-2
$\begingroup$

I do not see a clear distinction between and tags.

$\endgroup$
5
  • 4
    $\begingroup$ Predictions can be about unknown past values; forecasting can't. Clarifying the tag excerpts / wikis may be appropriate, though. $\endgroup$ Apr 11, 2016 at 15:01
  • 1
    $\begingroup$ It's funny that even though [forecasting] would seem me to a subset of [prediction], the former tag has almost twice as many threads as the latter one. $\endgroup$
    – amoeba
    Apr 11, 2016 at 17:20
  • 2
    $\begingroup$ @amoeba, that need not be surprising as it seems natural to use specific in place of general tags. I do not use time-series tag when I post something specific only to ARIMA or GARCH, even though both are time series models. $\endgroup$ Apr 11, 2016 at 18:37
  • $\begingroup$ As this is directly related to the newer Meta thread about wikis and that one is resolved, do you want to mark the answer here as accepted too? $\endgroup$
    – amoeba
    Jun 14, 2017 at 9:47
  • 1
    $\begingroup$ @amoeba, thanks for the heads-up. Done. $\endgroup$ Jun 14, 2017 at 10:02

1 Answer 1

11
$\begingroup$

Yes, we need both. I think forecasting is more specifically about time-series data -- "What happens next?" Prediction the task of deploying a model to assess unseen data in some way. With prediction, there need not be any time-series component at all; you might just be interested in how well a model does against a holdout set of observations.

Support of this can be found in Cressie & Wikle Statistics for Spatio-Temporal Data, p. 17:

Uncertainty in data, processes or parameters means that there will be uncertainty in conclusions. Statisticians call this drawing of conclusions in the presence of uncertainty, statistical inference (or just inference); in this book, inferences will be either estimation of fixed but unknown parameters, or prediction of unknown random quantities. (Notice that "forecasting," namely concluding something about the future, is a special case of "prediction.")

I've edited this description into the and wikis.

$\endgroup$
9
  • 7
    $\begingroup$ I agree. Although I tend to enjoy fussing over definitions as an acceptable blood sport, to me the argument is intensely practical. Imagine abolishing either of these tags. That would be futile. There are many, many time series people who would feel deprived without "forecasting" and many, many others who would feel entirely uncomfortable without "prediction". $\endgroup$
    – Nick Cox
    Apr 11, 2016 at 14:54
  • $\begingroup$ Fine. Then perhaps we should make the distinction between the two clearer? So far forecasting seems to be a subset of prediction. (I am making things a bit simpler than they are, but I want to highlight a point.) $\endgroup$ Apr 11, 2016 at 14:58
  • 3
    $\begingroup$ I just read the tags. "Prediction is concerned with assessing the probability of unknown values from known values and inferred relationships." I don't think that matches prediction as widely used at all. Prediction is about predicting [sic] the unknown, but there need be no probability explicit. If I predict systolic blood pressure from body weight in some model, the prediction is (to me) just the fitted value. Some probability is at most implicit. $\endgroup$
    – Nick Cox
    Apr 11, 2016 at 15:09
  • $\begingroup$ But in asserting that typically fitted and predicted are one and the same, I may be starting more hares and clashing with usages in particular fields of statistical science (to say nothing of machine learning). $\endgroup$
    – Nick Cox
    Apr 11, 2016 at 15:09
  • 2
    $\begingroup$ Naturally I am not in denial about prediction intervals and so forth. But they are part of how prediction is often done, not an essential part of the definition. $\endgroup$
    – Nick Cox
    Apr 11, 2016 at 15:16
  • $\begingroup$ @NickCox, "in asserting that typically fitted and predicted are one and the same, I may be starting more hares": indeed you are. Here is your first hare. $\endgroup$ Apr 12, 2016 at 6:31
  • 1
    $\begingroup$ @StephanKolassa Thanks for the reference. My premise is that forecasting and prediction overlap but are not identical. The comment cited is how I see the mundane practice at elementary or intermediate level of working with regression-like models. I tend to use fitted values and predicted values more or less interchangeably and find that in reading. I am sympathetic to anyone who wanted to introduce a precise distinction, but that would be flying in the face of much relaxed usage. This still leaves the Wiki text, which I find unsatisfactory, while being shy of trying to rewrite it. $\endgroup$
    – Nick Cox
    Apr 12, 2016 at 6:53
  • 1
    $\begingroup$ @NickCox I just happened upon a definition in a textbook I am reading, and thought you might be interested. $\endgroup$
    – Sycorax Mod
    Apr 12, 2016 at 18:19
  • $\begingroup$ I like it. Presumably "porceses" was a typo. If borrowing for Wiki, correct "either be" to "be either". $\endgroup$
    – Nick Cox
    Apr 12, 2016 at 18:24

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .