In market conversations a reassuring sentence is sometimes heard: “The range is very narrow, so we already know the price quite accurately.” That can miss the actual question. Perhaps the band only denotes the uncertainty of an estimated average value. For a specific flat that still does not establish a reliable price range.
The following reading workshop is aimed at interested parties and real estate companies in Austria and Germany. It does not explain any national valuation standard nor any specific portal function. All case descriptions are fictional. The methodology of an actually used report and the individual valuation must be examined separately.
Ask first: What does the range apply to?
A numerical band needs a reference quantity. For example, the observed lowest and highest values of a selection, a certain middle range of those data, an interval for an estimated average or a prediction for a new single case might be meant. These statements serve different purposes.
A headline such as “Price certainty 95 percent” is therefore not sufficient without further explanation. Ask about the exact term, the target quantity, the data basis and the model assumptions. If the publication does not contain this information, the interpretation remains open. A diagram that looks respectable does not make up for the missing definition.
In the example a property owner wants to know whether the proceeds of her flat will lie within a certain range. The interlocutor, however, shows a table about the precision of an estimated regional average. Before they discuss the quality of the number, they must clarify this difference in the question.
What a confidence interval for a mean expresses
NIST describes confidence limits for the mean as an interval estimate that makes visible the estimation uncertainty. The underlying estimate can differ for repeated samples. The confidence level refers to the long-term property of the procedure; it is not simply the probability that an already calculated specific interval contains the fixed true mean. [CI]
For readers this distinction is sufficient at first: the uncertainty of an average quantity and the heterogeneity of individual objects are not the same. A precisely estimated average can belong to a very varied group. It does not yet tell where a particular flat lies within that group.
Avoid the reverse inference as well. A large data set does not automatically make any selection a suitable basis for the entire market or for your object. In the review conversation the question remains therefore: Which population is to be represented, how was the selection formed and which prerequisites are necessary for the stated method?
Why the individual property is a different question
NIST distinguishes in the described regression model the uncertainty of the estimated function from the prediction for a single new measurement. For the single-case prediction additional dispersion is added; the prediction interval described there is therefore wider than the corresponding confidence interval for the function. [PI]
No calculation rule for real estate is derived from this here. The important question for the reader is: Does the presented range even claim a single-case prediction, or is it only used as such in conversation? A properly labelled interval also requires that the model used and its limits are understood.
In the fictional case the flat was extensively altered, whereas the comparison group mostly describes a different condition. This difference must be clarified for the concrete task. An additional decimal point or a narrower statistical band neither replaces the inspection of the object nor the examination of whether the data basis fits.
Three documents for a more sensible conversation
First, ask for the exact explanation of the displayed quantity. It should state whether an average, a single new case or an observed distribution is meant. Secondly you need a description of the considered objects and the time period. Thirdly the limited applicability to your own case should be explained.
This is not a request to demand foreign raw data with private individual details. Often a comprehensible methods and coverage description is enough to determine that the number cannot answer the current question. Which further documents are required for an individual valuation you clarify with suitable professional support.
A helpful follow-up question in the example is: “After your explanation this number describes the regional average. What information then does your statement carry about this particular flat, and what uncertainty remains?” It does not attack the source wholesale, but links the claim with the actually designated evidence.
Use the range as information with limits
If a band is later reproduced in your own knowledge contribution its full meaning remains. Do not strike out the methodological limit in favour of a brief success statement. If the interpretation is unresolved, the contribution can mention the open question or refrain from the respective conclusion.
Even a professionally calculated interval is not a promise about a future contract or the success of a marketing campaign. Which statement from which model can be used for a decision must be assessed separately. The gain from this reading workshop is smaller and more concrete: you recognise whether a number at all answers the question you are currently asking.