Survey results microdata, such as the NI Life and Times Survey, typically originate from a sample set of people that may differ slightly from the characteristics of the general population. For this reason, the analysed results take into account a balancing measure used to correct differences between the survey participant and the general population, called a weighting factor.
It is important that users of these datasets apply weighting factors when summarising or analysing results.
In the Life and Times survey, the weighting factor is present in a column called
wtfactor, a fractional number that indicates the strength of readjustment needed to match with the general population. Therefore, in summarising responses in a multiple choice column ("Strongly agree", "Agree", "Neither" etc.) we need to sum
wtfactor for each response, rather than simply sum the number of each of the responses. This column as it appears in the 2014 survey dataset is show in the image below.
Statistical software packages (e.g. SPSS) often contain tools for applying weightings to results during analysis, but the same results can be achieved with standard spreadsheet packages, such as Microsoft Excel and other non-proprietary software.
When using a CSV dataset in Excel or OpenOffice, the PivotTable tool offers an easy way to achieve this. If we first select the variable (i.e. survey question) that we want to summarise, by using the sum of
wtfactor the weighted results for that question will be returned.
To do this in Excel create a PivotTable when you have the dataset opened. Then select
wtfactor from the list of fields and ensure that that
value field is summarised by
sum. From the list the desired variable can then be selected for the rows of the PivotTable.
This is demonstrated in the screenshot below, counting results from the NI Life and Times Survey 2013 for the
sport1 variable ('My experience of sport at school has given me a lifelong love of sport'). The weighted and unweighted results for each response to this question are shown for comparison, but it is the weighted results which should be used in your analysis.
Checking against ARK's published results for this question, we can see that the percentage for each response (ignoring the NA values) matches once the weighting factor is applied.
The differences are relatively small, but they do matter in every case. If you have any further questions, we're happy to help.