A reader asked you “What does it mean when “X percent chance of rain” is predicted?” ( http://www.straightdope.com/classics/a2_115b.html ). Your reply was partially adequate, but would have been much improved if you had addressed the classic probability theory question to the effect of "If there is a 60% chance of rain today, and it is twice as likely to rain tomorrow, what is the probability of rain tomorrow? Obviously not 120%.). You could explain the simple answer (80%) and you could mention how questions like that relate to various theories of probability (yes, there are more than one, and for good reasons).
I came up with a rule of thumb that’s never let me down. If an expert cites a percentage chance of something happening, down to a specific number, said expert is probably lying. I came up with this rule dealing with doctors. Many doctors like to cite the precentage chance that something will happen, patient will get better, patient will die, etc., it comes across as impressive. A good doctor will say, “I don’t have a crystal ball”, meaning that the practice of medicine is pretty much a crap-shoot and a doctor’s best friends are hope and faith (not the TV charachters). In 1990 a young quack, a pulmonary specialist, told my mother-in-law that if her daughter did not stop smoking that she would likely die within 5 years. Trudy was really upset for a long time, she probably lost sleep. Connie died of general organ failure last summer (2004).
I worked in quality control n manufacturing for about seven years. When I wanted to, I could cite stats with the best. If someone tries to sell me a bill of goods citing percntages I will ask them to sho me their math. 
Cecil has a limited amount of space for his column. Doug-jensen, while your idea is doubtless a good one – the public need some education about statistics – there was just no way that Cecil was about to write a textbook (or even a chapter) on how basic statistics work.
Heck, if it comes to that, about once a month, Cecil gets asked the dumb old question, “If the temperature if zero today, and it’s twice as cold tomorrow, what will be the temperature tomorrow?” So, we’d not only have to explain how to deal with multiple occurrences in statistics, but in how to deal with assigning numbers to phenomena.
I amafraid you all are taking this probility stuff far too literally. To find out how your weather forcaster for your area has made his guess for the POP number in his forcast go to the weather service website http://www.nws.noaa.gov/ and type in your zip code to get your local forcast. Then look for the link to “Forcast Discussion” in the lower left of the screen. Go ahead- you’re paying for it.
Unless you are already very familiar with weather termininology you will also need to look at the bottom of the frame on the left of the page where there is a link to “glossary” to understand the terminology.
According to the USWS:
“Precipitation Probabilities (PoP): It is defined as the likelihood of occurrence (expressed as a percent) of a measurable amount of liquid precipitation (or the water equivalent of frozen precipitation) during a specified period of time at any given point in the forecast area. Measurable precipitation is defined as equal to or greater than .01 inch or .2 mm. Normally, the period of time is 12 hours, unless specified otherwise. The forecast area, or zone, is generally considered to be a county. In some geographically unique areas (mountains), the forecast area/zone may consist of portions of a county or two counties. At times, some NWS forecasters will use occasional or periods of to describe a precipitation event that has a high probability of occurrence, i.e., they expect any given location in a forecast zone area to most likely have precipitation, but it will be of an on and off nature. Usually, away from the mountains, each and every county is a forecast zone area itself.”
As far as I have seen, the forcaster will take all the information available to him, including computer models, radiosonde data, comments from nearby forcasters, etc. and mix it with a large shot of experience to distill it to one (somewhat meaningless) number. Reading the discussion will give you a much more accurate idea of what the forcaster really knows and what he doesn’t.
I’m not aware of any meaningful way to interpret “twice as likely to rain” by which it would mean 80% in the OP’s case. In fact, I can’t think of any way to meaningfully interpret that at all, given a 60% chance of rain today. What the OP is referring to is “half as likely to not rain”, but that’s a completely different concept.
