Currency Convertor Issued

BARB has finally issued a summary of the results of the last four weeks of the parallel run, comparing old and new panels.

Definitive comparisons can only be based upon the final four weeks to 4 August, since these represented the period where the size of the panels and the classification of panel homes were at their best, and the weighting and calculation methodology had been finally checked and found replicable.

The new system may be expected potentially to give different results from the old system because the two differ in a number of ways. * The new system is based largely upon entirely new panels, ultimately larger, whose results can all differ from previous results by reasons of sampling error. This cannot be measured but can be deduced when all methodological explanations have been exhausted.

* The new system measures viewing to non-terrestrial channels whereas the old system did not. The contribution of this viewing has therefore been assessed separately and thereafter excluded from many of the comparisons to compare like with like.

* The new system measures the contribution of time-shift viewing(within seven days of recording) where- as the previous system did not. The contribution of timeshift expressed in “consolidated” audiences has therefore also been assessed separately and thereafter excluded from comparison. * Data collection methods differ in respect of viewing by guests which is now attributed according to guest demographics rather than the viewing audience profile. * The new ITV area panels have been recruited on a disproportionate sampl- ing basis whereas the old panels were recruited proportionate to population. This results in over- representation of groups of particular interest, so that their sub-group data are better (whilstcorrectly reweighting them for total sample statistics). This has been achieved by under – recruiting for C2DE Early Inactive households. * The new system is based on targets from a new Establishment Survey. As well as possible contractor effects such as a higher incidence of AB social class, more up-to-date key statistics from the Office of Population and Censuses were incorporated. * The weighting system has also changed. Whereas ITV area data were formerly weighted only by main audience categories, now they are subject to full rim – weighting procedures which redress any imbalances. Previously, only the net-work data was extensively weighted. Now the more comprehensive weighting at all levels is made necessary by the dis- proportionate sampling design.

This has a number of other implications. Previously, the ITV area data (published unweighted) did not match the Network data (published weighted). In practice, total hours of viewing formerly averaged about 3% to 4% higher when published at the ITV area level than for the network.

This means that old versus new comparisons yield rather higher indices for the network where the old data were weighted downwards than the regionallevel where they were not.

COMPARISONS OF HOURS OF VIEWING

In order to compare like with like, it is necessary to relate the old panel total individual hours of viewing to the new panel individual hours of view- ing, excluding non – terrestrial and time-shift viewing (not measured on the old panel). Individuals Live Average Weekly Hours Of Viewing To Terrestrail TV(Network):

The difference between weighted and un- weighted data does not “explain” thedifference between the panel results. One possible source is that the more stringent weighting of the new panels on more variables redresses imbalances neither controlled nor weighted for on the old panels.

Another observed difference between the two panels has been that the new panel included fewer nil viewers than the old during the parallel run.

Taking the level of all-week nil view- ing homes as indicative of homes on holiday and assuming that holiday homes would have had an otherwise normal range of viewing levels, however, the homes on holiday differences could onlyhave accounted for a difference of around 1% of total viewing across the network (ie new panel +1% compared with old).

Another issue is that of possible sampling error. Based on calculations by AGB and incorporating a conservative estimate of the statistical efficiency of a panel in measuring an overall average, then an index of 100 comparing the two panels could vary by +/- 3% at two standard errors (a 1 in 20 level of likelihood).

In conclusion, therefore, the live terrestrial viewing level of the new panel was 10% higher than the old atthe end of the parallel run. A third of this difference could (but not necessarily did) arise from sampling variation. In the absence of other evidence, the rest could be attribut- able to different holiday habits and a combination of revised targets and more stringent weighting on the new panel.

THE EFFECT UPON WEEKLY HOURS OF VIEWING BY SUBGROUP

The degree to which the average difference of 10% for all individuals live terrestrial viewing, between new and old panels, varies by audience category is dependent on a number of factors. * The degree to which the subgroup hasdifferent live viewing patterns on the new panel versus the old. This can arise from sampling variation and can also be affected by the more comprehen- sive weighting used in the new system. * The effect of disturbing guest view- ing according to recorded demographics under the new system compared to pro- rata to the demographics of the viewing audience on the old system.

This must have had the effect of increasing audiences for younger people who are heavily represented among guests and decreasing audience levels for older people compared with the old system. The observed total differences in average hours of viewing, new panel indexed on old, for 4 weeks ending 4/8/91 were:

EFFECT ON WEEKLY HOURS OF VIEWING AT REGIONAL LEVEL

The index of the new panel on the old for weekly hours of live terrestrial viewing for all individuals was 10%. The comparisons for individual areas are:

The variations in the index are within the expectations of sampling variation as calculated by AGB. CHANNEL SHARES

Based on the last four weeks of the parallel run, the channel shares for live terrestrial viewing were very close on the old and new panels, although there has been a small shift away from BBC to commercial channels. Channel Shares Live Terrestl Viewing All Individs 4 W/E 4/08/91, %

GUEST VIEWING LEVELS

The percentage of live viewing attributable to guests was on average about 2.5% points lower on the new panel. Guest Viewing as % Of Total Live (inc. guest viewing)

There are a number of possible sources of this difference. On the old system, guests just had to press any spare button. On the new system they have to register themselves as a guest and then respond to requests for demograph- ics before being allocated a button. On the new system, levels of “unallocated button” pressing can represent up to 2%of viewing.

CONTRIBUTION OF OTHER CHANNELS

The “other channels” not previously measured by the old system include non- terrestrial channels on cable and satellite and in the case of Ulster, viewing to RTE stations. These are now measured on the new panels. Over the final four weeks of the parallel run these stations added 3.2% to total live terrestrial viewing by all individuals. The current level is 4%.

CONTRIBUTION OF CONSOLIDATED

The addition of timeshift audiences to the original live audience added 2.6% to network viewing by individuals. Thepercentage is slightly higher for the commercial channels at 2.9%. The audience groups most affected are ABC1 Adults and 16-44 Adults. Viewing by children is least affected.

There are very wide differences in the contribution of timeshift viewing according to programme type (from 0% to over 20%).

COMPARISON OF IMPACTS

Total Adults Impacts, New Indexed On Old Panel: