Weighting (AskiaAnalyse)

Weighting allows you to correct the weight of each individual (which by default is 1), for example to correct sampling errors. In AskiaAnalyse, you can create and modify weighting, and you can import these into AskiaVista.

For example, suppose we interviewed 60 men and 40 women. The population we want to describe is equally composed of men and women. To correct this, each time we run counts on a question, instead of adding 1 for each man, we will add 50/60, or 0.8333, and for each woman we will add 1.25. In other words, the under-represented categories will have a stronger effect than the over-represented ones. The process gets more complicated when more than one variable is involved; for an example, and details of the algorithm used, see weighting examples and algorithms.

Note: AskiaAnalyse can create weightings automatically if it finds variables with specific shortcuts. This is achieved by specifying the list of shortcuts in the implicit weight variables option in the options window.

This topic covers the following:

Defining a weighting scheme

To define a weighting scheme in AskiaAnalyse:

  1. Click , which is located on the same row as the weighting list in the general tab of table definition.

  2. Alternatively, in the tools menu, select weighting.

  3. The weighting window appears:

  4. To modify an existing weighting scheme, select it in the name list and then make changes to it as appropriate. See weighting settings below for details.

  5. To create a new weighting scheme, click . Then define your new scheme as appropriate. See weighting settings below for details.

  6. You can load an existing weighting scheme from your library by clicking . You can then edit the scheme, as appropriate. See weighting settings below for details.

Weighting settings

The following settings are available for your weighting scheme:

Option DESCRIPTION

Name

The weighting name (this will be visible in the table/chart).

Description

This optional field allows you to supply a lengthier description of the weighting (note this does not appear in tables or charts).
Level Specifies the data level at which the weight is calculated.
Options Allows you to specify further weighting options. See below for further details.
Base

Designates the number of interviews affected by the weighting, or the extrapolated base. To set this value, you need to clear (de-select) the auto option (see below).

Note: The weighting must affect the whole population. If you want to apply a weighting to part of the population, we invite you to intersect the variable(s) to be weighted by a question that includes the sub-population in question as answer category.
Auto Select this option if you want the weighting to take into account the current base (or the sum of the initial weight question).
Initial weight question

You have the option of using, as the initial weight, a numeric variable that contains the weight of each individual. This allows you to do either of the following:

  • Use weighting produced by another program.

  • If you have very complex weighting that takes a lot of time to calculate, you can save it as a numerical variable then reuse it in another weighting which will be very fast, because it will only need to read the weighting information from a question.

    To specify an initial weight question, click ..., then select the variable and click OK.

Note: The sum of weights included in this numerical variable must correspond to the sample base. If this is not so, the individuals not having a value will not be taken into account.
Set as default

When selected, this option will apply this weighting scheme to any new tables you create. Because the setting is saved to the .QES file, QEW file or .DAT folder, it will be retained as a default setting after the data file is closed and subsequently reopened. 

Note: There can only be one default weighting scheme. When a new default is set, it will replace the previous one. 

 

Weighting window buttons

The following buttons appear in the weighting window:

Button Description

Allows you to start an intersection.

Allows you to end an intersection.

Create a new weighting.

Save the current weighting.

Delete the selected weighting.

Select a weighting from the library.

Add the current weighting to the library.

 

Weighting window information

The following information is displayed in the weighting window:

Once a weighting contains one or more questions, its details appear in the right-hand pane:

Note: If the sum of objectives is different from the sample base  or from 100, AskiaAnalyse will sum the inserted counts and recalculate the weighting structure with reference to the inserted count sum.

 

Note: To know the weight of each individual or to construct a variable containing this information, you can create a weighting variable.
Note: You can load a weighting from your library by clicking fromdico, or store your weighting in the library by clicking intodico.

Further weighting settings (the weighting options window)

The weighting options window contains further settings that can be used to customise your weighting. It is accessed by clicking options in the main weighting window.

Weighting%20options

The following options are available:

OPTION DESCRIPTION

Accuracy (number of decimal digits

Determines the level of accuracy to be used (expressed as the number of decimal digits) that will be used in weighting calculations. A value of 1 is weak precision, a value of 25 is strong precision.

Accuracy (number of decimal digits

Determines the maximum number of iterations that will be used for each question in the weighting. For example, if the number of iterations is set to 30, and there are 3 questions, then only 90 iterations will be allowed. Normally, after a few iterations, the weights should converge to find an acceptable solution. However, this is not always the case. You can increase the number of iterations that askiaanalyse will use to find an acceptable solution here. Note that if you specify more iterations, the calculations will take longer to resolve.
Minimum weight The minimum weight that can be applied to an individual response category. Any weight lower than this value will be adjusted accordingly, and a warning message will be displayed when the table is generated (i.e. when you click the results tab). Warnings take the following form: "Weighting does not converge because of question [question name]".
Maximum weight The maximum weight that can be applied to an individual response category. Any weight greater than this value will be adjusted accordingly, and a warning message will be displayed when the table is generated (i.e. when you click the results tab). The warning takes the following form: "Weighting does not converge because of question [question name]".
Run each wave independently For a .QEW file (i.e. an AskiaSurf project), by default the weighing is applied to all waves together; select this option to apply it to each wave independently.
Use counts to define multiple question targets This option controls how multiple response questions are treated when used in a weighting scheme. If selected, the targets you define will be taken as absolute values (counts) which allows them to function independently of each other, as is required for a weighting scheme based on a multiple variable to be calculated. If deselected, the targets will be treated as percentages - or more properly, as a percentage of a sum of all the targets added together, which is how targets for single variables are treated. The option has no effect on single variables in weighting. 
Run question in their level Select this option if the weighting scheme needs to consider all of the responses for a question that belongs to a sub-level. Where multiple responses at the sub-level fall into more than one target, and the weights are applicable at the higher level, the weights at the lower level are averaged before being applied at the higher level. 
Weight aggregation When the option Run question in their level is selected, this option allows you to select the method by which the weights at a sub-level are averaged prior to their application at the higher level. A drop-down menu allows you to choose between geometric mean (the default) or arithmetic mean. 

 

Create your own Knowledge Base