Weighting allows you to re-balance the contibution that different sub-populations make to a table such that the total for one group that over-represented in the sample can reduced, while the total for another sub-population that is under-represented can be increased.
In AskiaAnalyse, you can create custom weighting schemes and apply these to any of the tables you set up.
In AskiaVista, you can only apply weighting schemes that have been created in AskiaAnalyse.
Weighting provides a way to adjust the contribution that each interview or survey record makes to the table so as to counteract any distortions that may arise from some subgroups being over-represented in the sample of interviews achieved, while others are over-represented, when compared to the population at large. These distortions could affect the overall results and any conclusions drawn from the analysis. Weighting provides an arithmetic way to adjust the proportions of the sample at an individual case level.
By default, each case (i.e. each respondent record) will contribute a count of 1 to the table. In other words, the default weight can be considered to be 1. By defining a weighting scheme, you are deliberately adjusting how each record is counted, such that some records will now contribute more than 1 to the total while others will count as less than 1. If a group of respondents is over-represented, their influence can be reduced by giving them a weight factor of less than 1. Similarly, respondents who are under-represented can have their influence increased by giving each one a weight factor of more than 1. Overall, the set of weights, when applied, will be calculated in such a way to ensure the overall sample size remains the same. Thus a survey of 1000 individuals would show both an unweighted and weighted base of 1000. However, the weighted and unweighted totals for other subgroups, such as those across the columns of a table, are likely to diverge, showing the effect of the weighting on each subgroup.
For example, a survey of 1000 adults was carried out in a country where 50% of the adult population is known to be male and 50% is female. Unfortunately, 100 more females participated than males, with the result that the survey shows a total of 550 (55%) females and 450 (45%) males. In order to reduce the distorting effect this could have on some answers in the survey (in particular for questions where being male or female is likely to make a difference to the responses given), weighting could be applied to even out this observed imbalance, by making the contribution of each sub-population exactly 50%. In this example, a weight factor of 0.9091 would be applied to each female respondent and a factor of 1.1000 applied to each male respondent to bring the totals to 500 females and 500 males. Each respective weight factor is calculated by dividing the required total (here, 500) by the actual total (e.g. 550 for females): 500 / 550 = 0.9091.
So long that the average weight applied remains 1 then the weighted and unweighted base values will remain the same, even though the proportions and therefore the totals for different subgroups will have changed.
For further examples and a description of how weights are calculated see Weighting examples and algorithms (AskiaAnalyse).