About Methodology of “2016 Legislative Council Election Rolling Survey”Back

August 19, 2016

The Public Opinion Programme (POP) of The University of Hong Kong has been conducting rolling surveys in all election years since 1995. This year is no exception, with similar methodology. However, some small differences in the design have caused some debates, so POP would like to provide some scientific explanations here.

On Reading Candidate Names

The number of valid candidate lists in geographic constituencies has increased from 67 in 2012 to 84 this year, or a growth of more than 25%. As a result, in our telephone surveys, we inevitably have to follow the current practice of most electronic media in reading out the first candidate’s name only in each list, instead of first two names. This is purely out of practical concern and following normal practices. Some people have suggested reading out two candidates’ names, while others have suggested reading out the names of all ex-Legco members. We believe that no practice, even reading out ALL the names in each list, is absolutely fair. What we are doing now is already the best solution under the current situation.

However, in view of the concerns of different people, we have spent extra resources to conduct a scientific experiment for five consecutive days from August 11 to 15. Apart from interviewing 1,027 registered voters using standard method, our interviewers randomly selected 469 respondents to conduct a test by “reading out two names”, some of them read out “one name then two names” while some read “two names then one name”. Result shows that whether reading out one name or two names had no significant differences statistically. After careful consideration, we have decided NOT to change the current question setting. We will release the detailed result of the experiment next Monday. We welcome anybody to conduct similar experiments to scientifically verify the results.

On Sample Size

Due to budget constraint, the sample size of rolling survey is small at the beginning, but it will gradually increase at a later stage to reach 3,500 per week. In the last few days, the sample size will increase to more than 600 per day, which is three times of that at the beginning. By then the sampling error will be reduced and thus the accuracy of the figures will be increased accordingly.

On Using Landline Numbers

The target sampling frame of our rolling survey is users of telephone landlines. Part of the reason is because there are even more constraints if we use mobile numbers, such as low response rate, high proportion of youngsters, one person with many mobile phones, and so on. Furthermore, same as other academic institutes, we use landlines to conduct opinion surveys. If suddenly use mobile numbers, it may lead to other unfairness. We believe that as long as the data is properly weighted, conducting surveys with landline numbers is still the best solution.

Media enquiry: Frank Lee Tel: 3917 7724 Email: [email protected]