The process of weighting consists of changing the weight of the interviews to correct sampling errors.
How to create a weight?
You can have a look at the demo 🎬 or the documentation 💡.
Can I use a numeric variable as weight?
Yes, it's possible. A common scenario is that you have imported your dataset and you have already a weight variable that you want to use in your tables.
In this case, you just have to select the appropriate numeric question and set it as the initial weight. You can also apply multiple weights.
The only requirement with imported weights is that the mean of the weights must be equal to 1.
Can I show weighted and unweighted base on my table?
Yes, you can fully customize the calculations on a table. Click here to learn about showing weighted and unweighted bases.
Where can I find the weighting report?
The report is generated in the survey.dat directory.
Check the option in Analyse/Tools/Options/Weighting => Write a report after each weighting.
How I can measure the weighting efficiency?
The weighting efficiency is an indication of the amount of skewing that had to be done to get the weights to converge; the closer this figure is to 100%, the less skewing that was needed to be done.
You'll find it in the Weighting report.
To know more about your sample reliability, see 🎓Weighting Efficiency article
My weighting does not converge
A weighting convergence error can occur for a number of reasons:
- When your weighting scheme falls outside the minimum or maximum bounds set in the weighting options.
- In order to converge, the weighting algorithm needs to iterate more than is allowed by your settings.
- You are attempting to weight a grouping to a target which is impossible e.g. the grouping has zero counts and you are weighting it > 0.
- Less common, the weighting never actually converges beyond a certain number of decimal places of accuracy. For this, you can reduce the accuracy setting.
Check your weighting:
1. Create a variable by weighting
Check the mean 'weight' score - i.e. add up all the weights and divide by the number of cases. The result must be equal to 1.
The maximum value must be under the maximum weight.
2. In the weighting report, look to see if there is any incoherency in the targets ascribed.
4. If the above three steps do not solve your problem . . .
Try to remove a variable or a group of responses, in order to set more reachable targets.