One of the problems with poll data is, that while the mathmatics says that a poll of 1,000+/- may be a statistical representative sample of the 300+ million of us, the details of who (and where) those thousand or so people are is vital to the accuracy of the sample.
And that also leaves out one other slightly important factor, the honesty of the people responding. For the polls showing how gun ownership has declined (number of households with a gun,etc...), I doubt every one. Nobody with two brain cells to rub together is going to tell an anynonomous voice on the phone that they own guns. That alone is going to seriously skew any poll on the subject.
If I were to poll a thousand people on the streets of New York City about any gun related issue, I'm sure I'd get quite a different result than if I polled a random thousand from Montana or Texas. Just as if I polled a thousand from the NRA's membership list, vs. a like number of people who donated money to Coalition to end gun violence (or whatever they are calling it this week).
The devil is in the details, and the only poll that really matters is held every 2 or 4 years in November.
All else being equal (and it almost never is) bigger bullets tend to work better.