From Science Magazine:
Brian McCabe and Jennifer Heerwig, sociology students at New York University in New York City, used statistical tools to unmask covert attitudes. They asked their control group of 500 people point blank whether they would vote for an African-American president, and 84% said yes—a lower percentage than other polls. McCabe and Heerwig attribute the difference to greater privacy: Their survey was taken over the Internet. They asked the second group three questions unrelated to race, such as whether presidential campaigns are too costly. The third group was asked those same three questions, plus a fourth about whether they’d vote for an African-American candidate. McCabe and Heerwig [asked] only how many statements the respondents in those two groups agreed with, not which ones. “If you have a socially undesirable answer, you can hide it,” says McCabe. With some statistical analysis, they determined that assuming the three groups were similar, 14% of their first cohort lied, and in truth just 70% would support an African-American. Democrats were much more likely to be dishonest than Republicans, proffering up the socially desirable answer, as were those with less education.
Very nice work. I agree with the authors on their assertion that this might not matter that much for the actual election in a month, so this gets the tag Science and not Politics.
Here’s the original paper. Here’s an interview with the author of the paper.
6 comments ↓
[...] ← Study of the Day [...]
Clever study and interesting results. I have a couple thoughts about it though:
1. The p-values are calculated with respect to the null-hypothesis that the subgroup has 0 social desirability bias. Comparisons between subgroups aren’t evaluated for significance. Is the bias of 0.19 for liberals really different from the conservatives at 0.12?
2. the study assumes the immigration, campaign finance, and Iraq questions have no desirability bias (or the effects cancel). However, I would argue all three of these questions have have a slight social bias for the “yes” answer. For example, politicians have found it expedient to support “more secure borders” even if they are pro-immigration. Some people, therefore, may voice a public support for more strict immigration laws, while privately believing they are too strong.
3. According to their numbers (overt support * proportion true support), potential support for a black candidate is about the same for liberals and conservatives (liberals: .89 * .70 = .62, conservatives: = .85 *.73 = .62)
1. Agreed.
2. I think you are wrong. You don’t need to assume anything about the 3 hypothesis, except that the two groups, on average, agree with each other.
3. I interpreted true support as the percentage of people who truly support the proposition, not the percentage of people who were speaking truthfully. That is, there is no multiplication. Your calculations flatly contradict their statement that conservatives have more true support than liberals.
3. Looking at it again, I think you are right. I misinterpreted “proportion true support” as the proportion of true support among people who claimed they would support a black candidate.
2. I still find some minor problems in comparing the study groups. The immigration question is particularly problematic. Immigration controversy in the US is about Mexicans, another minority group. So 2 out of 4 group C questions are related to sensitive race issues. While in group B, only 1 in 3 have this property. What is that effect?
Consider a person with some racial bias and a guilty feeling about it. Faced with group B questions, he answers no to the stronger immigration question (to feel better about himself). But faced with group C questions, he may alleviate his guilt by saying “I’ll vote for a black candidate”, and then feel more comfortable supporting stronger immigration laws. [this would underestimate bias.]
Now consider a person who is very angry about immigration. If this person is in group A, he is asked one question: “will you support a black candidate”. In isolation, the person approves. But if he is enrolled in group C, the immigration question may prime the person’s response to the “black candidate” question. (maybe he assumes a black candidate would be liberal and support minorities and maybe illegal immigration.)
I’m probably splitting hairs. I like the study, but I’d like to see it repeated with a better choice of control questions.
to be concise: the study assumes that presenting the 4th question does not change the way people answer the other 3, and vice versa.
There’s reason to worry about this with the Immigration question.
Of course, it is impossible to show that the answer to the 4th question isn’t influenced by the answer the 3 first questions in a party-dependent way.
Leave a Comment