In many fields of science, such as population genetics or climatology, we are uncertain of past conditions but have knowledge that makes some scenarios more likely than others. We can estimate the likelihood of various past states in a way that is very similar to predicting of future conditions from current knowledge. This gives useful insight into the processes leading to the present, which inform our predictions of the future. As an example, we may wish to infer the number of founder individuals of an invasive species based on genetic diversity observed in the current population.

Historians also use a similar approach when considering the consequences of particular events and actions, particularly through the lens of alternate histories.

Does anyone know of a specific word for this retrospective prediction?

  • It need not be qualified as "retrospective" -- prediction works both ways, for the future as well the past, in spite of the prefix pre-, especially in science. See use cases.
    – Kris
    Commented Jul 14, 2018 at 11:18
  • I don't think that prediction does really work both ways (or that is a newish usage). The Oxford English Dictionary (online) gives it as "a forecast" and "the prediction of future behaviour". Similarly, Collins ED "a forecase, prophecy, etc." Neither gives any hint of usage for inference of past events. Dictionaries often fail to capture common scientific usage but in this case I think prediction is an extension that requires qualification ("predicting the past"). Retrospection is also not right because it simply means looking back, without conditional inference. Historical inference is OK.
    – Spookpadda
    Commented Oct 14, 2018 at 9:21

5 Answers 5


I've frequently heard the term retrodiction used by physicists in particular, to refer to the deduction of a past state of affairs from a current state, and a set of reversible laws.


In the scientific community, uncertain explanations for past and present phenomena are referred to generally as hypotheses. This doesn’t specify a time-frame, but is applicable for past and present and sometimes future predictions. An alternative word is theory, which is merely something uncertain but believed to be true, past, present, and future. Hope this adequately targets your question!

  • Not really. The hypothesis is a more general description of a process, as you note it is applicable to past and future. Predictions are specific expectations based on a hypothesis. In this sense these predictions relate directly to measurable phenomena. In the example above the hypothesis might be that the invasive species is spread by vegetative reproduction. Current genetic diversity in the invasive species are the data. We wish to predict the number of founder individuals, given current diversity, the time since introduction and the hypothesized frequency of assexual reproduction.
    – Spookpadda
    Commented Jul 13, 2018 at 14:30
  • I see what you mean. In that case the word "estimate" should do the trick I guess Commented Jul 13, 2018 at 19:05

From your description, it sounds like you have already have a model that can be used to predict a later state from an earlier state, and a set of current observations from which you can compute some but maybe not all of the model parameters. Or perhaps there are some properties of the model that prevent it from simply being “run backwards” to compute its earlier states.

If so, your task would be to search the space of initial conditions. From each point in that space, you run the model forwards and compare the results with your observational data.

If you search takes the form of a grid search you could describe this as iterating over an assumed set of priors (or initial values).

If your search is based on random sampling, you may be using the Monte Carlo Method to estimate the distribution of initial conditions that corresponds to your present observations, with the possible goal of making a maximum liklihood estimate.

Others have already suggested estimate. However, in most problems where an earlier state cannot be calculated directly, there is usually some larger set of assumptions that determine which variables are random, what properties describe their randomness and so on. There may be a specialist term unique to your discipline.


You can regard it as afterburn:

In the field of Psychology, afterburn denotes a effect of past events, which influences the mental state and daily schedule of activities of an individual.

Eric Berne, the father of transactional analysis, coined this word.

Usage example:

  • The afterburn of her murder led her husband to the severe mental disorder.

How about imply

to involve or indicate by inference, association, or necessary consequence rather than by direct statement


to take as true or as a fact without actual proof

or postulate?

a hypothesis advanced as an essential presupposition, condition, or premise of a train of reasoning

Hope this helps.

(All definitions from Marriam-Webster)

  • Thanks but none of these really fit here. In this case a prediction (or retrodiction as used by physicists) is a statistically likely event given a conceptual model of the phenomenon and a set of other possible outcomes. The outcome has some degree of uncertainty, however large or small, therefore it is neither an implication nor a supposition.
    – Spookpadda
    Commented Feb 27, 2019 at 23:15

Not the answer you're looking for? Browse other questions tagged or ask your own question.