Weighting
Consilium takes on the (uni-, bi- or multivariate) weighting of your data!
In the quantitative market research, one is quite often faced with the need to adjust a sample in such a way that it is representative of the target group. In other words, whenever the target distribution suitable or desired for the population has not been achieved in the collected data. This is where subsequent weighting of the data comes into play.
It is still relatively easy to have to take care of the adjustment or weighting of just one characteristic, but it becomes difficult when there are several characteristics or variables that have to be weighted at the same time. Because the target distribution for each subgroup is not always known. For example, with the simultaneous weighting of the three characteristics gender, age and state one would have to know the target distributions of all subgroups (e.g. 30-39-year-old men from Hesse have a share of 3.4%), which is usually not the case for oneself to be able to create a weighting syntax or weighting matrix in SPSS or Excel.
If only the marginal distribution for each characteristic is known, but not the distribution in all possible cells, the technique of multivariate weighting, also called RIM weighting (RIM stands for Random Iterative Method), comes into play. A so-called iterative target weighting process is carried out with the help of software or a tool, with several iterations (repeated application of the same process on intermediate values already obtained) being run through in order to get closer and closer to the desired target values and finally to hit them with pinpoint accuracy. By adapting several characteristics at the same time, the multivariate weighting automatically derives the information about the proportional values in the individual cells of all combinations of characteristics.
When using RIM weighting, however, a few aspects must be taken into account. If a large number of target values are to be achieved, i.e. there are a large number of characteristics and characteristic values that must be weighted, several hundred iterations can easily be necessary. This could lead to sometimes high case weightings, which should generally be avoided in order not to attribute too great an influence on the result to individual cases in the data. The following also applies: the smaller the sample or number of cases in the data, the fewer target values are advisable. Because with a very large number of characteristics or characteristic values, there is a high probability that individual cells will only contain a few cases from the sample and that these may be given a very high weight.
The multivariate weighting is most effective and statistically most justifiable if the values actually collected do not deviate significantly from the target values. This also avoids high weights of individual cases. In addition, the multivariate weighting may not work if some variables or characteristics correlate strongly with one another, since it is not possible to achieve all target values even with many iterations.
Regardless of whether you are an institute market researcher, company market researcher, management consultant or student: We carry out weightings of any kind - univariate, bivariate or multivariate - for you and help you with the evaluation and optimization of the achieved weights and the weighting process. Contact us, because our expertise has a high weight!