Report margin of error when reporting poll numbers
eliaslevy opened this issue · comments
Statements like
Ted Cruz leads Beto O’Rourke by one percentage point in their Texas Senate race. Cruz leads O'Rourke 38 to 37%.
are misleading without reporting the margin of error of the poll. This particular poll had a 4.4% margin of error. Ted Cruz would be ahead 42.4% to 32.6%, or O'Rourke could be ahead 41.4% to 33.6%.
Would also be good to indicate that a single poll is seldom a good representation of the race, because of different biases inherent in how polls are structured and it is best to look at aggregated polls.
I hear you and it makes sense. I probably won't ever get into these specifics, however, because these are written as abstracts. Polling is inherently flawed, so I include these are basic directional information rather than any kind of scientific fact. lmk what you think
I agree that as summaries, it is not appropriate to go into too much detail, but simple adding "± 4.4%" to the poll numbers is relatively simple. E.g.:
Ted Cruz leads Beto O’Rourke by one percentage point in their Texas Senate race. Cruz leads O'Rourke 38 to 37%, ± 4.4%.
Or better yet, since either one could be leading given the margin of error, it would be better to word it differently:
Texas Senate race: Ted Cruz 38 ± 4.4%, Beto O’Rourke 37 ± 4.4%.
Eh but margins of error are inherently meaningless without the corresponding confidence level, which is never reported. It's probably always 95%, but still, nobody reports polls well.