cryptocurrencies, Economy and Other). Going a step further and doing it by topic is really too much overtime for me right now (Im doing all this in my spare time aside from a regular fulltime job).
You would need to:
1. Study all the thread titles/OPs that you want to take into consideration (you could assume that all posts within the thread are related by content to the OP to simplify). The title does not always say it all, so you need to go into the OP itself. Then you need to classify the thread and related to 1 (or n) topic/s.
2. Perform the above for all threads created or bumped in a certain timeframe (for example, since the Merit System kick-off, or last month). There are thousands of threads to consider, although if the focus were to be merit related, you could cut down quite a bit by performing the classification only on those merited threads.
3. And then cross the above with merits per post/thread.
It would certainly be interesting, but very long to perform for me right now if done properly.
A quick approximation is to look through the most merited threads (aggregate of posts in the thread) on the Merit Dashboard (see tab called Rankings) filter the Date Range (i.e. August) and/or subsection, and perform a tentative approximation from the list displayed. For example, if you filter Meta for the Month of August we get:
Seeking Reporter Badge Images -> 100 Merits in August.
Community generates suggestions to
-> 97.
Merit? This got to be a joke -> 94 Merits in August.
Analysis Correlation Matrix -> 77 Merits in August.
How recognized are you in the bitcointalk world? -> 69 Merits in August.
[guide] Bitcointalk account security -> 62 Merits in August.
<etc.>
You could create a tentative stat for Meta and other sections by manually classifying each thread (assign each thread one main core topic), and then adding all the Merit by Topic. Doing so for Meta may not be difficult (Merit, Account, Ideas, Plagiarism, Reporting, etc.), but trying to derive a proper topic on Bitcoin discussion or Ann threads may be rather painful, especially if you are not familiar with the content on the subsection.
Edit: I do encourage you nevertheless to give it a go, trying the dashboard approach I mentioned. Try it out on some subsections with which you are familiar, so that the context is known to you. Derived topics should be no more than 15 or 20 max. and can be categorized into groups to reduce the number. If you give it a go and find something interesting, open an OP to show us your findings.
The only fine-print needed is how topic assumption is derived by from the OP on each thread (not all merits in practice are necessary related to the OP of the thread, but its a workable assumption).
As a last note, its probably best in this case to limit data to recent months, in order to avoid heavy topic bias by early heavy meriting.
