1. (C) Summary: Taiwan’s vibrant polling industry has a spotty record in forecasting island-wide and local-level elections. Public opinion polls conducted by pro-Blue media organizations such as TVBS and United Daily News (UDN) often greatly overestimate support for KMT candidates and underestimate the strength of DPP candidates. Industry experts attribute polling problems to a combination of factors, including an urban, white-collar worker sampling bias, a large block of silent “undecided” voters who tend to support the DPP, and a “branding effect” generated by the parent media corporations. Compounding these problems, many survey centers release relatively unfiltered polling results without adequate weighting or accounting for key variables, such as likelihood of voting. Despite these deficiencies, election pundits will continue to rely heavily on often faulty Taiwan polls in their analyses and forecasts of key legislative and presidential elections at the end of 2007 and beginning of 2008 (Ref A). End Summary.Taiwan Polls Don’t Come With A Warning Label
——————————————–
2. (C) Taiwan’s public opinion polling industry has seen an explosion of commercial, media, think tank, and academic polling centers in recent years. Media organizations regularly provide the public with polling data on potential and actual candidates in important elections. Their track
record is spotty, to say the least. Most media polling is conducted by pro-Blue outlets such as TVBS and United Daily News, which tend to overestimate support for Kuomintang (KMT) candidates, while underestimating the strength of Democratic Progressive Party (DPP) candidates. While media polls were relatively accurate in predicting the 2005 county and city elections, more recently, their forecasts for the 2006 Taipei and Kaohsiung mayoral elections were far off the mark. Taipei DPP candidate Frank Hsieh garnered 41 percent of the vote despite polls in the preceding weeks that showed him in just the low 20s, and in Kaohsiung, DPP candidate Chen Chu squeezed out a narrow victory over her KMT rival although she trailed by double-digits in most polls (Ref B).
3. (C) Industry experts attribute polling problems to a combination of factors, including an urban, white-collar worker sampling bias, a large block of silent “undecided” voters who in past elections have tended to support the DPP, and a “branding effect” generated by the perceived political stance of the parent media corporations. Compounding these problems, many polling centers release relatively unfiltered polling results without adequate weighting to account for under-represented groups and without factoring in important variables, such as the likelihood of a respondent to vote. Some polling gurus have told AIT they do not expect the efforts by media organizations to improve their polling techniques to overcome these methodological challenges ahead of the 2007 legislative and 2008 presidential elections.
Urban Bias: Under-counting the DPP
———————————-
4. (C) Lower to lower-middle working class people in rural areas make up the core base of the DPP, while urban white-collar workers tend to favor the KMT. This latter group is over-represented in polls conducted by the major Taiwan media outlets because of relatively higher land phone
penetration and usage, even though the voter turnout rate for the urban population is generally 10-20 percent lower than that of rural residents. Global Views Monthly (GVM) Polling Center Director Tai Li-an, one of Taiwan’s foremost polling experts, explained to AIT that the urban bias often translates into polling numbers that, if not properly weighted, are skewed in favor the KMT. The polls also tend to be conducted in the early evening hours, a time when members of the working class may still be at work or returning home, further exacerbating the pro-KMT bias, added
TAIPEI 00000769 002 OF 003
Tai.
“Undecided” Voters: Tend to Lean DPP
————————————
5. (C) Taiwan public opinion polls also exhibit a high percentage of silent “undecided” voters, which generally ranges between 20-30 percent of those polled and can sometimes go as high as 40-50 percent. National Central University Election Study Center Director Yu Ch’ing-chin
explained to AIT that many polls fail to identify the voting preferences of this large segment of the voting population. Many “undecided” voters appear to be DPP supporters unwilling to express their party preference, especially when the DPP is the underdog or under attack by the opposition. GVM Director Tai noted that the percentage of voters who expressed a preference for the DPP dropped from 25 in 2005 to 15 in 2006, but the percentage of so-called “undecided” or “independent”
voters moved from 25 to 35 over the same period. By asking indirect questions, Tai found that that the new “undecided” voters nevertheless still expressed a slight preference for the DPP. These “undecided” voters are likely to vote for the DPP, but they are generally not counted as supporters for a DPP candidate in media polls.
“Branding Effect:” Coke versus Pepsi
————————————
6. (C) The political affinity and reputation of a polling organization’s parent media company also affects polling, generally skewing the results in favor of the KMT if the media organization is known as pro-Blue and in favor of the DPP if the organization is pro-Green. (Note: In practice,
pro-Green media organizations such as Liberty Times do very little polling.) Shih-Hsin University Professor Su Chien-chou likened this to the “branding effect” that occurs during a Coke-Pepsi taste test as consumers are more likely to prefer a particular cola if they know that the cola company is sponsoring the test. This phenomenon can affect polling results in one of two ways. First, supporters of the DPP or KMT will refuse in higher numbers to participate in a poll conducted by a media company they believe is associated with the other party. Second, survey respondents who do not have a strong political preference tend to answer in a manner that anticipates what they believe to be the political bias
of the media organization conducting the polls. While this “branding effect” applies equally to pro-Blue and pro-Green organizations in theory, Professor Su explained, the media, especially those organizations with polling centers, is predominantly pro-Blue, which means that most election polls produce results biased in favor of opposition candidates.
A Diamond in the Rough?
———————–
7. (C) The challenges of sampling biases faced by polling organizations are not insurmountable. TVBS Polling Center Director Wang Yeh-ting told AIT that major media-affiliated polling centers are beginning to recalibrate their polling methods and models after their “poor” performance in predicting the outcome of the 2006 December mayoral races. They are also considering whether or not to publish filtered, weighted polls that try to compensate for some of the persistent biases. GVM Director Tai, however, criticized his fellow pollsters for not making fundamental changes to their polling methodologies and models. Tai explained that
Taiwan’s polling biases can be mitigated with the use of a sophisticated statistical model that weights factors such as education level, voter turnout, and party preference. Most polling centers in Taiwan, Tai said, are still using first-generation models developed over a decade ago and they will continue in the near term to produce faulty poll results that are not properly filtered or weighted. Furthermore, Tai predicted that many polling centers will face difficulties in upgrading their telephone calling directories before the 2007= legislative elections, when new, smaller election districts are introduced for the first time. (Note: Tai has developed a complicated election prediction model that is in its sixth generation of development and was remarkably accurate in forecasting the results of the December 2006 mayoral races.)
TAIPEI 00000769 003 OF 003
Comment
——-
8. (C) The plethora of polls in Taiwan is a mixed blessing for watchers of Taiwan politics and elections. On the one hand, there is a great deal of polling data, which is at least suggestive of general trends. On the other hand, much of the data in media polls has not been properly weighted and also suffers from “branding effects” and other political biases. Politicians and commentators are all too ready to pick and choose those polling results that serve their own political interests. Most polling conducted by media organizations should be used with great caution. WANG
Hartford Group International Taipei 