"Statistical dead heat" is a dumb term

[cyberianpan]

I notice on P.ie and all over the media people use the term "statistical dead heat".

So if a poll has a margin of error of say +/- 4% and Candidate A has 45% and Candidate B has 48% the commentators say the race is a "statistical dead heat" ... what on earth do they mean ? That race is not a dead heat:

International Statistical Institute, 53rd Session 2001: Douglas Lonnstrom
With the worldwide proliferation of political polling, especially in the United States, the mass
media has coined the term ‘statistical dead heat’. The intended meaning is to call the race a tie if the
two candidates have percentages that fall within the sampling error. For example, if we have a sample
size of 600 the sampling error is +/- 4 percentage points. Therefore, if candidate A has 45% of the vote
and candidate B has 41%, the media is calling it even. I say this is not so. In fact, it is my contention
the media makes four errors when they use the term ‘statistical dead heat’. There certainly can be a tie
if each candidate has the same percentage, i.e. 45% to 45%. In any other case the candidate with the
higher vote is more likely to be ahead, even if the difference is only one percentage point.

ERROR 1 – Let’s start with the extreme case: A has 45% and B 41%, this is within the sampling
error margin for a sample of 600. Using the recognized statistical calculation called ‘power of the test’
(Berenson, 1996), the probability that B is tied with or ahead of A is 2%. Hence, there is a 98%
probability A is in the lead. Now, test the least extreme situation. Even though A is ahead 45% to 41%
assume the tie occurs at 43%. The ‘power of the test’ still shows A is the likely winner, although the
probability drops to 84%. A third scenario would be to have the poll show only a one point spread,
45% to 44%. Even then the ‘power of the test’ produces a probability of 69% that A is the leader. In
short, you would rather be ahead in the poll, no matter the margin, then behind.

ERROR 2 –...

Blog

And just in case Salon isn't convincing for you, here's the National Council on Public Polls.
Certainly, if the gap between the two candidates is less than the sampling error margin, you should not say that one candidate is ahead of the other. You can say the race is "close," the race is "roughly even," or there is "little difference between the candidates." But it should not be called a "dead heat" unless the candidates are tied with the same percentages. And it certainly is not a “statistical tie” unless both candidates have the same exact percentages.

Also see Illustrative table

I think that this is a stunning indictment of the chattering classes: why are they using terms that make no sense ?

It appears to show a fundamental misunderstanding of probability ... why do we even bother with opinion polls ?

cYp

[wombat]

Opinion polls use statistics but there are major differences between sampling washers and asking people how they will vote. Most sampling yields a yes/no answer whereas opinion polls require interpretation. Very few people who comment on polls are trained in maths, never mind statistics which is a specialist branch of the subject so you have to put up with imprecise descriptions.

[cyberianpan]

Opinion polls use statistics but there are major differences between sampling washers and asking people how they will vote. Most sampling yields a yes/no answer whereas opinion polls require interpretation. Very few people who comment on polls are trained in maths, never mind statistics which is a specialist branch of the subject so you have to put up with imprecise descriptions.

I know you understand area but It's not just that the term is imprecise... it's wrong !

Here look at these percentage leads[size=7]*[/size] (going across top, left to right) and see how probable it is that a candidate is actually in the lead for different margins of error:



Current in US politics many national polls have +/- 4% and often one candidate is 4% ahead: it is 85% likely that candidate is actually ahead !


cYp

[size=7]* they are simplified with no 3rd candidate and no unknown voters[/size]