I am wondering since quite some time, why my retention rate on Character Definitions is so low compared to all others and I don't get it up. Here is my statistics:
Character Writings 2770 2768 89777 94.8%
Character Definitions 2796 2782 19408 84.8%
Character Readings 2761 2756 56292 91.4%
Character Tones 2838 2837 106945 96.3%
Word Writings 5232 5225 32920 92.8%
Word Definitions 5365 5352 21105 92.0%
Word Readings 5214 5209 23563 96.3%
Word Tones 5297 5288 44232 95.7%
In November/December, I stopped adding new items, just making reviews and hoping, that the character definition retention rate would go up. It did, but not very much. Now I looked again at it and was wondering, why the review number for it is also so low. I assume, that e.g. when I do a Word writing, at the same time it is calculated also as Character writing, but anyway, I would assume, that Skritter would bring up the characters automatically for review, as the retention is so low.
I do not know, in which sequence skritter brings up the items for review and how due items are treated, if they are not studied on the same day. What I often do is, when I am somewhere below 100, I stop and only go on the next day. Let's assume, I have left 50 unstudied. When I start the next day, how are these 50 treated? Are they then on top of the list or do they remain in position 1 to 50? Because, if they are there and I again only go down to 50, they will still remain as unstudied. It happens quite often, that I don't find enough time to study for several, then my items due are in the thousands. What sequence would the items then have?
I have now started, to do my reviews with definitions only, bring them down to zero and only then do the others. Does this help?

Character definitions are not reviewed within other words, as you assume, so that's why you have a lower number of reviews for them. I guess the measured retention rates will tend to be lower on them for that reason. Another factor is that definitions are introduced later than writings, so they can get pushed back more frequently by the writings than the other way around, but that shouldn't have a large overall effect.

If your retention rates for character definitions are lower, then the scheduling will tend to lower those intervals more over time, trying to increase them. So you shouldn't have to do anything special; Skritter will be trying to figure it out. Also, a lower retention rate for character definitions is likely and normal. It will help a little bit to do definition reviews first, so unless it's unpleasant, it's probably a good idea.

The items are ordered basically by time_since_last_review / scheduled_interval_between_reviews. So if you do two items on Monday and one gets scheduled for Tuesday (one day) and the other for Wednesday (two days), then by Friday the first item will be 400% due and the second will be 200% due.

I did complicate this a bit by targeting items that are 350% due first, and doing items that are more or less due than that afterward. This prevents you from having to do all these really overdue (and thus really hard) items first and never getting to the easier and more salvageable items in a big review queue.

Hope this makes some sense. The takeaways would be: don't worry about the character definitions too much, and the scheduling will do the right thing when you've got a lot of items due for review.

Nick, thanks for your explanation.
I have this effect since longer time already - minimum half year - and the retention rate is always around 85%, although I target for 95%, but I have never got it up over 90%. It is not only the "ugly" number in itself, but I can also feel, that I'm not good at character definitions. I thought about it again: I more or less have more than 100 items due, which I don't study on the same day, as I'm doing it during day time, so the reviews, which become due on late afternoon or evening, will not be studied the same day. So might be, I'm not doing reviews on character definitions as often as I should do.
Would be interesting to see, whether other users are also facing the same problem. Might be you could then give the definitions a slight higher priority than the other reviews. But anyway, I will from now on finish my definitions as a first priority before moving to other reviews.

I can see exactly the same effect. On both character and word definitions, the retention level is right down and in red. There are also significantly fewer reviews recorded for definitions in the totals, when it should be the opposite if retention is lower.

Roland, doing the definition reviews first should definitely give a boost to your retention rates there. Character-level definitions, though, can be one of the most abstract answer types, and so may just be harder. They also don't come with any active recall built in. It might help to make sure to fully define the answer in your mind (or out loud) for your character definitions before checking, if you don't already.

wispfrog, you actually have more word-level definition reviews in the last month than writing or tone, which is what you would expect from the lower retention rate on them. The reason your character-level definition review count and retention rate would be lower than for the other character-level parts is because you don't do them as part of word-level definition prompts like you do with your writing and tone prompts. Does that make any sense?

I also have a lower retention rate for character definitions, but it's still within the efficient 80%+ range, so I don't sweat it.

Yeah, my results might well be skewed because over the last month I've been almost entirely working through a huge queue that I built up by over-enthusiastic use of the suggested words in the character popup. (Blame your blog entry!)

Most of those words I chose because I already knew the characters they contained. There's probably at least another two months worth to go too! But its been really good for nailing down characters in those words, which I guess was the idea.

But I'm reasonably sure that I noticed some amount of lag in the retention rate for definitions even before I started that. (The colour highlighting made it obvious) There were also noticeably fewer definition reviews in total, than writing or tones.

So I still think there really is something slightly odd going on.

Perhaps you could run some wider statistics over more people.

Is it the case that there are lots of people with a suspiciously low retention rate for definitions? Why isn't the algorithm adjusting to bring that retention rate up?

I'm not saying the results are skewed because of the last month, but because character definitions are intrinsically harder for many people, and the prompts are not as powerful at making memories. And the smaller number of reviews is because you don't get an extra review counted for each character definition whenever you do a multiple-character word, where you do for writing, reading, and tone prompts.

Ah, that all makes sense.

Still, I would have hoped that the low retention rate would cause more reviews that would correct the low retention rate upward.

But maybe the retention rate is misleading, if it includes things that are still 'being learnt', and learning is harder for definitions. Is that the case? If so, it might be nice to have a different figure as well.

Ideally it would only include items that were in a more mature phase of learning, but we didn't set up the calculations that way way back when, so it does count newer reviews. Tough to change something like that now; oh well.