This is whole thread a useful framing. So many debates in metascience have parallels in broader discussions about promoting and balancing prosocial behavior, autonomy, efficiency, policing, values etc…

Been pondering what bothers me abt this reasoning. I think it’s the presumption that avoidance of fraud is the first and biggest priority. It is a priority but not the biggest and not the sole. Consider an analogy with welfare state. If avoidance of benefit fraud and inefficient spending… 1/2

It’s a bit like putting all of the daily essentials like deodorant behind glass at CVS. Congratulations on preventing crime. I also no longer want to shop here.

Shoplifting deterrents deter in-store shopping with items behind locked cabinets

Locking up merchandise at stores hasn't curbed retail theft but is driving consumers to shop online.

Not that I’m aware of. That’s also harder to do well than most people realize.

DGCS: don’t go crosstab spelunking

Look, polls have biases. But slso, the MOE for a subgroup of n=96 is about +/- 10 points. Add in the design effect, it’s +/- 13 points. For the vote margin it’s +/- 26. For the change in the margin since the last poll, it’s about +/- 50 points. That’s just accounting for random sanpling.

And critically, those questions aren’t answerable until after the election.

Absolutely worth investigating. But also requires comparing not specific pollsters but polling methodologies. Getting a serious answer to these kinds of questions is both a) important and b) much more difficult than most people realize.

I think it really depends a lot on which polls/pollsters you are comparing it to. A lack of variability in some subgroups could likewise be indicative of a problem (e.g. using a method that has low variance but high bias).

I think it’s fair to criticize reporters who report something that is not actually statistically supported by the underlying data (idk if that’s the case in the poll you linked). But I’d encourage people to avoid armchair meta-analysis.