Multi-variable weighting

Multi-variable weighting weighs survey data according to the targets specified for answer categories in more than one question or profile variable.

Workflow

To create a multi-variable weight scheme:

  1. Choose the survey questions or profile variables to use as weight variables.
  2. Specify their order in the weighting calculations and set the target percentages that you want for each answer or category.
  3. Apply the multi-variable weight scheme, and the percentages and counts in the report are recalculated based on the specified targets.

Calculation

The system calculates multi-variable weight schemes using random iterative method (RIM) weighting. The RIM weighting algorithm attempts to calculate and balance the target values for all variables in a weight scheme to achieve convergence (that is, to make the data match the targets across all variables). One execution of the weighting algorithm across the data is called an iteration.

During multi-variable weighting, the system sets weight values for each participant based on the targets for the first weight variable. The weight values are then carried forward and readjusted to account for the second variable; this process is repeated for each subsequent variable in the weight scheme. By the time this process reaches the last variable, the proportions for the first variable may not be close to the targets you set. The system will continue running iterations of the weighting algorithm until all the variables match their targets as closely as possible.

Note:
  • The last variable in a multi-variable weight scheme is weighted most accurately, although the difference in weighting accuracy between the first and last variable is minimal.
  • In certain cases (for example, when the variables are too closely related to each other), it may not be possible to calculate weights successfully.
Multi-variable weighting by gender and province example

You distribute a survey to a group of participants. Contrary to expectations, the resulting data does not represent the general population in terms of their province and is not equally balanced between men and women. This could result from a sample that is not equally balanced between groups or from different response rates between groups.

To see what your survey data would look like if it had the expected proportions, you want to weight the data based on gender and province. Because the last variable in your multi-variable weight scheme is weighted most accurately, you put the "Home Province" variable after the "Gender" variable.

These are the targets that you set for "Gender" and "Home Province":

Gender Target Percentage Actual Percentage
Male 50% 45%
Female 50% 55%
Home Province Target Percentage Actual Percentage
Newfoundland 1.8% 2%
Prince Edward Island 0.4% 0.6%
Nova Scotia 2.2% 2.0%
New Brunswick 2.1% 2.1%
Quebec 23.1% 20%
Ontario 38.8% 41%
Manitoba 3.6% 3.0%
Saskatchewan 3.1% 2.3%
Alberta 11.0% 10%
British Columbia 13.3% 16%
YT/NT/NU 0.6% 1.0%
After you apply your multi-variable weight scheme:
  1. The system weights values for each participant based on the targets set for "Gender".
  2. The system carries forward the "Gender" values and readjusts them to account for the participants' weight values for "Home Province".
  3. The system returns to "Gender" to evaluate how closely the Male and Female proportions match the set targets.
  4. If the Male and Female proportions are not close to the targets, the system carries forward the weight values and adjusts them for "Gender" and then for "Home Province" again.
  5. The system runs iterations of the weighting algorithm until the proportions of participants in the weighted data are as close to the targets as possible.