Back

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

In relation to the controversies about the number of participants of the July 1 rally this year, several academics, who were also involved in the research together with the author, have released various articles explaining our research details and perspectives one by one. In order to benefit by opinion from various sources, we have already released the original statistics through HKU POP Site for public reference.

The counting method used in July 1 rally this year was indeed drawn from those used in July 1 rally last year, together with January 1 rally and June 4 Candle Light Vigil. Again, aiming at benefiting by opinions from various sources, the author would like to briefly explain how this headcounting experience was drawn.

Headcount on last year's July 1 Rally

Obviously, the rally on July 1 last year targeted at the legislation of Article 23. A group of academics who earlier re-analyzed the compendium of public's submission on Basic Law Article 23 decided to conduct headcounting during the July 1 rally. The research team members included Jennifer Chan and Boris Choi from the University of Hong Kong, Chan Kin Man and Clement So from the Chinese University of Hong Kong, Ma Ngok from the Hong Kong University of Science and Technology, Sammy Chiu from the Hong Kong Baptist University and the author himself. Through the research, the team tried to explain the huge discrepancy between the figures provided by the activity organizer and that provided by the police over the years. Is this discrepancy due to some mistakes of statistical counting, or would it just be the result of some making ups?

However, due to limited experience as well as the unexpected huge number of participants in the rally, the academics were unable to adopt a perfect method. The Research Team chose two bridges across the main route of the march, placing a video camera on each of the bridges and taped the march from above. A systematic sampling method is used for which 1 minute was taped for every 15 minutes. The tapes were then immediately sent to a location with viewing facilities. Using a large viewing screen, students counted the heads in the fixed frames. The research team planned to finish all the headcounting on the night of July 1 and announced the result.

Due to various constraints and difficulties, however, the counting result could not be reached by the end of the day despite the continuous hard work of the 7 academics and 10 students. Some of the major difficulties include poor location and angle of the video cameras, time-consuming manual counting process, rapid change of people flow, existence of large number of participants who were not involved in the whole process, etc. For further details, please refer to "How to Calculate the Number of Participants in July 1 Protest" published in Media Digest and POP website by Clement So and Jennifer Chan.

As a result, the research team gave up one of the bridges and focused on videoing on the one which is at the junction of Hennessy Road and O'Brien Road. After several days of counting, it was found out that 254,000 people had passed that point. The research team further referred to the 489 responses of participants from the community-wide random telephone poll conducted from July 2 to 5, which estimated that only 57.1% of total participants passed that taping point. After integrating the sampling error of the poll, we found out that the number of participants for the day ranged from 429,000 to 502,000.

The Research Team issued a press release on 4th July, stated that "the saying that 500,000 people took to the street is basically believable". Meanwhile, Clement So and Jennifer Chan of the team published an article explaining the counting process and result in Media Digest and POP website in mid-July.

Although the Research Team was not able to announce the counting result on July 1st, the pre-announced counting plan might have put some kind of pressure on the activity organizer and the officials. We also do not know whether this is the reason why the figures from the activity organizer and the police matched well for that occasion.

Headcount on this year's January 1 Rally

In order to improve the methodology for headcounting exercises in mass gatherings and rallies, the author conducted a counting at the same place during the January 1 rally this year. As the counting is only experimental in purpose, the author conducted the counting with 4 members of the newly-established HKU Student Research Team by adopting a modified methodology.

The Team firstly divided half of the Hennessey Road into 3 pedestrian lanes conceptually i.e. 3 west-directed driveways. Video taking and counting were then conducted for 1 minute in every 10 minutes i.e. 1 minute after every 9 minutes. Due to limited manpower, however, counting was in fact carried out every 20 minutes for each lane approximately. This is also because 2 people were needed for double checking the data. Had not been the 3rd lane opened for a short period of time, the average counting frequency would be even much lower.

This was the first time in which the counting task was conducted by manual tally counter, 2 people were responsible for 1 lane and at the same time 1 person was videoing for back up purpose. On the other hand, some team members were recording the road situation, including the number of lanes that were occupied by participants and how the police controlled the flow of participants.

Probably because the research team has not announced the counting exercise in advance, there was, again, a large discrepancy between the figures released by the activity organizer and the police. Upon receiving various enquiries from the media, the authors disclosed the counting methodology and the initial estimation on 2ndof January via POP Site as follows:

"Under 5% of counting error, the number of participants who had passed through the bridge at O'Brien Road was around 36,000 - 40,000. This figure, nevertheless, did not include those who did not pass through the counting point due to late participation or early departure. According to the random telephone poll conducted after the rally, about 43% of participants did not pass that counting point. There was no such poll conducted for the January 1 rally, but from what we have in hand at the time being, there were 886 on-site interviews conducted and 410 online questionnaires. From our initial estimation at least 20% of participants did not pass through the counting point. Using 20% as the lower bound and 43% as the upper bound and also integrating the 5-percentage-point error, the initial estimated number of participants for the whole day is between 45,000 and 74,000, subject to further sampling result.

Meanwhile, the press release also stated, "we welcome people expressing their counting methodology for comparison and verification…...we hope more professionals can be involved in the counting task, and thus, statistics can be provided in a more objective and effective way, so as to avoid distortion of figures due to political consideration."

Later on, the author conducted 4 random telephone surveys during January and discovered that out of around 4000 samples, 133 of them had jointed the rally and 86% of the participants had indeed passed the counting point. Since there were only 114 samples this time, the sampling error for 86.0% was +/- 6 percentage points at 95% confidence level. As a result, using the figure that number of participants passing the bridge at O'Brien Road is between 36,000 and 40,000, the total number of participants for January 1 rally is around 39,000-50,000.

However, one should bear in mind that as the research is experimental in nature, there were a number of constraints, including only 4 students were involved in the counting, sampling density was only 1 out of 20, crowd control was taken place near to the counting station, and inadequate samples in the telephone survey afterwards. All these affect the accuracy of the estimation.

The author's main aim of conducting the counting research is to gather experience so as to provide reference for similar events in the future. This kind of experience was further supplemented on 1st July this year, which will be discussed tomorrow.

Comparison between last year's July 1 Rally and this year's January 1 Rally

Date	July 1 2003	January 1 2004
Counting Parties	The compendium of public's submission on Basic Law Article 23	HKU Student Research Team
Academics Involved	Jennifer Chan, Boris Choi , Chan Kin Man, Sammy Chiu, Ma Ngok, Clement So and Robert Chung	Robert Chung (advisor)
Counting Helpers	10 students from different universities	4 Members of HKU Student Research Team
Sampling Location	Bridge at the junction of Hennessy Road and O'Brien Road	Bridge at the junction of Hennessy Road and O'Brien Road
Counting Method	1 minute for every 15 minutes is sampled for videotaping, counting is then done on viewing facilities at frozen mode	1 minute for every 10 minutes is sampled for manual tally counting, and videotaping is also carried out for back-up
Number of Participants Passed-by	264,000	38,000
Loss Ratio	43.9%	14.0%*
Estimated Number	429,000-502,000	39,000-50,000*

* The loss ratio was calculated basing on the 4 random telephone polls conducted in January. Hence, the figure released in early January has to be adjusted.

Headcounting Experiences from June 4 Candle Light Vigil and July 1 Rally this Year

Yesterday, the author discussed the headcounting experience from July 1 last year and January 1 this year. Today's discussion is mainly about the experience from June 4 and July 1 this year.

Headcount on June 4 Gathering this year

"The Research Team on Headcoung in Mass Gatherings and Rallies", a team voluntarily formed by a group of academics including Jennifer Chan from the University of Hong Kong, Winnie Kwok and Clement So from the Chinese University of Hong Kong, Leong Kwan Li from the Hong Kong Polytechnic University and the author, decided to experiment various head counting methodologies . The purpose was to investigate the difference between dynamic and static counting. The author was responsible for designing and implementing the "moving square sampling counting approach".

In simple terms, the moving square approach is like this: team members firstly spot a 3m x 3m square in the football court and outline it by a rope. By counting the number of participants in this square, we obtain the density in this square. We then deduce the total number of participants by measuring the gathering area occupied by the participants.

In greater details, team members had used 2 approaches: one is setting the square at the specific point and specific court where participants have filled in the area whereas another is before participants have filled in the area. The second approach caused fewer disturbances to participants but nonetheless more time-consuming. The counting square has to keep moving with the participants' movement and this is more complicated.

From our experiment, the first approach is more effective. The eight student helpers were divided into two groups. The two groups reached a total of 164 and 42 set of statistics for the first and second approaches respectively. After considerable analysis and evaluation, 132 and 36 of the two approaches were used to compute the average density. The answer of 25.3 people per square was reached.

The total area of the 6 football courts which were used for the gathering was 14,917 square meter, according to the counting result of another team. By multiplying the density and the total area, the number of participants of the gathering was approximately 42,000, excluding those sitting outside the football courts and in the passage of the courts. This group of people was being counted by another 3 teams of student helpers and there were about 9,400 of them. Hence, the Research Team estimated that 51,000 were involved in the gathering.

