方舟子:《总统辩论谁赢了?》

老闹子

知名会员
VIP
注册
2009-03-07
消息
4,178
荣誉分数
853
声望点数
123
总统辩论谁赢了?
方舟子 10月04日 09:43

http://fangzhouzi.baijia.baidu.com/article/645903
再多的民意调查结果也不会让川普承认自己输了,他本来就是极度自恋的人,从不认输也从不认错。而川粉对川普是一种类似邪教的追随,川普认为自己赢了,他们也就只能跟着喊他赢了

今年是美国大选年,将在11月8日投票选出新一届总统。按惯例两个主要政党的总统候选人要举行三场辩论。今年的第一场辩论在9月26日举行之后,几乎所有美国媒体请的裁判、评论员都一致认为民主党候选人希拉里大胜共和党候选人川普。但是辩论的胜负判断本来就是很主观的,即使媒体一致认为希拉里获胜也不会让川粉(川普的狂热支持者)接受,因为他们本来就认为美国主流媒体基本上都是自由派,都是支持希拉里的。举行辩论的主要目的是要影响选民,那么,我们可以给胜负定一个客观标准,那就是认为哪一方获胜的选民人数比较多,这一方就获胜了。第一场辩论刚刚结束,CNN就立即报道了他们通过一家民意调查机构ORC对521名选民做民意调查的结果,62%认为希拉里赢,27%认为川普赢,据此可以说希拉里大获全胜。

但是网上有很多对这次辩论结果的投票,有的票数几十万、几百万,都是一面倒地认为川普赢。川普就因此宣布自己赢了,说除了“没人看的CNN”,所有的“民意调查结果”都是他赢得了辩论,并嘲笑说CNN只对300人(其实是500多人)的调查怎么能算数呢。川粉也坚信网上投票结果,跟着川普嘲笑CNN的民意调查人数太少,甚至有一家一直力挺川普的微信公众号还抱怨人们想去CNN网站投票找不到门,大骂CNN。

其实那种网民自己去投的网上投票的结果才是完全不可信的,即使有几百万、几千万票数,也不可信。首先是网上投票很容易作弊,一个人可以反复地投,也就是刷票。有的网站为了避免刷票,限定一个IP只能投一次票,但是通过代理服务器是可以随便换IP的,这种限制难不倒懂点网络技术的人,何况还有现成的刷票软件可以自动换IP刷票,想刷多少票都可以。有一个关于支持谁当总统的网络投票有高达一千七百万次投票,95%是支持川普,谁会傻到去相信这样的投票结果没有作弊?即使投票的网民都很诚实,不作弊,网上投票结果也不可靠。支持者会到处拉票,比的是哪方支持者更狂热更爱拉票。川普支持者往往比希拉里支持者狂热、偏执,川普支持者就被发现在某些论坛拉票,列出了各个投票网站的链接号召川普支持者去投票。再退一步说,即使某个网站的网络投票没人作弊也没人拉票,其投票结果也只是反映该网站读者的意见,而不能反映全体选民的民意,因为一个网站的读者往往是有倾向性的,例如美国有一个著名的极右网站Breitbart是一直支持川普的(其主编是川普竞选团队的CEO),该网站的读者就基本上是支持川普的,投票的结果当然是大部分人(约75%)认为川普获胜,其实我们应该惊讶的是其读者中居然还有25%认为希拉里获胜。

所以网上投票说难听点是垃圾,说好听点是好玩,不能当真。想要知道民意,就要做民意调查。民意调查要由组织者通过打电话、发邮件等方式主动找人询问(以打电话为主,但是现在很多人都不接陌生电话了,所以也有网上调查,但是是通过发邮件邀请的方式,不是网民自己找上门),这样就可避免调查对象重复投票、拉票、倾向性等问题。如果要调查的人群很小,挨个挨个去问就可以了。例如对“你认为总统辩论谁赢了”这问题,你想统计一个班的学生的看法,让每个学生依次回答就可以了。但是如果你想了解全校学生的看法,假定一个学校有三万名学生,一个个去问显然是不现实的,即使能做到,工作量也太大了。在这种情况下就可以采取随机抽样的方式,只找一部分人做调查。如果找的人数太少,没有代表性,结果容易出现偏差;但是如果找的人数太多,则是浪费。那么找多少人合适呢?或者说,需要多大的样本呢?

