Why, in econometrics, survey statistics, and perhaps elsewhere, is the word "prediction" used to indicate a cross sectional estimate of the dependent variable, when it is therefore not a forecast?
There are many confusing terms in the jargon of various disciplines. In survey statistics, examples are "ignorable nonresponse," which is not at all 'ignorable,' "significance," which is perhaps the most commonly misleading term in all of science, and here we consider the word "prediction," when we do not mean a forecast. So where and how did this usage begin? Does anyone know the history? Does anyone care to speculate? It seems to be commonly used in econometrics and in survey statistics. Anywhere else?
In SAS PROC REG, and I would expect in other statistical software, "predicted" values are output, and STDI, in SAS PROC REG, is the individual square root of the estimated "variance of the prediction error," as found in econometrics. (In the first few pages of the paper at the first attached link, we see estimates for the variance of the prediction error for individual cases and for totals.)
So why would anyone think to call this a "prediction?" That seems to indicate an observed value will be forthcoming, when that is generally not the case.
Note that there are related questions on ResearchGate, considering "prediction," "forecasting," and similar sounding terms, so you may want to search for them as well.
Article Using Prediction-Oriented Software for Survey Estimation
Research When Prediction is Not Time Series Forecasting
Probably the word Prediction is reserved for Time forecasting, so we need a synonym, forecasting, in order to play with non temporal variables.
James,
I'm one of those who uses "prediction" when I don't want to assert causality (I'm trained in psychology), but you're right, the meaning people put on that word varies greatly, by and within discipline. When I use it that way, I'm referring to the sense of "predicted values" in a regression model, where you have both "predictor" and "outcome" in the analysis. Regression models typically don't incorporate any meaning of time in the statistical modeling itself. Those things are external to the statistical analysis.
"Is associated with" is preferred by many, but I find it often clumsy in text.
Hope that helps,
Pat
Probably the word Prediction is reserved for Time forecasting, so we need a synonym, forecasting, in order to play with non temporal variables.
Demetris,
To me, "forecasting" is worse for contemporaneous relations, as it implies ("fore") a relation to something that hasn't happened yet.
Pat
Within my little knowledge, econometricians use the word "prediction" for an individual's performance, score, income, etc., and the term "forecast" for the time series variables, such as GDP growth rate, FDI of a certain country or one particular concerned stock price. That is, we use "forecast" if the target variable is certain, and "prediction" for those could be anyone in a group.
I believe that this is not really an answer. I just would like to share my experience. I am curious about the history as well.
Yours sincerely.
Sounds like an answer to me, Kuan-Wei Tseng.
As for the history, I'd like to know that as well. Seems odd.
Etymologically, the terms prediction and forecast seem to have the same meaning. The prefix pre- means before (according to Oxford Dictionary) and dict comes from Latin to say. In Oxford Dictionary, the entry for predict reads: "to say that something will happen in the future", and the same dictionary considers the term as a synonym to forecast.
I think the uses of the terms is jargon-specific. In my experience, forecast has been more used with relation to time series analysis, while prediction has been more used in the statistics community. One point in common to both terms is that they refer to the future. Hence, I think none of them should be used in regression analysis.
Well, I did not read the attached paper, however this is a common question from my students. The answer I usually give is that diagnosis and prognosis is a very similar situation with the same rationale. Prognosis is something you dont know if it is going to happen and you want to make a best guess. Diagnosis is something you dont know for sure if it is really going on right now, but you want to make you best guess. Suppose you are attending a patient with fever and you want to make a probabilistic guess of dengue fever, malaria, zika virus, chkunguya virus, yellow fever virus, west nile virus... etc You are acttually making predictions of what is going on right now and betting on what is most likely. Surprisingly, doctors most often use predictors that occur after the event of interest to make predictions in the diagnosis rationale. For example, doctors use heart enzymes and ECG abnormalities to predict heart attack in a patient with chest pain. These abnormalities always occur after heart attack. Therefore, I understand that prediction is just an effort to assign a probability to an event.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics