Back

Note: This article represents the view of the author and not the University of Hong Kong.

We all know that the outcome of the United States presidential race is decided by the Electoral College. That is why no matter how accurate opinion surveys could be, they could at best predict the popular vote of the candidates, not the number of electoral votes. Four years ago, Al Gore won the popular vote, but George Bush was elected as president because he had more electoral votes. Whether history would repeat itself or not would soon be known.

Yesterday, the author explained the operation of rolling polls conducted by the most credible pollsters in America. Today, we shift our attention from the prediction of popular votes to the study of electoral votes. We will also look at the related issue of random sampling errors. The author's wish is to provide more substance to local experts, scholars and policy research directors, so that they can examine the American polling system in more detail, and make more contributions to objective and comprehensive discussions on the development of opinion polling back in Hong Kong.

The analysis of electoral votes usually quoted by the Hong Kong media are generally from the polls conducted by Zogby and CNN. Thus, this article will focus on discussing the methods adopted by these 2 pollsters. Yet, it must first be stated that both pollsters are in fact conducting the nation-wide public opinion polls at the same time, and the Zogby's nation-wide 3-day rolling survey released by Reuters is even the focus of the Hong Kong media. Yesterday, the author has already had an in-depth discussion on this survey. This article will only concentrate on discussing the electoral vote-based state-wide survey.

In Zogby's view, due to the system of the Electoral College, the result of this presidential election in fact hinges on the voting result of 10 key states, which are named by Zogby as "battleground states". Therefore, Zogby has kept on conducting rolling surveys in these key states in a 4-day rolling basis, aiming at accumulating 600 likely voters in 4 days. That means around 150 people have to be interviewed per day. The survey is conducted by telephone based on random sampling within the states, with a sampling error of +/- 4.1 percentage points. There are not much controversies arising from the rolling survey method both inside and outside America.

According to the latest Zogby Battleground Poll released by Reuters, Kerry led in 6 key states, but lagged behind Bush in the 4 remaining states. The survey was conducted from 27 to 30 October and released on 31 October. It must be noted that the Zogby's nation-wide rolling survey is on a 3-day basis, with a sample size of 1,200 cases. However, the number of rolling days in its state-wide survey has increased to 4 days and the sample size is reduced to 600 cases. While these changes are made in the light of the specific conditions, the same principle is also applied in the author's rolling survey of the Legislative Council elections in Hong Kong.

Now, let us take a look at the CNN's integrated analysis, which is also frequently quoted by the Hong Kong media. During the elections, CNN has created a "CNN Electoral Map" every week. The latest one was released on 29 October, showing that if the election were held on that day, Bush and Kerry could obtain 227 and 207 electoral votes respectively. Among the 8 key states, Bush and Kerry each led in 4 of them. According to the illustration of CNN, the "CNN Electoral Map" was a macro-analysis integrating the public opinion polls' result in each state, the expert opinions, as well as the voting record and political parties' election engineering of each sate. CNN has even stressed that these analyses might not match the election result.

Same as Zogby, CNN has based on the past voting record and this election's situation to select 15 key states, namely "showdown states", and conducted surveys in each key state. Generally, 1,300 registered voters were interviewed in 4 days, and among them, 1,100 were voters intending to vote. The latest round of state-wide survey was conducted from 27 to 30 October and released on 1 November, with a sampling error of +/- 3 percentage points. The survey reveals that, among the 10 states, only 2 could tell who the winner was from the public opinion polls, while Bush and Kerry each took one.

Apart from providing the Electoral Map and state-wide survey figures, CNN has also made heavy use of the analysis of "poll of polls". This is in fact some simple daily average figures based on a combination of results of various nation-wide surveys conducted by large pollsters, so as to demonstrate the candidates' strength. Take the release of 1 November as an example, the integrated analysis shows that the support rating of Bush was 48%, while that of Kerry was 46%. The data was the average of figures released on 31 October by 6 pollsters, including CNN itself, Fox News, New York Times and other pollsters.

CNN explained that the "poll of polls" could cover larger sample size, and so it was more stable and a more macro view could be captured. However, no matter how stringent the filtering process of the "substance survey" is, the problem arising from variations of the sampling, rolling and weighting methods among different surveys still cannot be solved, and thus the integrated figures can only be taken as a kind of reference.

Nevertheless, CNN's surveys in the 15 states have indeed provided very systematic reference materials. In the website of the "poll of polls", the reader can easily select any one sate to examine the result of various public opinion polls. The author has browsed this website and found that the samples of many state-wide surveys are controlled at around 600 cases, with a sampling error of +/- 4 percentage points. The sample size of those smaller surveys is lowered to 400 cases, with a sampling error of +/- 5 percentage points.

This brings out the final issue which the author would like to discuss, that is the question of sampling error. The author released an article entitled "the Basics of Rolling Polls" in mid- September. In that article, the author pointed out that when the sample size was around 600 people, the sampling error of +/- 4 percentage points would appear in those figures near to 50:50, which was the present situation of the United States presidential race. However, in Hong Kong's Legislative Council election, if one's support rating was 5%, the sampling error of a 300-people constituency-wide survey would only be +/- 2.5 percentage points. By the same token, if one's support rating was 15%, the sampling error would then be +/- 4 percentage points. Back to the rolling survey conducted by the author during the Legislative Council elections, since the source of votes was rather dispersed in each constituency, the sampling errors of support ratings of each candidates' list were basically controlled at +/- 4 percentage points. Even at the final stage of the election when the author divided the latest constituencies' figures into 5-day, 4-day and 3-day rolling samples to release, the sampling errors of those smaller constituencies were also below +/- 3, 4, 4 percentage points respectively.

There are not much criticisms in America concerning the +/- 4 percentage points, because commentators there have a better understanding of the value of public opinion polls and the concept of random sampling. If the Chinese really wants to surpass the British and the American, and if Hong Kong really wants to become more superior than New York and London to become China's Manhattan, maybe it is better for us to learn seriously the theories and methodologies of public opinion polls, in order to avoid meaningless political attacks.