Learn how weighting is applied to your report data.
Random Iterative Method (RIM) weighting is used to
calculate the weight factors that are applied to adjust your data when you
create a weighting scheme with one or more variables. This process is also
commonly referred to as
Raking or
Iterative Proportional Fitting.
When RIM weighting is applied, it iteratively adjusts the weights of
respondents step-by-step until the sample matches the target population for
each variable in the weighting scheme. The algorithm applies the following
processing to the survey data:
- It adjusts the weights for
one variable at a time, so that the sample distribution matches the population
target for that variable.
- After adjusting one
variable, it adjusts the next variable, and continues to iterate through this
process for each variable.
- The iterative adjustment
of each variable in turn continues until the sample matches the target
proportions for all variables simultaneously, within an acceptable margin of
error. This point when all variables closely match the target proportions is
called convergence.
Note:
- In multi-variable
weighting, the last variable is weighted most accurately, but 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 produce a weighting scheme.
RIM weighting for gender and age example |
---|
You have conducted a survey and you want your sample to match the
known population proportions for age within each gender.
Survey response proportions
|
Target population proportions
|
The results of your survey include the following
percentages:
- Gender:
- Age:
- 18-34: 40%
- 35-54: 31%
- 55+: 29%
|
The desired percentages after weighting are the following:
- Gender:
- Age:
- 18-34: 30%
- 35-54: 40%
- 55+: 30%
|
The distribution of the unweighted survey responses includes the
following values:
Gender
|
Age
|
Unweighted responses
|
Male
|
18-34
|
20
|
Male
|
35-54
|
16
|
Male
|
55+
|
14
|
Female
|
18-34
|
14
|
Female
|
35-54
|
10
|
Female
|
55+
|
10
|
The RIM weighting algorithm calculates the following adjustments by
dividing the target population proportions by the survey proportions:
-
Iteration 1: Adjust the Gender variable
Adjust the weight factors to match the gender targets:
-
Males are overrepresented by 10%, so their weight is
adjusted down. The weight of each male response is reduced.
Male weight = 0.5/0.6 = 0.83
-
Females are underrepresented by 10%, so their weight is
adjusted up. The weight of each female response is increased.
Female weight = 0.5/0.4 = 1.25
In iteration 1 the unweighted responses are multiplied by the
Gender weight factors:
Gender
|
Age
|
Weighting Calculations
|
Weighted Count
|
Male
|
18-34
|
20 * 0.83 = 16.6
|
16.6
|
Male
|
35-54
|
16 * 0.83 = 13.28
|
13.28
|
Male
|
55+
|
14 * 0.83 = 11.62
|
11.62
|
Female
|
18-34
|
14 * 1.25 = 17.5
|
17.5
|
Female
|
35-54
|
10 * 1.25 = 12.5
|
12.5
|
Female
|
55+
|
10 * 1.25 = 12.5
|
12.5
|
The
Weighted Count for the Male entries
increases, and it decreases for the Female entries.
-
Iteration 2: Adjust the Age variable
Adjust the weight factors to match the age targets, using the
current weights from the gender adjustment:
-
Reduce the weight of the 18-34 group.
18-34 weight = 0.30/0.40 = 0.75
-
Increase the weight of the 35-54 group.
35-54 weight = 0.40/0.31 = 1.29
-
Increase the weight of the 55+ group.
55+ weight = 0.30/0.29 = 1.03
In iteration 2 the previously calculated weighted response
values for Gender are multiplied by the Age weight factors:
Gender
|
Age
|
Weighting Calculations
|
Weighted Count
|
Male
|
18-34
|
16.6 * 0.75 = 12.45
|
12.45
|
Male
|
35-54
|
13.28 * 1.29 = 17.13
|
17.13
|
Male
|
55+
|
11.62 * 1.03 = 11.97
|
11.97
|
Female
|
18-34
|
17.5 * 0.75 = 13.13
|
13.13
|
Female
|
35-54
|
12.5 * 1.29 = 16.12
|
16.12
|
Female
|
55+
|
12.5 * 1.03 = 12.88
|
12.88
|
The values in the
Weighting Calculations column are the
values calculated in iteration 1 multiplied by the Age weight factors.
-
Iteration 3: Re-adjust the gender variable
Repeat the adjustment for gender using the updated weight
factors calculated for the Age variable.
-
Iteration
n: Re-adjust age and gender until convergence
Continue to adjust each variable in turn using the weight
factors calculated for the previous variable until the sample matches the
target proportions within an acceptable margin of error.
The final set of weight factors can be downloaded using the
Export Weight Factor .XLSX link. The output
includes the final weight factor for each combination of variables (e.g. Male
18-34, Female 55+, etc.) after all of the iterations have completed and
convergence is reached.
|