The “twice as cold” one is easy, though. If the “0 degrees” mentioned is Celsius, then “twice as cold” is about -22 degrees Celsius. If the “0 degrees” is Farenheit, then “twice as cold” would be a numbing -72 Farenheit, or so. We don’t discuss the temperature in terms of some arbitrary reference point of a scale, but rather, relative to a temperature comfortable to humans. One would not refer to a temperature twice as cold as 30 C, or a temperature twice as hot as 10 C, because 30 C is not regarded by humans as cold, and 10 C is not regarded as hot. 30 C is regarded as hot because it’s hotter than the human comfort temperature of about 22, so twice as hot would be twice as far from that comfort temperature, or 38.
I have to agree with the pilot. The percentage of probability is based upon the percentage of the area which gets rain from the oncoming air mass. When the forecast is for 60% chance of rain tonight, check the radar map for the air mass approaching your area. You will see that 60% of the area gets rain at any one time. Meteorologists definitely don’t use statistical chances from past occurrences. They are not almanac composers. They also don’t base weather forecasts on personal experiences. Meteorology is an objective science; imperfect, to be sure, but objective.
I am a meteorologist with the National Weather Service, and have been for 11 years. I saw your response and felt compelled to respond, since it is loaded with incorrect statements.
The pilot’s interpretation is incorrect. Gustinator’s answer references the official definition of probability of precipitation as used by the National Weather Service. It is nothing more than the probability of more than 0.01" of an inch of rain at your location. This is about enough rain to wet the sidewalks and streets for a short time.
Meteorologists definitely used statistical chances from past occurrences! One of the main tools we have for forecasting precipitation probability comes from something called “Model Output Statistics” or MOS, which uses various atmospheric parameters and compares them to past occurrances of what happened.
And yes, meteorologists do use personal experiences to aid them in forecasting. Studies have proven that a forecaster who is new to an area is not going to do nearly as well as a forecaster who has spent a few years getting to know the weather patterns in that area.
I knew it wouldn’t be long before we heard from a real weatherman, I imagine they are a very strong Cecil type demographic. And right, that’s NWS not USWS.
I got interested in weather a few years ago when I got suprised by a sudden change in conditions while I was in my small boat. There is an unbelievable amount of information on the web about weather and a lot of it is totally free. Plus lots of cool charts with all the different model outputs for the next few days, radar, satellite and so forth. I know I can even get the radiosonde data from my area (NYC) although I certainly don’t know what to do with it.
I haven’t got to the stage of putting in my own weather station and hooking it up to the web, but plenty of people have…
I find the forcasters in my area are very forthright about their little tweaks and interpretations of the numbers. A lot of the time the various computer models the forcasters use don’t agree all that well, so they must pick the one they “feel” best about and then adjust the numbers to their own liking. What you can get from reading the discussion is how certain the forcaster is about his prediction. If he says something like,“I went with the AVN and tweaked the POPS up a little due the excess moisture” you know he may be out on a little bit of a limb. If he says, " all models are in agreement that we won’t see any chance of precipitation until next Wednesday" you can put money on it.
I have admit I didn’t know about the model output statistics. Does each model have its own set or is it all rolled into one somehow?
For the first 47 years of my life, I was a Yankee, i.e., I lived in one state or other in the northeast part of the country. If the meteorologist said there was less than a 60 percent chance of rain, I could be confident that it probably wasn’t going to rain.
Five years ago, I moved to St. Petersburg, Florida, a beautiful city that’s on a peninsula on a peninsula. Florida is also very flat compared to what I’m used to. Here, a “hill” is what I would call a slight rise. And there are the more “even” days, because I’m closer to the equator. Obviously, the weather patterns are very different.
Here in Florida, they still provide probabilities, but they mean something very different, particularly in the summertime. Here, if it’s summer and the meteorologist says there’s a 30 percent chance of rain, the effective meaning is: It’s definitely going to rain, and there’s a roughly one in three chance it will rain ON ME.
Mary
I always thought the percentages ment that they took a poll of the weatherpersons who had corns, authritius(or however its spelled) and bunions (again), and if ten % said they hurt, then it’s 10% etc.
spelling and grammer subject to change without notice. (and how).
The answer is very simple. Each local weather bureau is issued ten cows. First thing in the morning the meteorologist goes out and counts how many cows are lying down. If three cows are lying down there’s a 30% chance of rain, six cows, 60% and so on. That’s why the percentage is always given in increments of 10%. In smaller markets they only get five cows so the increment is 20%; two cows, 40%, 3 cows, 60%.
