It is possible to have both voter privacy & traceability. Here is just one academic article (among many) describing how you can use bisimulation & graph theory to mathematically prove the correctness of privacy properties of electronic voting protocols is here: http://www.lsv.fr/Projects/anr-avote/PUBLIS/DKR-jcs08.pdfhere.
The cryptography behind the scenes is quite complex. The protocols can provide plausible deniability. It is not decentralized, voters need to have a special token (e.g. an electronic card or something) emitted by a central authority (which we assume you can trust).
Also, the procol assumeprotocol assumes the use of voting booths, so there is no remote voting. The act of casting the actual vote is done securely and hidden from malicious eyes.
When the election completes, all the votes & protocol messages are made public, not just the tally. Any voter can then verify that their vote is present and is correct. They can prove whether it is missing or incorrect using their token.
(Right now I don't have time for a more elaborate answer. I suggest you take a look into the link article for the details for now).
