Learn some statistics people.

[Teaser]

If I have to read one more post wondering "What caused" some minor fluctuation in a poll overnight I'm going to grab my lazer tag gear, find the highest clock tower on campus and...

Fluctuations in polls happen. Usually for no reason.

Repeat it with me:

Fluctuations in polls happen. Usually for no reason.
Fluctuations in polls happen. Usually for no reason.
Fluctuations in polls happen. Usually for no reason.

[JuniorPlankton]

1. Yep

It's just process error.

[RafterMan]

2. Yes, but

fluctuations also happen in polls for reasons. Are you so confident you know the difference?

Ease up a little.

[MissouriTeacher]

3. OMG! The 'nuisance' comment is killing our campaign!!!

:eyes:

[endnote]

4. Especially if the fluctuations are well within the margin of error!

Anything within the margin of error, by definition, is noise! it can go a bit up or a bit down... it means nothing...

[SnowGoose]

5. Margin of Error

What I wish people had a better grasp of is the meaning of "margin of error". It does not mean what it would appear to based on lay terminology. It does NOT (!) take into account nonsampling errors.

Put another way, this thing about not surveying people who use cellphones exclusively is not accounted for in the MOE. The MOE assumes that the sample of people you survey is representative of the whole population you wish to study. Given that HUGE assumption, the MOE describes the likelihood that, just by chance, you happened to pick people out of the population who, when aggregated, produce a measure that differs from the overall population. That's why when you have a larger sample, you tend to have a smaller MOE: because the chance that the sample differs from the population goes down as you gather a bigger sample.

But bad design (including question wording, method of communication, time of day, order of the questions, and a million other things) which alters the results is not accounted for in MOE.

EOR
(end of rant)