Revisiting the movie method: how to order the onyomi

I was told about the movie method for learning the jouyou kanji and their Heisig's keywords and primary/secondary onyomi. (Here is Nukemarine's convenient movie method spreadsheet.) It's a topic that's been discussed a few times on these forums, perhaps most recently-detailedly in 2013.

In brief, it involves:

1. Memorizing RTK's 200-odd primitives (not the real kangxi radicals, but Heisig's more hierarchical primitives) up front.
2. For the 300-odd onyomi (Chinese-derived pronunciations for kanji used by themselves or with compound words) that are shared by the 2200 jouyou kanji, assign a movie (dorama, anime, whatever) to each onyomi. Then learn all kanji that share that onyomi by assigning each kanji to a scene from the movie.
3. Rejoice upon completion because you've learned RTK volumes 1 and 2.

I had some quantitative questions about this method that the people employing it weren't answering in their blogs so I thought I'd share a bit of data analysis using Nukemarine's RTK Spreadsheet RTK 1 and 3 spreadsheets (in this post). Specifically, I was curious about how many kanji might a single onyomi "contain", how were those kanji distributed on a commonness basis, and how to order the onyomi readings when you go to learn them.

(Large plots follow. If they're too big for your screen please view them, with an abbreviated discussion, at http://imgur.com/a/v2ybH)

First, Nukemarine's kanji spreadsheet contained both (newspaper?) frequency as well as the Kanji Kentei (kanken) levels for many kanji, including most of the jouyou. A preliminary question was how far could kanji's kanken levels be used as a proxy for commonness. One of the complaints against RTK that I'm sensitive to is that the ordering sacrifices usability: common kanji are potentially learned late in the sequence. Since the movie method necessarily throws away the overall RTK ordering (building on primitives), we might as well learn the onyomi ordered by commonness.

I see that while the relationship between a kanji's kanken level and frequency is complicatedly interesting, the answer is yes, kanken levels are a reasonable discretization of the frequency range:



---

Next, grabbing the first reading in the "Kana" column of Nukemarine's spreadsheet (that column contained multiple readings, and I assumed that the first reading was the primary one; this also results in kanji without onyomi being assigned a kunyomi) and sorting the resulting onyomi by the number of kanji it contained shows a nice decay in the size of onyomi groups:



The top plot shows the number of kanji in each onyomi group. The bottom plot shows how they cumulatively add up to the 2200 jouyou kanji.

I see that there are 290 primary onyomi. 290! The total number of films, doramas, anime, books, walks, and buildings I remember clearly is probably <30! (And half of those are the individual Firefly episodes... )

Here are the top five onyomi:
コウ (74 kanji)
ショウ (58 kanji)
シ (55 kanji)
カン (49 kanji)
キ (40 kanji)

But the last 150 onyomi have four or less kanji in them. 60 onyomi have a single jouyou kanji in them (e.g., く.う, イク, ネイ, にお.う, とち; what does a dot in the middle of a kunyomi reading mean?)

So any weaponization of the movie method faces the twin problems of requiring at least 150 cultural items to serve as memory palaces which will contain between 5 and 74 kanji, and of then somehow chunking another 150 onyomi with fewer kanji associated with them.

---

Finally, I needed some way to decide which onyomi group to learn first. There's a lot of different visualizations that help but here's a simple one that I like: for each onyomi group, sum up the kanken levels of all kanji in it. The more kanji an onyomi group contains, the higher you expect the sum to be; so the higher the average kanken level for an onyomi reading, the more elementary kanji it contains. The plot below shows, for each onyomi reading, its kanji's sum of kanken levels (horizontal axis) as well as, on the vertical axis, the median kanken level (dot) and 25- and 75-percentile kanken levels (vertical lines).



This confirms the expectation that a single onyomi may be used for kanji with widely-varying commonness. I could start at the top-right of the plot, and sweep down towards the lower-left to order the onyomi into a learning order. The angle of the sweep trades off number of kanji in an onyomi group and how elementary the kanji are therein.

Someone preferring to learn common kanji earlier can pick up the high-kanken level onyomi with fewer kanji in them (smaller sum of kanken levels) before onyomi with a larger stable of kanji but which are in general lower kanken.

Someone more spartan could start from the right and work left, sorting the onyomi only in terms of total kanji in them.

The spreadsheet purportedly made available by Alex Elkholy, the first proponent of the movie method, organized the onyomi alphabetically and dispensing with the number of kanji associated with it. That might be a way to learn too.

---

I am interested in this because (1) I'd like to incorporate the method of loci into my kanji system and (2) learn the onyomi along with the kanji, instead of waiting for the post-RTK sentences/immersion phase.

However, I cannot see an easy way to surmount the two obstacles mentioned above: how to come up with hundreds of movies/doramas/etc. if one doesn't spend much time indulging these or in general has a poor memory for things like that, and how to systematically deal with onyomi that have just one or two kanji associated with them. Perhaps instead of associating onyomi with movies and individual kanji with scenes, one could associate onyomi with scenes/tropes/memories and the individual kanji with stories involving them. Like シ = my best friend whose name was Shi, and about whom I could probably think of 55 stories to tie to the 55 kanji with that onyomi.

Or I could spend less time on kanji and more time watching movies for a few weeks...

---

(Edit: of the 290 onyomi I cite above, eight are provisionally kunyomi (kanji whose "Kana" column in NukeMarine's database were in hiragana). With RTK volumes 1 (sixth edition) and 3 kanji, we get 315 primary onyomi/kunyomi, with 26 kunyomi. So adding RTK3 adds seven more onyomi to learn, though I suspect the name readings of those extra kanji are far more numerous and important.)