A reminder, if a poll shows a candidate being +2 or -1 since last month, it means nothing's changed [View all]

[Godhumor]

Last edited Fri Dec 11, 2015, 12:11 PM - Edit history (2)

I'm not going to go through the real way of checking the stability of numbers in a poll month over month, because it requires a fair bit of calculation across multiple polls. However, there is a white lie stand-in that is almost as good.

Basically, look at a poll's margin of error. If a candidate's change in support, in either direction, is less than or equal to the MOE, the change is not significant. If it is more than the MOE, then there is some real change occurring (though maybe not as much as you think.).

For example, A poll with a 5% MOE has Clinton support at +3 since last month, Sanders at -2, and O'Malley at -1. All three candidates have changes within the MOE range, so, the race would be said to be stable from last month, as it is possible no actual change occurred.

Another poll has an MOE of 4%. Sanders is +6 since last month, Hillary is -3 as is O'Malley. Sanders has significant change while Clinton and O'Malley don't. However, with the MOE stand-in, Sanders real change could be as low as +2 (His net improvement minus the MOE). Furthermore, HRC and MOM both have nonsignificant changes. The instinct is to say Sanders benefited from both of them equally, but, due to their insignificant changes, we don't actually know if that is true. It is possible he gained from both, or from just one or from neither (Could be he picked up some undecided voters.).

Be careful interpreting trends month over month. In reality, this race has been ridiculously stable the past two months. Almost boringly so.

On edit: if you're trying to interpret something like Reuters/Ipsos, just set your computer on fire and walk away.

On 2nd edit: The white lie had been called out, so let's add an addendum. To see what portion of a month over month change may be significant, you can take the absolute value of the MOE and subtract out the change, or vice versa. That gives you the total net change that might be significant depending on where in the MOE values actually fall. Since we don't know where in the MOE the values are, this is a dubious calculation. That is why the White lie to knowing if change is significant is if it is completely outside the range of the MOE.