Here's Abhijit Banerjee et al describing the results of a study:
During last year’s Delhi elections, SNS, an NGO, filed a series of Right to Information cases in order to extract information about the performance of the sitting MLAs. This included how they had spent their annual Rs 2 crore development funds, as well as how often they had participated in the committees where they are supposed to represent their constituents. Hindustan combined this information with data from the mandatory disclosures by all candidates regarding criminal history, education, and personal assets, and printed the same in a concise format for a set of ten constituencies, in the week before the election.
A group of eight NGOs then distributed free copies of these newspapers in 200 polling stations, chosen by lottery to ensure that they were comparable to all the others where the newspapers were not distributed. While candidates with serious criminal records got about 9 per cent more votes in general, the newspaper distribution appears to have effectively cut the electorate’s preference for criminals in half. However, one could make the case that this effect is the result of priming: it is conceivable that the newspaper campaign simply made people more sensitive to this issue.
The piece also discuss some other findings about the role of caste in UP politics -- especially in the perceived rise of criminalization since the 1980s. Here's the abstract of the academic study behind these findings:
This paper examines how increased voter ethnicization, defined as a greater preference for the party representing one's ethnic group, affects politician quality. If politics is characterized by incomplete policy commitment, then ethnicization reduces average winner quality for the pro-majority party with the opposite true for the minority party. The effect increases with greater numerical dominance of the majority (and so social homogeneity). Empirical evidence from a survey on politician corruption that we conducted in North India is remarkably consistent with our theoretical predictions.