Tuesday, September 09, 2008

是集體智慧,抑或純綷好運?

很多人試圖解答這個問題,唯有這篇文章令我眼前一亮:
Fooled by Randomness(by Justin)

“假如有一班人投票,目的是希望泛民能保住關鍵小數的地位,而對泛民個別的候選人沒有特別偏好時,隨機投票可能比配票容易。每個人都從泛民裏隨便一張名單,只要人夠多,每張名單的得票自然相近(如新界東,廿萬張選票均勻地分佈在五張名單上)。選配票之難,在於它所需之巨大資源,也在於人類難以預測的行為;反而,在泛民在無辦法配票的情況下,人們各自各「隨機」,各候選人所得的票數卻能平均分配,無招勝有招,能得到超乎意料之外的結果。”

不少政界與傳媒事後孔明的評論流於吹水,唯它例外地以科學解釋群眾行為,層次都不一樣。佩服!

Reminder:沒有讀過Fooled by Randomness / The Black Swan的,可能看不懂原文。不過沒學過量子力學與牛頓三定律的倒沒問題。我認為。
:)

7 comments:

s tsui said...

The problem is, when Fan Man advocates that people do not answer exit polls, the single most effective measuring tool for voter behavior is sabotaged. We will never be able to come up with a scientific description of what happened in this election because of that irrational objection.

Leona said...

但Justin純綷以選舉結果作出推斷,不涉exit poll啊
當然,你可以說,沒有民調可依,誰知道泛民支持者心裏想的是什麼?
那其他評論員也是不知道的,他們又憑什麼說那是因為群眾配票的智慧etc.呢?

或許這樣說,原因是什麼誰也不知道,大家都是用不同的理論去解釋之,Justin的也是其中之一,而這個講法能說服我

想一想,其實科學理論統統都是從實驗結果作出推斷直到被推翻...又怎能說這不科學呢?
:)

said...

這種quantative analysis我不擅長。但他的說法我覺得是有漏洞。明顯看得出作者有考慮到大數法則(law of large numbers),寫到「只要人夠多」,可惜廿萬這個是不是夠多確實是未知之數。極端一點,5張名張可能其中一張達到七八萬亦有可能發生。
反而在配票前,做一點統計。當然不是指全部,而是利用一些sampling method做一些抽樣調查。既可以減低費用,準確性又有一定保証。

s tsui said...

the essence of nnt's writings is that he (and i :P) is staunchly against armchair speculators. trying to guess what happened by looking at the outcome is possible but definitely not advisable. one can only find out what happened by testing various hypotheses, tinkering with the input and see if the output changes accordingly, and always be ready to accept refutation.

without exit poll or other kinds of scientifically valid (and useful) data, i wouldn't engage in speculating about how Fan Man managed to win a certain number of seats.

just logically, when justin stated that "人們各自各「隨機」,各候選人所得的票數卻能平均分配" - this statement actually does not tell us anything. randomness most definitely does not equal "balanced distribution". therefore, it is not obvious to me what is his rationale supporting the thesis - "隨機投票可能比配票容易 (in helping Fan Man 保住關鍵小數的地位)".

:P

Martin Oei said...

沒民調可依,還是有觀察方法的。

Leona會比較清楚我是怎推出來的,因為事前事後兩篇分析文章,我都是交給她的。 =)

我不是用計量方法,因為現時有人有系統地搞砸民調時,你手上的Sample有問題,你用什麼方式糾正都有問題。

由於有互聯網,用觀察互聯網的輿情數字,例如facebook group的參與人士,再參照有其他類似情況的地方,由觀察其他地方同類現象所歸納出的東西,拿香港來test便知。

我是拿馬來西亞峇東埔補選觀察結果,直接應用在香港,所以我已經在前一天講,選舉結果對泛民有利。

現在搞社會科學,過份依賴統計學是走火入魔。正如金融市場,CFA Level 1和2的計量方式我全部都識,問題實際金融市場操作時,你純用計量的結果係被人屈機。在今年Barcamp Hong Kong 2008,有個人用黑客界常用逆向工程手法,意圖推出機構投資者常用的買賣程式,然後上下其手去屈機。呢次民調事件,都係類似情況。

得閒睇張五常《賣桔者言》,我在大學教書時,時常要求學生睇呢本書。

tintinbright said...

我的門外漢observation:

很有趣的解釋.不過取得五票這'結果'祇是'一次'的outcome. 重複N次投票會不會也是取得五票呢? 當然我們無從實驗, randomness 也看似一個很合理的解釋.

其實泛民各有擁護者, 大家表現平均, 亦各自打出告急牌, 選票'恰好'頗平均的分配亦是自然現象. 所有的解釋 (包括我的)一樣是吹水 :-)

fooled by "Fooled by Randomness"? :D

Justin said...

S Tsui:
Yes, my method is not that very accurate actually. I presume all the 204030 votes have no preference among all pan-democrats list, and they vote randomly. Then the votes will be evenly distributed among all lists.

Despite of my errors in the article, the main thesis still stands. The main thesis of my post was, distributing votes is too difficult to be carried out in practice, and actually a similar result could be attained by asking the voters to choose randomly among the five lists.