So first, we walked up to people, we asked them about their voting intention and then when they started filling in, 首先，我们找到一些路人，询问他们的投票意向，然后当他们填写的时候，

we would fill in a set of answers going in the opposite direction. 我们会填写一份相反的答案。

We would put it under the notepad. 并放在写字板的下方。

And when we get the questionnaire, we would simply glue it on top of the participant's own answer. 然后，当我们拿到填好的问卷时，会直接把它粘到参与者自己的答案上面。

So there, it's gone. And then we would ask about each of the questions: 于是乎，它不见了。然后，我们会再问他们这几个问题：

How did you reason here? And they'll state the reasons, together we will sum up their overall score. 这里你给出的理由是什么？然后他们会陈述理由，同时，我们还会算他们的总分。

And in the end, they will state their voting intention again. 最后，他们还会再次陈述自己的投票意向。

So what we find first of all here, is that very few of these manipulations are detected. 那么，我们首先了解到的是，这些小把戏很少会被揭穿。

And they're not detected in the sense that they realize, "OK, you must have changed my answer," 即便被发现，他们也不会觉得，“好吧，你肯定是换掉了我的答案”

it was more the case that, "OK, I must've misunderstood the question the first time I read it. Can I please change it?" 更可能是这样，“好吧，我第一次读题目的时候一定是误解它了。我可以换回答案吗？”

And even if a few of these manipulations were changed, the overall majority was missed. 即便部分被篡改的答案被改回来了，总的来说，大部分还是被忽略了。

So we managed to switch 90 percent of the participants' answers from left to right, right to left, their overall profile. 我们成功替换了90%参与者的答案，从左翼到右翼，从右翼到左翼，他们整个的概述。