I’m planning to do multivariate testing on improving my Anki scores. I think I will mainly focus on speed and mature cards correct. I might however also take the plunge and delve deeper into it to develop a better metric for myself.
I plan to test for 60 days, starting next Monday.
At the moment I have a few ideas of interventions worth testing:
Anki Setting#1: Showing/Hiding the time till the card reappears when answering Again/Hard/Good/Easy
Anki Setting#2: Showing/Hiding the progress bar
Treadmill desk
Background music (I use audio for a fair bunch of my cards, so I would guess that the music doesn’t help)
Daily Vitamin D3 #5000 UI Yes/No
A capsule of a vitamin B mix from ratiopharme. Yes/No
Spending ten minutes after the Anki session doing nothing. Yes/No
Spending 3:50 minutes dancing bachata foodwork before starting the Anki session.
I won’t control the time of the day. Unfortuantely I think there are a few days where I’m not at home and therefore can’t do Anki.
Does anybody have other suggestions of other controlled tests that I can do in parallel.
How long do you do Anki for? If you’re going to this much effort to control interventions, it seems like it would be worth including a few other metrics at the same time to see if any of these interventions affect things other than your Anki scores. I’ll suggest some:
A) How well you predict you’ll do at Anki, reported right before starting. Maybe a 1-10, or a smaller Likert.
B) How well you felt you did at Anki, reported immediately afterwards.
C) How much fun it was. For me, this would vary a lot based on the music and treadmill desk.
D) A brief cognitive testing session after the Anki, either before or after the 10 minutes of nothing.
As for other interventions: if you’re not going to control the time of day, you should at least record it and include it in the analysis. Circadian effects can be strong compared to some of the things you’ll be testing. You might also be interested to analyze performance vs. time-since-last-orgasm.
There are other uncontrolled variables you might want to analyze, but you’re a pro tracker, so I’m guessing you already track those and have thought of them for this experiment.
How are you going to control your intervention schedule?
I was reading something the other day that said that the longer it takes you to recall & the more effort involved, the better. I don’t remember if this was backed with any citations, but I believe their idea was that if you recall a fact immediately, it means that you didn’t really need to “refresh” that memory pathway by studying it. If you are just on the verge of forgetting, then that very effortful recall is going to be much more valuable & stabilizing.
[quote]How long do you do Anki for?[/quote]I think around 1 1/2 years. The primary deck that I’m using has 75000 reviews of cards. It has at the moment a total of 9500 cards.
[quote]As for other interventions: if you’re not going to control the time of day, you should at least record it and include it in the analysis. Circadian effects can be strong compared to some of the things you’ll be testing.[/quote]Good point. Anki records the data in it’s database. But I should probably just take a note of it in case I dedice against delving deep into the database.
[quote]You might also be interested to analyze performance vs. time-since-last-orgasm.[/quote]To you journalists already focus on, “quantified self is about being public with your sex data” that I would stay out of giving that argument any ammunition.
[quote]There are other uncontrolled variables you might want to analyze, but you’re a pro tracker, so I’m guessing you already track those and have thought of them for this experiment.[/quote]Yes, I’m tracking a bunch of variables. Mostly every day after awakening. It’s something I do regardless of this experiment.
[quote]I was reading something the other day that said that the longer it takes you to recall & the more effort involved, the better. I don’t remember if this was backed with any citations, but I believe their idea was that if you recall a fact immediately, it means that you didn’t really need to “refresh” that memory pathway by studying it. [/quote]Taking longer to recall is not the same thing as taking more time.
If I understand Piotr Wozniak right speed does not correlate with how well you know the card when you look at it the next time.
At the same time Piotr Wozniak is a strong believer in calculating the forgetting index to find the point when you are just on the verge of forgetting and believe that’s more valuable for stabilizing the memory.
Running 1 km in 2 minutes is more effortful than running 1 km in 4 minutes.
When my head feels very clear, my speed at answering cards is higher. I think that’s a good sign. Even if I would temporaily raise the amount of cards that I remember by slowing down a bit, I think that would mean less deliberate practice at doing Anki.
I think the fact that I’m getting faster at doing Anki over time is a good sign.
