Stats and Data Science
Playing the weighting game: How Rim Weighting Works
Ben Ithier
October 1, 2021

The moment you tell someone outside our industry how TV audiences are measured they normally are surprised that a panel of around 5,100 homes can accurately measure the country’s viewing. This blog explains how the BARB system ensures that audiences are accurately calculated by applying weighting to each individual on the panel and represent the UK population across a variety of demographic categories by region.

##### How many people could be watching?

A pre-requisite of accurate weighting is knowing the population size that your sample represents. This isn’t as simple as knowing how many of one demographic there is in the population (eg how many 16-24 year olds), we also need to know how they are regionally dispersed, whether they are in homes with Sky and so on. The BARB system uses population estimates (universe projections) sourced from government as well as specially commissioned Establishment Survey data. The latter provides detailed information required for BARB that isn’t accurately available elsewhere, for example number of TVs in homes, satellite and cable penetration and so on.

##### Why weight the sample?

Weighting is necessary to ensure that the weighted panel as a whole represents the latest regional population estimates by key demographics such as age, ethnicity and reception. It compensates for disproportionate weighting that is designed into the system (for example to ensure that every TV region has a sample size large enough for its audiences to be reported) and also compensates for any imbalances between the panel as it is designed and the actual reporting sample. In an ideal world the reporting panel would be exactly as it is designed to be, but in practice at any one time there are factors that mean that there is imbalance; this could be anything from some sample groups being harder to recruit to temporary glitches such as power cuts. Without weighting some groups would be over- or under-represented and this would bias the measurement.

##### Weighting Approaches

So how do you ensure that a single rim weight assigned to each panel member is representative of multiple demographic categories?

One potential method would be to employ single matrix weighting, where the weights are computed so that the sample totals conform to the population “universe” totals for each demographic “control”group. This means that the weighted sample for each combination of interlaced controls would match the target total. For example, only taking gender, age and social grade as the control groups, would mean that the weighted sample of any interlaced group such as Men 16-34 ABC1 would match the universe target. While this is ideal for when there are only a few control groups there are problems in terms of fragmentations when more demographic groups are targeted, as would be the case with BARB. This approach would be likely to cause issues: the fragmented interlaced groups having missing samples, extreme weights, multiple universes and so on.

BARB adopts an alternative process called rim (random iterative method) weighting, which is an iterative mathematical technique/algorithmic method that targets each of the marginal control groups (e.g. gender, age and social grade) separately. Using this method means there is no requirement to directly control interlaced demographic groups such as Men 16-34 ABC1, the process converges on the best possible weighting solution to deliver correct universes.

The BARB rim weighting process targets a series of marginal demographic profiles at regional and network level. One of the main benefits of this method is that it enables many variables to be controlled simultaneously. These rim weighting target groups have been identified as being key determinants of viewing and form part of a specially designed rim weighting configuration. The degree of weighting is dependent upon reporting requirements and whether population data is available for appropriate universe estimates to be calculated. So, there is scope for further controls to be added into the weighting process as and when they become available.

Rim weighting works iteratively to gradually ensure the sums of weights for all reporting panel members match each of the demographic categories that are being controlled.

##### RSMB’s role

RSMB is responsible for the design and maintenance of the BARB panel rim weighting to ensure its optimal performance for those reporting BARB panel homes; this requires regular reviews of the panel sample and configuration to ensure the risk of any rim weighting failure is low.In exceptional circumstances causing large changes to the panel, such as the COVID-19 outbreak, RSMB has set up diagnostic tools and robust processes to aid further investigation. Our highly experienced team is ready to tune the rim weighting configuration to ensure day to day operations remain unaffected by these extreme events. In addition, experts at RSMB are always on hand to find solutions if ever the mathematical solution fails – which it rarely does.

Every single day the rim weighting process is applied to the network and ITV regions and typically are over 1,400 unique target profiles with weights calculated for each of the BARB sample of over 12,000 individuals!