Nevertheless, there might be 10% discrepancy in the estimated figure. Possible reasons include: uneven density of the crowd in the court, as our figures has revealed that the density of courts in the southern side was higher, with 6 people more per square than those in the northern side; the density in the inner and outer parts of the court also varied, while it is even difficult to count the number of participants outside the court; since people kept on coming in and out from the court, it is hard to judge the maximum size of court occupied by the crowd, while figures would also be underestimated if the turnout was estimated basing on the maximum area occupied. As a matter of fact, for the two sets of raw figures we are now adopting, the error for one of them was +/- 2 people, i.e. around 8 percentage points, under 95% confidence level.

In spite of the above limitations, after experimenting different counting approaches, the Research Team draws a conclusion that careful sampling is more cost effective than whole counting.

Headcount on July 1 this year

After incorporating various experiences from headcounting, "the Research Team on Headcoung in Mass Gatherings and Rallies" decided to conduct another counting exercise on the number of participants of July 1 Rally once again with the aim of finding the truth scientifically. Moreover, team members decided to fully disclose the methodology and statistics no matter how the result would be. Nevertheless, team members understand that it takes time to reach the final result, in particular, a time consuming telephone survey for calculating the loss ratio. Hence, research team decided not to announce the result in short term.

Furthermore, the team believes that constraints exist in every kind of counting method. The strength and quality of the people's demand do not necessarily affect the amount of participants in rallies. On the contrary, over-emphasizing the number of participants does not help in solving immediate problems. Yet, situation this year was different from last year's. There, as in the past, is a great discrepancy between the figures released by the activity organizer and the government. We are not sure whether this is because the research team have not announced in advance that result would be publicized.

In fact, the approach that research team used is of the same kind as those using for last year's July 1 and this year's January 1 rallies. The three of them all belonged to fixed station counting approach, a kind of dynamic counting method. Learning from last year's experience, the team had increased the number of frontline helpers to 12 and the sampling frequency had also been increased to 1 minute count after every 4 minutes. The counting task was, again, using tally counters, with 6 people counting different lanes at the same time, 3 people double checking the result, 1 person videotaping the process for back up checking and 2 people recording road circumstances .

Initially, the team planned to set the counting point at the bridge at the junction of Hennessy Road and O'Brien Road again. However, as there was construction work at the site, the team chose the one situated at the junction of Hennessy Road and Arsenal Street i.e. where people are entering Admiralty from Wan Chai.

First of all, the research team divided the Hennessy Road at the area into 8 pedestrian lanes. That is indeed equal to 6 East-West driveways plus 2 tramways. Video taking and counting were carried for 1 minute after every 4 minutes. It turned out that the 2 tramways were not used for rally but as emergency pathways. The total number of people flow for the remaining 6 lanes were about 149,000. In order to assess the adjustment factor, the research team conducted a random telephone poll from 2nd -11th July. 231 out of 3,512 samples aged 18 or above have joined the July 1 rally. 77.4% of this group of people said they passed through the counting point. As there were 178 people in our sample, the sampling error at 95% confidence is +/- 6%. Suppose all the estimating error is attributed to sampling error of telephone poll, total number of participants of July 1 rally for the whole day should thus be adjusted to between 180,000 and 210,000. Raw figures of the above estimation have already been fully released via the HKU POP Site.

Dealing with the number of participants in rallies properly

A group of academics with common interest take the initiative and volunteer to form the research team. Firstly they re-analyzed the compendium of public's submission on Basic Law Article 23. They then conducted research on the number of participants in rallies and published articles afterwards. All in all, they just want to promote civil society and the truth seeking spirit in science.

Frankly speaking, with limited resources, there is limitation in each and every single counting methodology. As one of the subjects of science, different statistical methods involve different definition, theory and practical consideration. As long as the research processes, methodology as well as the raw figures are released, all people can then benefit from various sources and gradually develop a set of statistical method which suits the local society.

The author always thinks it will be better if the organizers and the government could leave those counting work to professionals. This could avoid unnecessary political controversy. What they should do instead is to release all the raw figures and counting methodology in the past so as to facilitate academic research.

The number of participants in rallies should never be used as a political instrument. This kind of numbers will become significant only if there is no democracy and where public opinion could not be reflected in elections. As what the author has once written, "In an advanced and civilized society, one idea is one idea, one single opinion is one single opinion. For good ideas, we cannot afford to miss even one; but for bad practices, one is already too many." It is a pity for the society to regard the number of participants in rallies as a kind of public opinion benchmark for policy administration.