Explore key variables, examine relationships between variables, data manipulation.

Introduction

Types of variable

Variable distributions

Survey weights

The terms survey weights, weighting variables and weighted data may be unfamiliar or confusing.

You might be tempted to avoid thinking about survey weights. However, you should be aware that your results are likely to be incorrect if you do.

Luckily, with some background information and a little work, using survey weights is relatively straightforward.

What do weights do?

Survey data comes from a sample of a population. Weights help make the sample data representative of the population.

The main way of getting a representative sample is to select cases at random. However, a randomly selected sample might not be representative because:

- some people are less likely than others to agree to participate in surveys (often termed non-response bias)
- common way of selecting a sample can make some cases more or less likely to be selected (referred to as unequal selection probabilities)

Survey producers create weights in response to these problems. The weights adjust a sample to make it more representative of the population it was designed to reflect.

How do weights work?

Weights appear in survey datasets as a variable, which assigns a value to each case to indicate how much ‘weight’ it should have during data analysis.

A weighting variable can make several adjustments to the data; for example, it can simultaneously adjust for non-response and unequal selection probabilities.

The weights in your dataset

Most large survey datasets include at least one weighting variable. You need to use the weight for your results to reflect the population accurately (unless you take other steps to avoid biased results).

Procedures for applying weights are relatively simple but vary across statistics software packages.

Typically, you will specify the weight before you start analyses.

Which weight?

Survey datasets often contain more than one weight. Different weights may relate to

- different samples (such as a ‘core’ and ‘ethnic boost’ sample)
- sample units (such as individuals or households).

The details of any weights will be included in the survey documentation. It is important to read the documentation carefully to find out which weight to use for your analysis.

If unsure, you can contact the UK Data Service helpdesk.

Grossing weights

Weights can also adjust a sample to make it look the same size as the population. These are called grossing weights.

Grossing weights are useful when describing the prevalence of social phenomena in the population, such as incidences of crime. They are used in some official surveys including the Crime Survey for England and Wales and the Labour Force Survey.

When using grossing weights, results from data analysis will look like they come from a sample of millions, rather than a few thousand; thus, results can appear more precise than they might be in reality. As a result, some researchers prefer to rescale the weights to stop the artificial inflating of the sample size.

The ‘Data Analysis Workbook’ that accompanies this guide includes more on using grossing weights.

Further reading and resources

The following UK Data Service guides focus on weights in more detail

- An introduction to using weights in social surveys (Video)
- What is weighting?
- The guides to using Stata and SPSS also include sections on weighting.

Missing data

More than one variable

Creating new variables

Worksheet