Often, data collected from surveys isn’t exactly representative of the target population even in those instances where quotas have been applied (see our blog on how to get representative samples for more on quotas). Weighting is a statistical technique that can be used to correct any imbalances in sample profiles after data collection.
Imagine we have a target population that is evenly split by gender. If we then interview a sample of 400 people within this population, 300 of whom are male and 100 female then we’d know that our sample over-represents men.
Weighting the resulting data can help us to correct this imbalance. The target proportions for both men and women are 50%. The proportion of men would therefore need to be “downweighted” from 75% (300 out of 400 interviews) to 50% while the proportion of women needs to be “upweighted” from 25% to 50%.
In this case weighting would multiply the existing female interviews by 2, so that the female response is amplified in the data. For example, on the gender question we have 100 people answering female but after weighting this becomes 200 as the "female" data is counted twice.
The male interviews need to be correspondingly downweighted. In this instance we need to get 300 responses to effectively count as 200 so we multiply the male responses by two-thirds (or 0.67). Before weighting we have 300 males coded on the gender question. Multiplying by two-thirds gives us 200 males, equalling the number of female responses after weighting.
The numbers used to multiply the responses from each proportion of the sample are called weighting factors. A summary of the weighting factors for this example is shown below:This is a very simplistic example, used only to illustrate the concept of weighting. Typically weighting is used to match the population profile on more than 1 variable to get as representative a sample as possible. For example, to get a representative sample of a country’s population we might weight on a number of demographic variables such as gender, age, region and social grade.
This post has been written to provide a very basic understanding of weighting. In reality analysts weight survey data using specialist software such as SPSS. To be able to instruct these analysts how to weight your data you will need an understanding of the different types of weighting available and be able to understand the effect of weighting on your data. Weighting can change the structure of your data in an adverse way, so caution is required when applying it. It is, for example, inadvisable to upweight small groups of respondents so that they account for a significant proportion of the total sample as this will mean that survey results are disproportionately affected by a small minority of respondents.
Resources available on the web that go into more detail on weighting seem to be rare but for those interested in further learning a more detailed discussion on weighting can be accessed here - http://www.spsstools.net/Tutorials/WEIGHTING.pdf.