Poll questions are often worded and/or contextualized in a way that highly influences the immediate answer.
I had a poll call recently where the pollster cited crime statistics and community involvement by police, then asked something like "do you want deprive law enforcement with the funding it needs to keep the community safe"?
I explained to the guy on the phone that "keeping the community safe" requires reallocating LE money to social support services, but of course that was not a response option.
Most people asked that question would probably just say "no" because obviously nobody wants their community to be unsafe.
Even assuming the questions are fair, the results can be presented in a way that would not be consistent with the desires of those polled.
For example, according to a
Marist poll cited by 538, "Medicare for All, replacing private insurance" polled at 41% overall, whereas "Medicare for All who choose it, allowing private insurance" polled at 70% overall.
One could report from this poll that M4A has 70% support with misleading accuracy.