My second reaction: they lied to subjects about the purpose of the experiment and didn't fully inform respondents about what the surveys were about; their human ethics committee is far more awesome than others.
My third reaction: I hope this gets published despite that many of the main effects are statistically insignificant.
I've read a bit of the literature on preferences for homogeneity. It depresses me, but I'd hoped that longer exposure to more diverse groups would overcome some of those effects. And as it's hard to tease out causality in population based measures, you could argue the toss on it.
Ryan Enos at Harvard's Department of Government approached things a bit more scientifically. Here's his experimental design. Get a bunch of train stations in ethnically homogenous white areas of Boston. Half of them are controls. The other half get a pair of obviously ethnic Hispanic confederates who are going to ride the morning commute for a week while chatting with each other in Spanish - other commuters could assume that a Hispanic family had moved into the neighbourhood and were commuting into town. Pre-survey people at all stations and get them to give an email address for follow-up surveys. Among the questions, ask about attitudes towards immigration. Half of respondents were invited to answer the survey again three days after treatment began; the other half were invited to do so ten days later.
- Ideological conservatives' views did not much respond to treatment, but remained hostile towards immigration. Liberals and moderates became more hostile to immigration as consequence of treatment. But results were not always significant.
- Negative responses were sharper in the three day than in the ten day treatment. But the 95% CI around the effect only failed to include zero for one survey question and only in the three-day treatment: "Should the number of immigrants be increased?"
- Income didn't moderate the effects: the same negative shift in attitudes was pretty uniform across respondent income.
- Treated respondents were more likely to say their neighbourhood had recently become poorer, more diverse, and more unsafe. But the confidence interval is wide.
- Effects were largest among those with low numbers of Hispanic friends; for those with few Hispanic friends, effects were significantly negative on all three questions.
- There was no change in policy attitudes other than those associated with immigration.