# Forgetting curve

## Formula

**Forgetting curve** describes the decline in the probability of recall over time (source: Wozniak, Gorzelanczyk, Murakowski, 1995):

`R=exp(-t/S)`

where:

- R - retrievability of memory (recall)
- t - time
- S - stability of memory

This text is part of: "*I would never send my kids to school*" by Piotr Wozniak (2017)

## Explanation

In a mass of remembered details, the shape of the forgetting curve will depend on (1) memory complexity (i.e. how difficult it is to uniformly bring individual knowledge details from memory), and (2) memory stability (i.e. how well individual details have been established in memory). For example, a set of easy French words, memorized on the same day, may align into a curve that meets the above formula. Those French words will have low complexity (because they are easy), and low stability (because they have just been learned). Those French words will be lost to memory, one by one, at equal probability over time. The chance of recalling a given word will be R (retrievability) after time t. With time going to infinity, the recall will approach zero. However, if all words are reviewed again, their stability will increase and recall time will be extended. This is used in spaced repetition to minimize the cost of indefinite recall of memories.

## Power or Exponential?

Forgetting is exponential, however, superposition of forgetting rates for different stabilities will make forgetting follow the power law. In other words, when memories of different complexity are mixed, the forgetting curve will change its shape, and may be better approximated with a negative power function (as originally discovered by Hermann Ebbinghaus in 1885). Plotting the forgetting curve for memories of different stability is of less interest. It can be compared to establishing a single expiration date for products of different shelf life produced at different times. Power approximations face the problem of t=0 point. On the other hand, exponential forgetting may seem devastating in its power. Luckily, for well-formulated material, decay constants are very low due to high memory stabilities developed after just a few reviews.

**Forgetting is exponential**due to the random nature of memory interference

## Data

Spaced repetition software SuperMemo routinely collects data and displays a set of forgetting curves that depend on memory stability and knowledge complexity.

Examples of curves collected with SuperMemo:

- Power curve for the first review of heterogeneous knowledge
- Exponential curve for the second review of homogeneous knowledge
- Cumulative normalized curve for different levels of stability and memory complexity (over 214,000 data points taken from over 380,000 repetitions)
- 3D curve for various levels of stability (retrievability log axis reversed in reference to time)

See also: Error of Ebbinghaus forgetting curve

## Example

Figure:The firstforgetting curvefor newly learned knowledge collected with SuperMemo. Power approximation is used in this case due to the heterogeneity of the learning material freshly introduced in the learning process. Lack of separation by memory complexity results in superposition of exponential forgetting with different decay constants. On a semi-log graph, the power regression curve is logarithmic (in yellow), and appearing almost straight. The curve shows that in the presented case recall drops merely to 58% in four years, which can be explained by a high reuse of memorized knowledge in real life. The first optimum review interval for retrievability of 90% is 3.96 days. Theforgetting curvecan be described with the formula R=0.9907*power(interval,-0.07), where 0.9907 is the recall after one day, while -0.07 is the decay constant. In this is case, the formula yields 90% recall after 4 days. 80,399 repetition cases were used to plot the presented graph. Steeper drop in recall will occur if the material contains a higher proportion of difficult knowledge (esp. poorly formulated knowledge), or in new students with lesser mnemonic skills. Curve irregularity at intervals 15-20 comes from a smaller sample of repetitions (later interval categories on a log scale encompass a wider range of intervals)