Weight Variables Overview
When a survey is fielded, it can be difficult or impossible to get the correct number of participants from any demographic or persona to represent a group of consumers accurately. For this reason, respondents can be given different values, called weight values, which affect their influence on the response data. In short, weighting adjusts the sample so that it better reflects the desired population.
An example of this might be the following:
On a survey from a car brand, 700 females and 300 males were asked about what type of car they drive to work. This group of 1000 respondents is called the sample.
However, the car brand knows that the market is comprised of 50% male and 50% female. This 50/50 makeup is called the target population.
Because of this discrepancy between the sample and the target population, the respondents must be weighted so that males have an equal representation as females. This math can be quite complicated, but it will result in an output that more accurately represents the market.
KnowledgeHound offers those with the "Manager" role the ability to switch between weighting calculations, called weight variables, and create their own weight variables* while analyzing the data. This article will describe the capabilities of the feature, and you can also watch this video for a demonstration.
Weight Variable Toggling
Many datasets loaded to KnowledgeHound have variables intended to be used for weighting. KnowledgeHound recognizes those and makes them available to be used in that way. One of these variables is chosen during data processing to be the default weight.
The option to use whatever the default weight is, or select another available weight for a given dataset can be found in the "Edit weighting" drop-down menu, pictured here:
Note that these variables can only be used for the dataset on which they appear and cannot be used elsewhere.
Occasionally, a weight variable will be shown in red with a message saying it cannot be used. If the data was reloaded after creating the variable and the variable no longer includes weight values for the entire population, it can no longer be used.
As pictured above, you can also choose not to weight the dataset at all (which may also be the dataset's default option).
Selecting a different weight variable will cause the analysis page to reload the visualization with new weight values assigned, so percentages and base sizes will change. If you exit the analysis page and come back later, the weight variable will reset to the default. If a different weight variable should be made the default, reach out to KnowledgeHound using the chat bubble on the bottom-right corner of the page.
Important note about pinning items to Stories and Summary Dashboards:
Within the weight selection drop-down that there is an option called "Use dataset default".
When a visualization is added to a story or summary with this "Use dataset default" option selected, that visualization is subject to change if the default weight for the dataset ever changes. However, if any other weight option is applied to the pinned item, it will not change even if the default changes.
Weight Variable Creation
Beyond the pre-determined weights, KnowledgeHound allows users to create their own weight variables. This is helpful for making sure you can accurately represent populations that would otherwise be very difficult to account for.
Below the list of standard weights in the "Edit weighting" drop-down menu are custom weights along with the ability to create a custom weight variable.
Custom weights are user-created weight variables and can be deleted only by their author.
To build a custom weight variable, select the "Create a custom weight" button to get started. You'll be brought to the weight creation page where you'll first select the variables needed to determine the target population. Up to 2 single-select categorical variables from the survey can be used. Variables are only eligible to be used as weights if the entire sample has been exposed to and responded to them.
When the appropriate variables have been selected, their responses will show on the right side of the page. If two variables have been selected, the possible response combinations will be shown there. These responses represent the groups of respondents who responded in those ways. Next to each respondent group is a percentage field that allows you to choose what percent of the population is made up of a particular group. These percentages must equal 100% to account for a complete sample. By default, the target population will be evenly distributed* between each group of respondents, but those percentages can be changed to reflect the desired targets.
When the target populations are set and equal 100%, you may name and save the weight variable. Now that your variable is created it can be used by other managers using this dataset, but can only be deleted by you.
*: with an odd number of responses, even distribution of the target population is not currently possible.