1. The term "oversampling" re: polling is almost always misused. Truthfully, even pollsters help to perpetuate that misunderstanding because it's so commonly used, even by Corporate Big Media.
A certain group absolutely can be "overrepresented" or "underrepresented", but...
A certain group absolutely can be "overrepresented" or "underrepresented", but...
2. but... it's not the sample thing as oversampling, which for instance, is what we do in Florida re: Hispanics to ensure we collect enough diverse opinions among subgroups of Latinos.
Continued...
Continued...
3. A group is "over" or "underrepresented" typically B/C the pollster isn't willing or doesn't want to diversity data collection modes. This isn't speculation. It's true. Different groups take surveys at different rates depending on the mode of collection, i.e. cell, IVR, online.
4. The same is true of the interview type, i.e. whether it's live-caller or anonymous. Some of this is at a very basic level, some not so basic.
Women are more willing to interview than men, even more-so over the phone. Their participation rates often mean the raw is 60/40.
Women are more willing to interview than men, even more-so over the phone. Their participation rates often mean the raw is 60/40.
5. I can go on and on about this. But bottom line, Big Media believes and professes live-caller is the "Gold Standard." As most of you know, I disagree. I prefer mixed-mode collection on almost every project to ensure the opinions are truly representative.
Continued...
Continued...
6. Is it ever the case some modes turn out to be more accurate than others? Sure. But my experience and research - both at BDP & partners - indicate mixed-mode is almost always a better approach, more "representative," than just one mode w/ it's own participation bias.
Continue.
Continue.
Otherwise, you end up with Donnelly+7 in Indiana because you called too many people on their cell phone in Indianapolis.
8. Just to drive the point home, because you know how much I love visualizations, this is a screenshot from a survey last night in Florida.
The district is 54% female and 46% male.
But look how much higher than participation rate is for women. This is very, VERY common.
Cont.
The district is 54% female and 46% male.
But look how much higher than participation rate is for women. This is very, VERY common.
Cont.
9. We only call in the evening on weekdays while online panels run all day. But the online panel is about 55.5%/45.5%, much more representative of the district, itself. True, we can and do micro target the panel while the phone is randomly selected from voter files.
But still.
But still.