样本的大小和几个因素有关。一个因素是置信度,它表示的是你对调查结果的信心。因为不是对所有的人做调查,置信度不可能是百分之百,但是如果调查了足够多的人数,你就会对调查结果的可靠性很有信心。一般取置信度95%,意思是有95%的几率是这个结果。另一个因素是抽样误差,据此可以知道调查结果的正负变化范围(所谓置信区间)。可接受的误差越大,需要的样本就越小,反之,可接受的误差越小,需要的样本就越大。民意调查可接受的误差通常是2%~5%。“你认为总统辩论谁赢了”是个比率问题(支持希拉里和支持川普之比),在调查前我们是不知道比率的,这时候可以设置为50%,因为在比率为50%时需要的样本最大。知道了人群总数(总体)、置信度、误差和比率这些参数之后,统计学上就有公式可以计算需要的样本大小。网上有现成的民意调查计算器(例如http://www.surveysystem.com/sscalc.htm ),把这些参数输进去,就可以算出样本大小。假定总体有3万人,置信度95%,误差5%,算得样本大小为379。也就是说,你要知道3万名学生对“你认为总统辩论谁赢了”这个问题的看法,随机抽样379人做调查就够了,调查结果有95%的可能是成立的,有5%的误差——假如调查结果是50%的人认为希拉里赢,意思是如果重复调查20次,有19次的结果是45%~55%的人认为希拉里赢。

如果我们想要知道一个30万人的城市对这个问题的民意呢?其他参数不变,把总体改成30万,算得的样本大小是384,只需要多调查5个人。如果要调查的是全美国人口呢?美国人口大约是3亿人,算得的样本大小还是384。在总体足够大以后(几万人以上),需要的样本大小基本不变了,调查全国和调查一个城市需要的样本大小其实是一样的,这虽然违反直觉,但是事实就是如此。

所以,要对全美国做民意调查,随机抽查300人也是可以的,川普、川粉的嘲笑只不过暴露了他们对统计学的无知。不过,抽查300人的误差有些大,全国性民意调查的误差一般控制在3%左右,因此需要的样本大小通常是1000人左右。CNN/ORC调查了521人,误差是4.5%,误差稍大,但是因为调查的结果比率悬殊(62%:27%),这个误差并不会改变希拉里大胜的结论。CNN/ORC调查采用的是简单随机抽样的方式,这样抽得的随机样本不一定能很好地符合人群分布特征,例如其中认同民主党的人数偏高。所以有时候为了获得能代表人群分布的样本,会采取分层抽样的方式,先根据人群特征设定各个群体的比例,然后再在各个群体随机抽样。

因为是抽样调查,即使精心地设计,也不能保证每次调查结果就一定可靠。但是如果有别的调查机构也做了类似调查,那就可以相互验证。如果各个调查的结果都一致,就说明非常可靠。CNN是第一个公布总统辩论的民意调查结果的,随后其他民意调查也都得到了类似的结果:NBC News的结果52%:21%,ppp的结果51%:40%,YouGov的结果49%:22%,Politico的结果49%:26%,Echelon Insights的结果48%:22%,路透社的结果:48%:23%,福克斯的结果61%:21%,Post-ABC的结果53%:18%。所有的调查结果都认为希拉里赢了,而且除了ppp的调查,都认为希拉里大获全胜,认为希拉里赢的选民人数是认为川普赢的选民人数的2~3倍。不知为何川普、川普支持者只顾着骂最先出头的CNN,而无视其他的调查结果?特别是保守派电视台福克斯一直是支持川普、获得川普的信任的,其民意调查结果比分比CNN还悬殊,认为希拉里赢的人数是认为川普赢的人数的3倍,川普、川粉怎么不去骂福克斯?

当然,再多的民意调查结果也不会让川普承认自己输了,他本来就是极度自恋的人,从不认输也从不认错。而川粉对川普是一种类似邪教的追随,川普认为自己赢了,他们也就只能跟着喊他赢了,然后认为所有的民意调查结果都是不可靠的,都是被民主党收买的,只有他们刷出来的网上投票才能代表民意。

2016.10.3
 
再多的民意调查结果也不会让川普承认自己输了,他本来就是极度自恋的人,从不认输也从不认错。而川粉对川普是一种类似邪教的追随,川普认为自己赢了,他们也就只能跟着喊他赢了.

