Graphical presentation depicts quantitative data effectively. It can help readers build up a clearer impression of the trend and distribution of the data than columns of figures can. However, if not done properly, it can easily cause an optical illusion and mislead readers. Look at the two bar charts (Diagrams 1 and 2) presented above. Both diagrams present the sales turnover of a company from 1995 to 1998. A quick glance at Diagram 1 may give readers the wrong impression that the turnover in 1998 was double that in 1997 and four times that in 1995. This is because the bar depicting the turnover in 1998 is two times as high as that in 1997 and four times as high as that in 1995. On the contrary, casual viewers of Diagram 2 may think that there had not been any notable changes in turnover from 1995 to 1998 since the heights of the bars look almost the same (particularly those depicting turnover from 1995 to 1997). What causes the differences in the messages perceived by readers of these two diagrams? It is the initial value of the y-axes of the bar charts. In Diagram 1, a 'non-zero' starting point is used while in Diagram 2, a 'zero' starting point is used. Because the initial value of the y-axis in Diagram 1 is set at $1,000 million, all the relative differences in turnover (for turnover above $1,000 million) have been shown in the bar chart in an exaggerated manner. Although the variations in turnover from 1995 to 1998 are not very large (less than 15 per cent), the readers are given a false impression that the changes are very significant (300 per cent) by simply comparing the size of the respective bars. In Diagram 2, a 'zero' initial value is used for the y-axis. Because a large scale is also used for the y-axis, a small variation in the height of the bars will represent a large difference in turnover. As a result, the heights of the bars in Diagram 2 do not show any conspicuous differences since there were only moderate changes in turnover. A more appropriate way of presenting the data is shown in Diagram 3. A 'zero' initial value and a broken-scale are used for the y-axis of the bar chart. This method of presentation shows more clearly the changes of data over the years while drawing readers' attention to the possible illusions that could be caused by reading the bar chart casually. For more information on this series of articles, please write to the Census and Statistics Department at 21/F, Wan Chai Tower, 12 Harbour Road, Wan Chai, Hong Kong, or call 2582-4732. The department's homepage is at http://www.info.gov.hk/censtatd/