ok cupid generated a lot of publicity and interest by number crunching all the data they collected to answer questions about attractiveness and dating

I would love to see skritter do something like this.

Finally we could get some concrete answers about learning Chinese

E.G

Is Heisig really the best way to learn?

what is the optimal time of day to study?

What is the optimal duration of study?



What other questions would you like to see answered?

Anyone with a background in statistics on the forum care to weigh in?

the ok cupid blog OK trends blog.okcupid.com

Under what circumstances do someones "characters known" tend to decrease? (my problem)

I'd love to see those posts, too. If I had more time, I'd love to do them. George wants to do them, but he would have to wait for me to get the data out for him to chew on.

Also, the results from these things wouldn't be individualized--they might not hold true for you. But with SRS, it's easy to construct experiments to test different method or study times. You just get two equal lists of characters or words to learn, then learn one list with one approach and another one with another approach and see how long or how many you learn. Since Skritter keeps the stats for you, it's one of the simplest self-experiments you can do. I'm going to be doing some of those myself. If anyone wants to perform their own experiments, I'll be happy to post their results on the blog.

@joshwhitson13 Do you mean according to the Skritter definition for "characters known"? For me it decreases every time I start studying after a hiatus.

I would be interested in the following question: Given a fixed amount of study time per day (or week), what is the effect of splitting that time up into multiple study sessions? If two sessions are better than one, does that mean that three sessions would be even better than two, ten even better than that, etc?

I read a study that asked that question about studying mathematics and they showed that spreading material out over time increased long-term retention. However, I don't think they were using spaced repetition.

James

James, that's a hard question for us to answer given the amount of data we would need to mull--we'd have to go to the level of individual reviews, and it would be hard to identify periods of mostly reviewing or also adding. But it's one of the easiest questions to answer personally using an experiment with learning two equivalent groups of characters, one with X/3 minutes 3 times a day and one with X minutes once a day.

On a related note, I've been meaning to ask. Does Skritter involve the difficulty of an item in its scheduling? I was thinking that you must now have quite a lot of data on the relative difficulty of different items, for example most people remember 你好 pretty easily and maybe forget 瓤 quite a lot. So is this taken into account when scheduling?

Or after testing you on an item a few times, can the algorithm then assess your individual ability to remember it?

I would expect that a word being difficult or easy is defined by the number of reviews needed by an individual person before "knowing" a word. An SRS picks this up and will schedule accordingly. As some people might find a word difficult and other people don't, I don't see how a word in itself can be difficult. The reason why I need more reviews for certain words don't matter that much I believe. Different reasons, different words, I trust the SRS to deal with it. There's always stuff I never seem to learn. :-)

There is a high correlation between how difficult different users find different words. It should be very valuable to incorporate this information into the scheduling.

That said, I haven't done it yet. It's something I really want to do, but it's a lot of work to get it right, and it's not a very visible selling point to improve the scheduling algorithm.

maybe not a such a great selling point

a great customer retention point though

happy customers= word of mouth