:dx:
 
转一个方舟子刚发的推特
====================
新语丝之光001
30分钟前 来自 手机微博触屏版
中医粉说我拿了美国的好处,反转控说我拿了孟山都的好处,川粉说我批拿了民主党的好处,还有的干脆说我是为了能在美国吃福利,想象力都不错,我拿的好处真多。幸好还没说我是非法移民怕被川普遣返。 @自由满洲 民主党给了你什么好处 还不如接着骂韩寒//@方舟子:再多的民意调查结果也不会让川普承认自己...
 
再多的民意调查结果也不会让川普承认自己输了,他本来就是极度自恋的人,从不认输也从不认错。而川粉对川普是一种类似邪教的追随,川普认为自己赢了,他们也就只能跟着喊他赢了.

:dx:
教主总结自己还是很到位的。
 
表面看起来,方舟子和川普有某些相似的地方:他俩都很坚持自己的主张,都有很多反对者和很多支持者;他们面对的对手情况也很相似,也同样有很多的反对者和支持者。以方舟子和崔永元、韩春雨等的争斗为例,如果也弄一个民调来调查中国网民对他们各自的支持率,基本可以肯定方舟子的民调支持率要输给崔永元和韩春雨。

但是,对于自己在一般民众中的支持率不高或民调上的落后,方舟子似乎从未否定过。如果要问方舟子和川普的区别在哪里,这可说是方舟子和川普的第一个区别。
第二个区别,就要讲到民调能够解决什么?什么问题不能通过民调来解决?民调输赢和事实对错的区别,等我以后有空再讲。
第三个区别,就要讲到方舟子和川普各自在坚持什么?他们的坚持有什么区别?争当总统和坚持科学事实的区别,等我以后有空再讲。
第四个区别,就要讲到为什么民调输赢对川普很重要?为什么民调输赢对方舟子不重要?争取选民和“虽千万人吾往矣”的区别,等我以后有空再讲。

所谓外行看热闹,内行看门道。看热闹者,不求甚解,看什么都差不多。看方舟子也好,看川普也好,真正要看出点实质区别来,至少得把上面这些看清楚了。
 
嗯,你是外行。

表面看起来,方舟子和川普有某些相似的地方:他俩都很坚持自己的主张,都有很多反对者和很多支持者;他们面对的对手情况也很相似,也同样有很多的反对者和支持者。以方舟子和崔永元、韩春雨等的争斗为例,如果也弄一个民调来调查中国网民对他们各自的支持率,基本可以肯定方舟子的民调支持率要输给崔永元和韩春雨。

但是,对于自己在一般民众中的支持率不高或民调上的落后,方舟子似乎从未否定过。如果要问方舟子和川普的区别在哪里,这可说是方舟子和川普的第一个区别。
第二个区别,就要讲到民调能够解决什么?什么问题不能通过民调来解决?民调输赢和事实对错的区别,等我以后有空再讲。
第三个区别,就要讲到方舟子和川普各自在坚持什么?他们的坚持有什么区别?争当总统和坚持科学事实的区别,等我以后有空再讲。
第四个区别,就要讲到为什么民调输赢对川普很重要?为什么民调输赢对方舟子不重要?争取选民和“虽千万人吾往矣”的区别,等我以后有空再讲。

所谓外行看热闹,内行看门道。看热闹者,不求甚解,看什么都差不多。看方舟子也好,看川普也好,真正要看出点实质区别来,至少得把上面这些看清楚了。
 
一篇解释和分析针对美国总统辩论网上投票和真实民调的文章。
http://www.cbc.ca/news/world/trump-clinton-debate-polls-election-1.3789852

The clock is still ticking, but has time run out for Trump?
There are several reasons to believe Clinton might be unstoppable come November
By Keith Boag, CBC News

I voted for Donald Trump as the winner in last week's presidential debate. In fact, I voted for him several times, and again this morning.

I did so while experimenting with some polls on websites (Slate, Drudge, and others) where it's fairly easy to vote multiple times. So, if you were fooled into thinking Trump was, in fact, the winner of the debate last week, that's on me.

Those online surveys aside, what most people who watched the debate really think (including me), is that Hillary Clinton won.

Even if you didn't watch, you might have heard the punditry, supported by a couple of legitimate post-debate polls, declare Clinton won the thing hands down.

