I remember an incident from my career in forecasting (not elections) that might help: my forecast said that the cost of a contract could be anywhere from $5m to $75m with the median* at $35m. The senior managers couldn't cope with that. They wanted to know "the cost" or the "maximum cost" (which in theory would be infinite, $75m was the …
I remember an incident from my career in forecasting (not elections) that might help: my forecast said that the cost of a contract could be anywhere from $5m to $75m with the median* at $35m. The senior managers couldn't cope with that. They wanted to know "the cost" or the "maximum cost" (which in theory would be infinite, $75m was the 95th percentile) and couldn't deal with a statistical distribution of outcomes. It's the same with elections. Pollsters usually tell you the "average" outcome that their poll predicts, and they tell you the error in the "average" due to sample size. They don't tell you the shape of the distribution. They don't tell you if their average is a median or a mean. They don't admit to any bias in their sampling (they try to correct for it... often unsuccessfully). They don't tell you if it's very "peaky", or "flat" with a much higher probability of results far away from the average value. They don't tell you if it's a roughly symmetrical distribution, or skewed with a lot more results to one side of the peak. So, take all "single answer" polls with a large pinch of salt, and understand that the "answer" is only a point on a distribution of possible indications.
*median is the 50th percentile value of a distribution, that is 50 percent of possible outcomes are less than this value, and 50 percent are more.
I remember an incident from my career in forecasting (not elections) that might help: my forecast said that the cost of a contract could be anywhere from $5m to $75m with the median* at $35m. The senior managers couldn't cope with that. They wanted to know "the cost" or the "maximum cost" (which in theory would be infinite, $75m was the 95th percentile) and couldn't deal with a statistical distribution of outcomes. It's the same with elections. Pollsters usually tell you the "average" outcome that their poll predicts, and they tell you the error in the "average" due to sample size. They don't tell you the shape of the distribution. They don't tell you if their average is a median or a mean. They don't admit to any bias in their sampling (they try to correct for it... often unsuccessfully). They don't tell you if it's very "peaky", or "flat" with a much higher probability of results far away from the average value. They don't tell you if it's a roughly symmetrical distribution, or skewed with a lot more results to one side of the peak. So, take all "single answer" polls with a large pinch of salt, and understand that the "answer" is only a point on a distribution of possible indications.
*median is the 50th percentile value of a distribution, that is 50 percent of possible outcomes are less than this value, and 50 percent are more.
Thank you, Louise, for your helpful perspective!