[domestic election] observers reduce the probability of overvoting (more votes cast than registered voters) at observed stations by 60 percent. They also reduce unnaturally high levels of turnout and, in political party strongholds, instances of ballot stuffing.
…observers induce the strategic relocation of fraud to nearby but unobserved polling places. Displacement is concentrated in the strongholds of Ghana’s two major political parties. Displacement decreases as the share of observed polling places in a constituency increases, and at the highest level of observer intensity is reduced to zero even in party strongholds.
Results from an experiment around the 2012 election in Ghana.
One of the nicer aspects of the paper is they look to see whether observers just move the fraud around or not.