Sunday, May 13, 2007

Chutzpocrisy and the Taiwan media

Ya gotta love the chutzpah and hypocrisy of the Taiwan media. The other day ESWN translated a piece from the pro-KMT China Times on the DPP primary polling....

Some background. The Pan-Blue (KMT + PFP) world is the world where conspiracy thinking passes for analysis, where Chen Shui-bian won the 2004 election because of an assassination attempt on him the night before, arranged by Chen himself; where Chen Chu won the Kaohsiung election because of accusations of vote buying that broke the night before; and now, we're going to hear about how Frank Hsieh won the DPP primary because the DPP primary vote was rigged. The China Times' "expose" has to be understood against that background. Such thinking is pre-industrial thinking, where the world is a place Where There Are No Accidents and Everything Happens Because Someone Arranged It That Way. When I was in the Peace Corps I used to get this from my Kenyan colleagues -- I was clearly a CIA agent, and everyone knows that the CIA arranges all events in the world to its liking.....

The China Times wrote of the survey of the population which formed a portion of the DPP primary data:

Suppose this was a general population survey of the entire adult population. It is general knowledge throughout the world that telephone surveys experience differential response by population sub-groups. For example, young people are less likely to be at home to take and coooperate with telephone calls; poor people are less likely to have telephones; etc. The differential responses rates will show up by monitoring the demographic distribution of the sample compared to the population (according to the census). For example, the sample shows 8% male 18-29 but the census says 10%. Usually, the sampling statistician would weight the sample. For example, each male 18-29 is now worth 10/8 = 1.25 persons and so on. This is accepted practice around the world.

In the DPP primary election, there is a wrinkle. Instead of surveying the general population, the DPP is filtering out the pan-blues (those who identify with the Nationalist and People First Parties). That is their prerogative. But they have to deal with the statistical consequences.

As a survey sample, this one will experience differential response rates like any other survey sample. Therefore, they should weight the sample to correct the biases. But to what values? In the general population, the percent of male 18-29 might be 10%. But what is the percent in the sub-population with the pan-blues being filtered out? The government census does not enumerate this. The percent of male 18-29 could be 8%, or 12%, or whatever.

So now there is a dilemma. If they don't weight, they are accepting the biases from the differential response rates; if they weight (for example, to the distribution in the general population as a whole), they may be introducing a different sort of bias. There is no clear correct answer here. The numbers can be produced as weighted or unweighted, and they will be different. Whichever number is chosen, it will be different from the other one. So someone is going to gain and someone else is going to lose as a result. This will be a point of contention. But the rules should be spelled out beforehand to remove the accusation that the choice was made post facto to favor certain candidates.

In other words, as the China Times hinted in that last sentence, the DPP screwed with the numbers to support favored candidates. Who that favored candidate would be is a bit mysterious, since the media here has been claiming that Chen Shui-bian supported Su and dislikes Hsieh, so why the all-powerful Chen, who can arrange assassination attempts to his own benefit (according to KMT mythology), can't arrange his own party's primary so Su would win it, is beyond me. But the really enjoyable thing about Conspiracy Thinking is that logical consistency is not a priority.

The chutzpocrisy here is really beyond the power of words to express. So instead, I'll supply you with that natural antidote to conspiracy thinking, numbers. Here are the numbers from the polls before the Kaohsiung mayoral election last year:

Liberty Times (12/5): Huang Jun-yin (34.38%) versus Chen Chu (34.09%)
China Times (12/4): Huang Jun-yin (43%) versus Chen Chu (29%)
TVBS (12/3-12/4): Huang Jun-yin (51%) versus Chen Chu (31%)
United Daily News (12/2): Huang Jun-yin (39%) versus Chen Chu (27%)
Era TV (11/26-11/27): Huang Jun-yin (41%) versus Chen Chu (26%)
ETTV (11/24): Huang Jun-yin (41%) versus Chen Chu (29%)

The pan-Blue media outlets (#2-6 on that list) all had Huang Jun-yin up by massive amounts, with the Chinese-owned TVBS taking pride of place at Chen Chu losing by 20 points. The China Times is there awarding Huang Jun-yin a 14% lead over Chen Chu.

The actual score? Chen Chu won by 1%. Thus, the polling experts at the China Times: Huang +14. Reality? Huang -1.

