How to Compute Share of Wallet, Spend, Mouth, etc. (i.e., Volumetric Analysis)
Where percentages are computed from numeric data, rather than categorical data, the percentages are variously known as:
- Share of wallet (particularly in banking).
- Share of category requirements (particularly in packaged goods).
- Share of throat (particularly in beverages).
There are two different widely used formulas for computing such share statistics:
- The volumetric formula: this method gives greater weight to respondents with higher values. For example, if one person has drunk 1 Pepsi and 0 Coke, that person's share for the brands is 100% and 0% respectively. If a second person drunk 0 Pepsi but 9 Coke, their shares are 0% and 100%. The combined shares are then 10% and 90% for the two brands (i.e., 10% of all consumption was for Pepsi and 90% for Coke).
- The unweighted formula: this method computes shares for each respondent and then computes their averages. This, continuing the example, with this method each brand has a share of 50%.
Using the unweighted formula in Q
If a question is is set up as a Number - Multi question in Q, and the data has been created in a way that it sums to the same value for each respondent (i.e., is constant-sum data), then the unweighted formula is automatically computed by selecting the containing the quantity information in the Blue or Brown Drop-down Menus and selecting whichever is appropriate of the following Statistics:
If the data is volume weighted (i.e., respondents values do not all sum to the same amount), the volume component can be removed by either:
- Using Create New Variables - Case-Level Shares.
- Creating a weight which is the inverse of the volume.
Using the volumetric formula in Q
If a question is is set up as a Number - Multi question in Q and the data is volume weighted, the volumetric formula is automatically computed by selecting the containing the quantity information in the Blue or Brown Drop-down Menus and selecting whichever is appropriate of the following Statistics:
Where the data is not volume-weighted, the most straightforward approach is to create a variable that reflects the volume, and set this variable as a weight and then choose the appropriate share Statistic. Where there is already a weight, it can be combined with the volumetric weight using the method described in How To Combine Multiple Weights.
Stack the data to simplify analysis
Where the questionnaire contains a looped structure (e.g., asking people to list all their purchases in a series of occasions or brands) it is often useful to first stack the data prior to attempting to compute share figures. Where there is a lot of data it may be advantageous to treat it as Panel Data.
Both stacking and treating the data as panel data will greatly simplify further analyses, allowing comparisons of share figures by segment and other questions through crosstabs and the use of filters, without the need to construct new variables.
Warning about missing data
All of the methods described on this page will give the wrong answer if respondents have some missing data for some values. That is, respondents need to either have complete data or completely missing data.
Warning about statistical significance
Some caution needs to be taken when interpreting statistical tests for share figures. Q's default approach to statistical testing on tables showing Averages, tests for differences between averages. Where the Average is not a multiple of the share, the results of the significance tests will be misleading.
An alternative approach to conducting the testing is to instead transform the data. See How To Test Differences in % Column Shares.