But the deception of a Trump triumph continues, aided, of course, by Trump as he trumpets the bogus online polls.
Such a great honor. Final debate polls are in - and the MOVEMENT wins!#AmericaFirst #MAGA #ImWithYouhttps://t.co/3KWOl2ibaW pic.twitter.com/sfCEE3I5pF
@realDonaldTrump

That particular part of the deception was enabled by websites such as Conservative Treehouse, which directed web traffic explicitly to skew Trump's numbers in more than a dozen Internet polls.

"Vote for Donald Trump right now," urged a site called TruthFeed, and linked to the websites where you could do so.

No wonder so many Trump supporters seem to believe that 81% of the country (the Drudge poll) agrees with them that Trump won the debate. And many of those supporters will now presumably dismiss legitimate polls that show a post-debate bounce for Clinton.

Never mind that those are probably the same people who swore by the scientific polls when they showed Trump driving a bulldozer through the field of Republican candidates during the primaries earlier this year.

That's just human nature at work.

usa-election-trump.jpg

Supporters of Republican presidential nominee Donald Trump cheer as Trump speaks at a campaign rally in Manheim, Pa., last Saturday. (Mike Segar/Reuters)


But there's a real story that polls are telling us: the presidential race has stayed pretty consistent throughout the campaign. Except for a day or two after the Republican convention, Clinton has led Trump in national polls since primary season when both clinched their nominations. The size of her lead has fluctuated, but the overall shape of the race has not.

The evidence is found in poll aggregation sites such as Éric Grenier's CBC Presidential Poll Tracker and the RealClearPolitics.com polling average.

A single poll, the LA Times/USC poll, has consistently put Trump ahead of Clinton by about five points. But that's not changed the story of the polling averages and there are reasons to believe that particular poll's methodology is what accounts for its different results. It's likely what pollsters call a "house effect"— a poll that consistently leans in the same measurable way, and doesn't represent the discovery of some warp in the electorate that the rest of the pollsters somehow miss over and over again.

Polling in swing states has been more topsy-turvy and is arguably more important. But when all the data is force-fed into the forecasting models used by FiveThirtyEight.com, Daily Kos, the Princeton Election Consortium and others, Clinton still comes out on top.

Lesser of two evils

We should consider other data, too.

There is a higher percentage than usual of undecided voters and those who say they support third party candidates this election.

Their numbers typically shrink by election day and that should be expected this time, too. After all, voters know that neither Green Party Leader Jill Stein nor Libertarian candidate Gary Johnson is going to be president. So if those voters want to make a difference, they might just decide to hold their noses and mark an X for whichever of the top two candidates doesn't make them want to gag.

That means personal popularity matters.

It's true, both Trump and Clinton are historically unpopular party choices, but he wins this dubious honour. Once the choice is seen clearly — it's either chicken or fish, as New Jersey Governor Chris Christie, a Trump ally, likes to say — then Clinton probably has an advantage as the lesser of two evils.

usa-election-debate.jpg

Republican U.S. presidential nominee Donald Trump and Democratic nominee Hillary Clinton are both unpopular choices for their parties. (Mike Segar/Reuters)


Sam Wang at the Princeton Election Consortium adds another important point: as exciting as this election has been, it hasn't made voters more excitable. In fact, the electorate has been on a 20-year glide toward stability. The 2016 polling has showed an electorate more stable in its opinions than at any time since the data was collected.

That makes sense when you consider how polarized the electorate has become since the 1990s.

But it's also a fact that argues against sudden and surprising plot twists in the remaining weeks of the campaign.

And there is another bit of data to pile on top of all that: In the past 40 years — 10 presidential elections — the candidate leading in public opinion after the first debate has won the White House every time.

Again, this time that's Clinton, not Trump.

Widening data gap

But maybe we're looking at the wrong data. That's the view of David Plouffe, who figured prominently in Barack Obama's campaigns for the White House in 2008 and 2012.

The president's former senior adviser argues that between what's publicly known and what's privately known, there exists an ever-widening data gap. Democrats especially have new and richly granular detail about voters that polling can never match.

Plouffe rates Clinton's chances of winning in November at 100 per cent. When he says so, it doesn't sound like hubris.

All of these things taken together are part of Trump's new reality show.

He is trailing Clinton and last week's debate might have been his last best chance to turn things around. He didn't; he knows it. Maybe that explains his wildly self-destructive behaviour since.

And that makes you wonder what's to come.

(David Plouffe, back when he was senior adviser to U.S. President Barack Obama in 2012. He's says Clinton is a lock to win this election.) (Jason Reed/Reuters)
 
后退
顶部