Allan Murphy

Who they are

This page records the public professional record only. Nothing adverse is recorded and no motive is attributed.

His affiliation and wider career were not verified for this page, so none is asserted here. What can be said is what the paper did. It brought together results from operational and experimental forecasting programmes in the United States and asked whether forecasters' stated probabilities matched what happened. Checking forecasts against outcomes in this way is called forecast verification, and the paper belongs to that field.

Key papers

Murphy, A. H., & Winkler, R. L. (1977), 'Reliability of Subjective Probability Forecasts of Precipitation and Temperature', Journal of the Royal Statistical Society, Series C (Applied Statistics), 26(1), 41-47, doi:10.2307/2346866. Read at the level of the published abstract.

His wider publication record was not surveyed for this page.

Key ideas

The paper's subject is reliability, the property now usually called forecast calibration. A forecaster is reliable if, among the occasions on which a given probability was stated, the event happened about that proportion of the time. Of the days given a thirty per cent chance of rain, about three in ten should be wet.

The forecasts examined were subjective probability forecasts for two kinds of quantity: whether precipitation would occur, and the day's maximum and minimum temperatures. The authors' conclusion, in the words of the abstract, was that weather forecasters can formulate such forecasts in a reliable manner.

The result matters to this site because of what a retailer does next. A sales forecast that uses the weather has a trustworthy input at one end, and the studies of retail sales measure how much of the gain survives the step to units. That chain is set out in What a Weather Forecast Does for Sales. The limits are those of the paper: it concerns the American programmes it summarised and the two quantities named, and it does not say that a calibrated forecast is also a sharp one.

How to read the result

Three questions help in using the 1977 paper without over-reading it. The first is what was scored. The forecasts were subjective probabilities for whether precipitation would occur and for the day's highest and lowest temperatures, and reliability was judged by comparing stated probabilities with what happened. The second is what the verdict covers. It covers forecasters in the American programmes the paper summarised, in the period it covered, and the abstract says they could formulate such forecasts reliably. It does not say every forecaster, in every country, in every era, is reliable.

The third question is the one a retailer should ask first: reliable for what use? A reliable forecast is not necessarily a useful one. A forecaster who named the average chance of rain every day would be reliable and tell a planner nothing. That is why the retail studies cited in What a Weather Forecast Does for Sales measure how far sales-forecast error falls when weather is added, not whether the weather forecast is honest. Murphy and Winkler supply the first link in the chain, and the later studies supply the rest.

The paper also shows why weather became the standard test bed for forecast verification. Rain either happens or does not, and the temperature range is recorded, so every forecast can be scored against the day it was made for.

What the record does not show

This page does not cover the author's other work, his institution or his later influence, because none of it was verified for this site. It also does not claim that calibrated forecasts are the best forecasts. The claim carried forward is narrow: of the two inputs a weather-adjusted sales forecast needs, the weather forecast and the sales response, the first has a published record of reliability and the second has to be measured by the retailer.

---