Even though Barb has at last got its new improved panel up and running, it is still not clear just how many people are watching low rating niche channels. Are we being asked to take too much on trust?
For most people in the television industry, Barb ratings are the standard currency. Producers rely on them to see how their programmes perform, channels rely on them to attract advertising and the advertisers themselves use them to decide how to assign vast sums of money. While the Barb system is still headline news and people are questioning how it works, it's a good time to delve behind the figures we often take at face value and see exactly what they mean.At first this can be frightening - partly because it requires the use of statistical jargon and partly because the results are often not what people expect. When looking at the accuracy of the Barb panel, a key measure is the Relative Standard Error - this is expressed as a percentage and shows how much a given television rating (TVR) could statistically vary because of the vagaries of the panel system, without the real audience varying at all.For any TVR it is possible to calculate the potential low audience and the possible high audience that could result from this Relative Standard Error, within the 95 per cent confidence interval. In other words, we can be 95 per cent sure that the channel's audience was somewhere between these two high and low audience figures in the minute the TVR was recorded.For channels with a low level of viewing, the panel fails to give an accurate measure over one minute. A TVR of 0.04 per cent (ie 0.04 per cent of people with access to multichannel were watching) could represent an audience of between 700 and 16,962 viewers for one minute.The higher the audience, the lower the error rate - a typical primetime minute on Channel 5 will be accurate to within a reasonable 7.3 per cent (based on a TVR of 3.14 per cent), while a primetime ITV 1 minute with a TVR of 12 per cent will be accurate to within a very respectable 2.8 per cent. In this last case, we can be 95 per cent confident that the audience falls between 6.3 million and 7 million in that minute.These are worst case scenarios - there is a higher probability that the actual audience will fall in the middle of the range. These figures also only apply to a single minute - to achieve an audience at the margins of those outlined in the table would be unlikely and there would be a low probability of this occurring again.According to a 1980 study by Jictar, the organisation that preceded Barb, the standard error rate for a four-week average of a 15-minute slot would be the single-minute standard error multiplied by 0.8. When the ratings are averaged (for example, to give a quarter-hour figure or a programme rating), these error rates are reduced.In January 2002, the Barb network panel totalled 9,116 individuals - some way short of the eventual new panel size (it is expected to be up to full strength in March) and down on the 10,490-strong panel from December 2001. The multichannel panel was similarly under strength in January, comprising 4,358 members compared with 5,092 on the old panel. When the new panel is fully implemented, both panels will increase above 2001 levels.The way the panel is comprised has also been tweaked, particularly the multichannel panel. Comparing the multichannel panel from January 2002 with the December 2001 equivalent, the most noticeable change is a reduction in the dominance of younger viewers (particularly four to nines) and an increase in viewers aged 35 plus. Viewers aged 65 plus accounted for 6.6 per cent of the old multichannel panel and 8.4 per cent of the new panel in January.Of course, each group is weighted to represent the same total number of viewers (based on the establishment survey which determines the demographic split of the panel), but an increase in the number of panel members in a demographic group results in a decrease in the 'weights' (the number of viewers a panel member represents) and an increase in accuracy.