seems not clearly better than normal polls, but there are several considerations

problems with neighbor polls:

• one side might be more vocal about their views, giving people skewed perceptions
• one side might have larger families/friends on average; a random sample of someone’s friends is not a random sample of the population
• you might get exaggerated results, e.g. if 70% of people are voting Red, 100% might say they expect their friends to vote Red

good things though:

• people might be more truthful
• might sample the population better because you’re not just getting data on people who answer calls from unknown numbers

sometimes they’ve been more accurate, sometimes less