Okay, I’m quite confused. In Cecil’s column, he gives the clear statement that a PoP of x% means that in the last 100 times the same conditions have been seen, it rained x% of the time. Got it.
But then he adds that addendum from Joyce K. in Seattle that states an entirely incompatible answer to Cecil’s, with no additional comment.
What gives? What’s the actual answer and why didn’t Cecil try to sort it out?
Hard to read the Master’s mind, but normally after a column where Cecil says, “The answer is X”, there are half a dozen letters that say, “No, no, my grandma told me it was Y.” I think Cecil was posting one such, because it amused him.
To follow up on what Dex said – I took Joyce K’s response to be a total joke, based on the idea that it rains every day in Seattle, so the weatherman’s forecasts only tell you how much of the day it will rain.
I don’t care what Cecil says, or anyone else, whether a meteorologist or not. I’m still convinced I’m right.
Besides, the way I heard it is that the percentage chance of rain depends upon how many weathermen have gone fishing for the day.
Sorry Barbitu8, but you are not right.
MOS (which is the basis for the rain probabilities in forecasts, see link I posted above) is created through a complex linear equation based on what the computer model predicts the conditions of the atmosphere will be, compared to how many times it rained when the computer model made the same prediction.
Basically, the computer model makes a forecast. Then the MOS looks at all the times the computer model made that forecast and compares it to how many times it rained at a particular reporting station (almost universally, this means an airport; this is where the “current conditions” displayed on the weather channel come from). This is the reason that it can’t possibly represent “really” represent an area, only a point. MOS is only looking at a particular point when it makes the forecast POP.
Now, in some conditions (such as Florida), it can seem more like an aereal forecast than a point forecast. The reason is simple: Florida and much of the South, there are only pop-up thunderstorms during the summer. This means that any particular point may or may not see rain, but someone probably will.
In the Northeast much of the time, and in most places during the winter, it is different, because instead we have large-scale synoptic storm systems moving through. Then, it really looks like a point-chance and not an aereal coverage. Because the computer model is forecasting a particular storm system to move a certain way, the MOS is in effect predicting the chance that, based on the forecast, the storm system will in fact move that way.
But, the basic fact remains unchanged. MOS (and rainfall chances based off of it) is really a point forecast chance, not an aereal coverage estimate.
I had a few gramatical errors in the original; this post just cleans that up…
Sorry Barbitu8, but you are not right.
MOS (which is the basis of the rain probabilities in forecasts, see link I posted above) is created through a complex linear equation based on what the computer model predicts the conditions of the atmosphere will be, compared to how many times it rained when the computer model made the same prediction.
Basically, the computer model makes a forecast. Then the MOS looks at all the times the computer model made that forecast and compares it to how many times it rained at a particular reporting station (almost universally, this means an airport; these same sites are where the “current conditions” displayed on The Weather Channel come from). This is the reason that it can’t possibly represent an area, only a point: MOS is only looking at a particular point when it makes the forecast POP.
Now, in some conditions (such as those commonly seen in Florida), it can seem more like an aerial coverage forecast than a point probability forecast. The reason for this is simple: in Florida and much of the South, pop-up thunderstorms make up 90-95% of the rainfall during the summer. This effectively means that while any particularly point may not see rain, someone probably will.
In the Northeast most of the time, and in the rest of the country during the winter, it doesn’t work out the same, because instead of pop-up thunderstorms, we have large-scale synoptic storm systems moving through. In those situations, the POP really looks like the point-chance forecast that it is, and not an aerial coverage estimate. Because the computer model is forecasting a particular storm system to move a certain way, the MOS is in effect calculating the chance that, based on what happens when the computer model forecasts the storm to move that way, the percentage of the time that the storm system will in fact move that way.
But, the basic fact remains unchanged. MOS (and rainfall chances based off of it) is really a single-point probability of rain forecast, and not an aerial coverage of rain forecast.