Dont use "levers". Just say, first we consider "legal accountability". then say despite very detailed legislation, and easy access to courts and a litigation culture, biased polls are not challenged in court. Suggesting that the source of the bias is hard to prove in court. ... Second, we consider reputation. Then explain why pollsters would have a reputational incentive (if pollsters are known to produce biased polls they would not be trusted by voters---so there is no reason to pay fora pollster with a reputation for biasing in favor of the one who pays.) We find evidence consistent with such a reputational mechanism at play. then explain the large pollsters , media etc . and also that bias is smaller where there are multiple independent polls (making the bias more obviouss)

How can bias in polls be reduced?

Just end with "this suggest that the source of bias is subtle sources of bias --- sample fram composition , ... ---that are not possible to detect from the formal survey representation"

Delete this part about firm selection

Here we should say "Another explanation could be survey design ....". dont say "channel A" never in paper.

Not sure if I like this test

Is this too strong?

just say "One explanation could be ...". then use plainer words ("post-field editing" is hard to read)

delete this. just start "I assemble ..."

Is this really important to keep in intro? Would it not be sufficient to say sponsor bias has not been documented before? I thought we concluded this was not done

Either dont say 2024 and municpal elections here (just "Brazil") or we must explain why 2020 brazil, state elections in brazil etc wont offer the same opportunity.

Maybe just skip this, and just state that the literature does not answer this yet (backed up with a summary of our literature reivew in a footnote)

maybe replace with primary input to voter beliefs and donor decisions (drop campaign strategy). Would be maybe nice witha footnote explaining why, especially stuff that indicates that strategic voters tend to focus on those highest on the polls + donors wanting to donate to winners. Ideally we should say something here that convinces readers that polls are important and can actually influence the outcome of elections (maybe do a search for any literature claiming this we can cite in the footnote).

say "AND the polls are not legally challenged, suggesting that the bias works through subtle channles not easily detectable or proven in court". Then "Large-volume firms .... , suggesting that reputational converns of pollsters could limit bias." then "Overall, the analysis suggests that informing voters of pollster bias through mandatory disclosure could be ..."

maybe say: "compared to the actual electoral outcomes and to independent polls sruvey both the weeks before and the weeks after" and drop "the estimate holds ...."

I think this formulation is awkward. any more elegant way of stating it formally without the i and j stuff?

Why must it learn two mappings? maybe this statement is wrong

this could depend on settings. maybe for some settings the evidence mapping is very easy, but not a simple mapping from types. so make this less strong statements