Yup. The China Times is certainly in a position to complain about weighting and survey results.

But let's revisit the logic of their primary claim, which is intended to support a sly conspiratorial insinuation:

So now there is a dilemma. If they don't weight, they are accepting the biases from the differential response rates; if they weight (for example, to the distribution in the general population as a whole), they may be introducing a different sort of bias. There is no clear correct answer here. The numbers can be produced as weighted or unweighted, and they will be different. Whichever number is chosen, it will be different from the other one. So someone is going to gain and someone else is going to lose as a result. This will be a point of contention. But the rules should be spelled out beforehand to remove the accusation that the choice was made post facto to favor certain candidates.

Read it carefully. First the China Times admits that whether you weight or not, you introduce bias into the poll. Quite true. But had the DPP weighted the responses from the sample, the weights would almost certainly have been arbitrary, and based on some other poll or social science research whose results were themselves weighted in ways that are not very solid (what level and type of belief makes one a Blue? What definition of Blue is acceptable to all viewing the poll?). Whenever one defines a sample and a methodology, one runs into the problem of arguable definitions. The China Times mentioned this to kick up dust, and make it seem like there is a problem, so as to support their underhanded claim: the DPP primary is actually a conspiracy. Merely claiming Bias in the poll! is easy; all polls have bias to some extent. What the China Times would have to demonstrate is that the bias affected the results -- which it cannot do because it does not have reliable alternative numbers. In fact, the China Times is probably incapable of producing such numbers, as its performance in recent elections here shows. The discussion here is simply the analytical equivalent of white noise.

Having kicked up dust and cast doubt, the China Times then claims that the rules should be spelled out beforehand. But this is a bit of disingenuous nonsense, since the rules were spelled out beforehand: the poll wasn't weighted and every candidate knew it. All candidates faced the same poll and from that perspective, the poll was perfectly fair. The China Times is simply hoping to cause discord in the Green camp, and feed its own readers' penchant for conspiratorial nonsense.


6 comments:

Anonymous said...

What explains the discrepancies between the pan-blue poll numbers and the actual election results? Are they using faulty polling techniques, much like Literary Digest did in the 1936 US presidential election (when it predicted Alf Landon would trounce FDR)? Or are they hoping that by manipulating survey data, a bandwagon effect can be created that favors pan-blue candidates?

channing said...

IMO, most polls are hit-and-miss. Especially in Taiwan, when there are vote-buying and vote-counting accusations flying back and forth all the time, as well as the fact that Chen Chu won the KH mayorship by a figurative few votes...

...you should note that every poll listed had Huang Jun-ying as the winner, and the vote recount was never completed...basically, when you poll on a fishy election, the results are bound to be inaccurate.

Michael Turton said...

Kaminoge, it is the latter. The Blue media simply makes stuff up to suit its political goals.

Channing, the Kaohsiung election was not fishy. I suggest you stop paying attention to Blue bullshit, and track down the actual numbers. There you will see that Chen Chu had 379K votes, Huang had 378K votes, and the TSU candidate had 7K votes. The total Green vote was 386K, same as in 2002. The Blues did a much better job of getting out the vote, and vote buying, this time around, giving them an huge improvement in their vote. The Blue media polls are simply there to provide fodder for post election conspiracy shit intended to discredit any Green victory. Just look at the results of previous KH elections -- they are almost always close.

Michael

Michael Turton said...

In other words, the Kaohsiung election was close not because the Greens bought votes but because the Green vote was split. In Taiwan politics polls should be ignored.

channing said...

Well, your perspective is a green-vote vs. a blue-vote, whereas my intention was strictly KMT vs. DPP in the Kaohsiung election, hence my mention of the small margin of victory. I'm not quite sure what "Blue bull" you were referring to, but Wiki says the vote recount was completed, increasing the victory margin. I suppose that settles some of the dust.

Michael Turton said...

No, Channing, my perspective is DPP vs. KMT, but it is necessary to understand the whole Green vote in order to understand why the Kaohsiung election was so close.

It was you that used the word "fishy" to describe the Kaohsiung election, which is why I told you to stop listening to Blue nonsense. Nothing fishy about it; it is perfectly understandable if you look at previous elections.

Michael