|
|
Market
Research
Business Intelligence
Tools & Resources
Member of MRIA
|
Marginal Weighting
Map Your Survey Data to the Whole Population
Objectives of marginal weighting
One of the biggest challenges in any Marketing Study or Survey or Poll is how to make the survey results represent the general population. The solution to this challenge is to assign marginal weights to each respondent.
The objective of the marginal weighting process is first to generate the sum of the weights so that the weighted survey results from the respondents can be considered as if generated from the whole population. Secondly the weighting must ensure that the gaps between the aggregated weights and the population by the given geo-demographic conditions when slicing and dicing the respondent file are as close as possible.
Another benefit of this weighting process is that it can adjust some distortions in the respondent distributions. For example, while the general male/female distribution is 49%/51%, the respondent distribution may actually be 53%/47%. The marginal weighting process can eliminate this kind of distortion by assigning low weights to male respondents and high weights to female respondents.
Special conditions for marginal weighting
Three conditions give rise to challenges when applying marginal weights and require specialized treatment.
Overlapping geographic areas:
Most studies use standard geographic definitions such as Census Subdivision, which do not have cross-over problems. Some studies however are based on customized markets which are special areas offering certain products/services and these products/services markets can overlap each other. In such a case, it is difficult to allocate the population at market level when assigning the proper weights in these markets.
2 or more demographic variables:
It is easy to assign weights based on a single demographic variable (for example, Income). Utilizing a simple tool such as Excel can do the job. But when the weighting process has to be applied to 2 or more demographic variables, the balanced weights will fluctuate as the demographic variable is altered from one to another.
Insufficient respondents:
If there are too few respondents in some cells, the “empty cell” problem arises. The lack of responses will prevent the weights from being assigned properly by the pre-defined geographic areas and demographic breakdowns.
SM Research’s solution: Multivariate Converging Iteration Process
SM Research has a specially designed Multivariate Converging Iteration Process. In addition to simple weight methods that can assign weights by one selected geographic level such as Census Subdivision, City or a customer defined market and one demographic condition, this process can assign multiple key demographic variables. These key variables can include the most commonly used variables of Age & Gender, Household by Size and Household Income or customer specified variables such as Day of Interview and Transit Pass Holder distribution.
SM Research’s Multivariate Converging Iteration weighting method has been used for many national surveys and transit related studies, such as Combase Newspaper Coverage Study, Vancouver Transit Rider Study, BBM RTS Re-weighting for COMB Transit Model, etc.
E-mail or contact us for more information.
|
|