All robo-call polling is Useless

[cthulu2016]

It is illegal to robo-call a cell phone.
__________________________


In other news, as of October 1 Gallup (not a robo-caller) started using samples of 50-50 land line and cell phone. This was an increase in cell phones from the previous 40%-60%.

(Presumably, if the night's quota for land-line or cell is full you start ending further surveys when you find out they have the 'wrong' kind of phone, until you have the number of each requested.)

This does NOT however mean that cell phones are suddenly having a larger effect on Gallup Results. No notable change in results would have happened on October 1.

Cell phone vs. land-line is a weighted demographic category. A pollster has a number, based on all sorts of data, of the % of cell-phone usage, and that weight is applied to the set of results.

If cell phones are 48% of phones and your sample is only 40% cell-phones then you give those cell phone calls a greater weight in the results. Whatever that 40% of calls said is counted as 48% of opinion. (And the land-line calls are reduced in weight.)

But you don't want to always be making large weighting adjustments if it can be helped. Ideally you want your sample in the ballpark, and only needing a slight weighting adjustment.

So Gallup seeking 50-50 samples suggests that the current mix of random dialing to households is currently about 50-50, cell and land-line.

_____

http://livewire.talkingpointsmemo.com/entry/gallup-increases-cell-phone-respondents-in-tracking-polls

[unblock]

1. some interesting statistical challenges here. e.g., what about those with both types of phones?

while it's highly unlikely, given the tiny sampling rates, that any particular household be called both on a cell phone and a landline, such households are nevertheless overrepresented using this method unless a further adjustment is made.

whatever metric you have describing the demographics of cell phone owners and demographics of landline owners, you are double-counting any household with both, which presumably are the more affluent households.

[cthulu2016]

2. That's an interesting point

With a national sample of 1,000, the odds of calling the same household twice in a survey is negligible.

But every two-phone household is, as you note, twice as likely to be called by pollsters in general.

Very provocative. Thanks for the observation.