[quote=“Christian_Kleineidam, post:5, topic:490”]Taking longer to recall is not the same thing as taking more time.
If I understand Piotr Wozniak right speed does not correlate with how well you know the card when you look at it the next time.
At the same time Piotr Wozniak is a strong believer in calculating the forgetting index to find the point when you are just on the verge of forgetting and believe that’s more valuable for stabilizing the memory.
Running 1 km in 2 minutes is more effortful than running 1 km in 4 minutes.
When my head feels very clear, my speed at answering cards is higher. I think that’s a good sign. Even if I would temporaily raise the amount of cards that I remember by slowing down a bit, I think that would mean less deliberate practice at doing Anki.
I think the fact that I’m getting faster at doing Anki over time is a good sign.
[/quote]
[quote]Myth: High fluency reflects high memory strength. Our daily observations seem to indicate that if we recall things easily, if we show high fluency, we are likely to remember things for long.
Fact: Fluency is not related to memory strength! The two-component model of long-term memory shows that fluency is related to the memory variable called retrievability, while the length of the period in which we can retain memories is related to another variable called stability. These two variables are independent. This means that we cannot derive memory stability from the current fluency (retrievability). The misconception comes from the fact that in traditional learning, i.e. learning that is not based on spaced repetition, we tend to remember only memories that are relatively easy to remember. Those memories will usually show high fluency (retrievability). They will also last for long for reasons of importance, repetition, emotional attachment, etc. No wonder that we tend to believe that high fluency is correlated with memory strength. Users of SuperMemo can testify that despite excellent fluency that follows a repetition, the actual length of the interval in which we recall an item will rather depend on the history of previous repetitions, i.e. we remember better those items that have been repeated many times.[/quote]
This is why I make multiple cards for the same item & learn things both forwards & backwards – sure I can remember it when I’m sitting at home alone in complete silence, but can I remember it when an angry doctor is on the phone yelling at me?
In summary, they are separate but both important depending on your goals.
Improving both recall speed & recall accuracy seem like rational goals.
[quote]This is why I make multiple cards for the same item & learn things both forwards & backwards – sure I can remember it when I’m sitting at home alone in complete silence, but can I remember it when an angry doctor is on the phone yelling at me?
[/quote]I’m also using multiple cards per items.
In case of foreign vocabulary it’s even more than just forwards&backwards. I also have a card that maps sound of the new word to it’s spelling.
Today I tried for the first time to do Anki+Treadmill desk for 20 minutes. Afterwards my right shoulder got tense.
Maybe I have to do a few days of 10 minutes per day Anki+Treadmill desk to adept to the process before I will start my formal experiment.
Today I also had a TV journalist over and explained to him QS with this Anki + treadmill desk example.
The experiment seems to be perfect in terms of having something that looks exiting.
If I tell the story that I read on the internet that a treadmill desk improves Anki performance it’s also very easy to explain that I want to have my own empirical data to verify the claim that Anki performance gets improved by the treadmill desk.
Get the book “Practical Experiment Designs for Engineers an Scientists” by William J. Diamond. In your initiating post, you mention 8 variables you want to test, each with a high/low value (e.g. on/off, yes/no, etc.) If you’re only interested in main effects (i.e. you have reason to assume interactions between your variables are not significant), you can use a resolution III design. The book is basically a recipe book: you have your sigma (standard deviation) because you’ve done ANKI test duplicates before, use alpha=beta=0.1 and delta=2.5*sigma to get your sample size (N-high=N-low). Then follow the recipe to lay out, perform then analyze the results. These values for alpha (manufacturer’s risk; you say something is different when it’s not), beta (consumer’s risk; you say there is no difference when there is) and delta (difference you want to detect) are good starting points for an initial round of experimentation.
You should do residuals analysis before reaching conclusions to ensure there isn’t a time effect. If there is a time effect (sequential repeats improve the response) you’ll need to account for that in follow-up experiments.
The book has a computer program you can run. Alternatively, the recipe is so easy to follow you can do it with some columns on a piece of paper with a pencil.
If you’re going to do 60 days of work, you might as well be able to say something with a quantifiable amount of confidence. If you’re going to do a multivariate test, use a multivariate experimental design.