To weight or not to weight your survey data has long been a topic of discussion. Weighting data means adjusting segments, such as demographics, to match desired breakouts. For example, say you have 40% men and 60% women in your survey responses, but you want to mirror US Census (49% men, 51% women), you can adjust the data, so men have a weight larger than one (1.23) and women a weight smaller than one (0.85). The breakouts now reflect the general population and allow you to extrapolate your findings accordingly.
This works well for probability samples, helping to address some elements of non-response and selection bias. The caveats being that too much weighing will skew the data, and weighting doesn’t solve for groups or characteristics who didn’t respond at all. Further, there are questions as to whether weighing is viable with non-probability samples.
In a recent webinar, University of Georgia MRII Exec Director Jeffrey Henning provides a thoughtful overview of different sampling types and approaches to weighting data. He shows that recent research-on-research suggests non-probability samples should be weighted and that simple Rim (Random Iterative Method) weighting dramatically impacts the projectability of your data.
Rim weighting, as opposed to targeted weighting, allows researchers to adjust multiple characteristics in a dataset while keeping the characteristics proportionate as a whole. This means that you can run separate weights by data subsets and tailor your weighting schemes by wave, country, or any other data segment within your survey.
The FocusVision Decipher Crosstabs tool uses Rim weighting to apply weighting schemes to report data. There are two easy options:
- Create a Weighting Scheme Editor to create a weighting scheme by selecting questions and assigning weighting percentages to each of their response options.
- Upload a pre-configured weighting file directly to Crosstabs.