Tags:
• Creator
• Explorer
• Viewer

In this article

1. Unit of Count
2. Ratio
3. Multi-Level Analysis Example
4. Multi-Level Default Weights
5. Relationship between Levels
6. Low Sample Size Calculation

1. Unit of Count

In an analysis using a multi-level project, you get the choice of seeing the results based on the different ways of counting that are inherent to the levels of the data in your project. For example, the respondents (people) and/or their consumption occasions (occasions). 

At the top of every analysis, you can identify the count that is used.  The default unit of count is the primary level, for example, Counting: People.

You can interact with the different levels (e.g., people/occasions), by changing the unit of count using the counting option (selector) in the interactive title. There is no need to specifically include an item from a child level.  

The default mode for the unit of count selector is replace. To see multiple units of count in your analysis, your need to switch to add and then select the relevant options.

2. Ratio

On a multi-level project, you can calculate the relationship between levels by selecting the ratio option (N:1) under the modify menu.

This option calculates the number of times one value contains or is contained within the other, for example, the Ratio of Occasions to People.

The number of ratio options is dependent on the number of levels in the project.

3. Multi-Level Analysis Example 

Ice Cream Brand Tracker is a multi-level survey that includes questions about people, brand awareness, and brand consumption as well as questions about their ice cream consumption occasions within a given day (24 hours).  The survey includes respondents aged 18 to 44 years old.

Ice Cream Category Consumers

Cool Nut Consumption Past 24 Hours

For every occasion, respondents are asked what size of Ice Cream they consumed. Cool Nut is a key ice cream brand.

• Small ice cream -  is the equivalent of 150 ml.
• Medium ice cream - is the equivalent of 240 ml.
• Large ice cream - is the equivalent of 360 ml.

This table describes the results from the Harmoni analysis included below this table.

In the table below, the Total row has the percentages set to Total Base (Total Only) which allows you to see a row percentage for the total. Total percentages are calculated across all the brands in the row (Note: only Cool Nut is displayed here, but there are 9 brands in the axis.)

Values have also been added to the Size axis to calculate the volume using the ∑ (Sum Value) calculation type.

4. Multi-Level Default Weights

Multi-Level projects commonly have weights at the respondent (parent) level, although it is possible to have weights at the child level. When creating the project, the project owner includes the required default weight(s).

Here the default weight for the project is defined at the respondent level with the unit of count labeled People.

The following examples outline the application of default weights.

• If each unit of counting has its own default weight specified, the analysis will use each level-specific weight.

• If a unit of counting doesn’t have its own default weight specified, its closest ancestor’s weight will apply.

Overriding Default Weights

Users can override the default weight by dragging a weight into the analysis.  In the first release, only one override weight can be dragged in.  If a user drags a weight into the analysis, that weight will apply to the counts for that unit of counting, and every descendent below that unit, overriding all default weights.

For example:

• If the user drags weight A into the analysis, A, B, C, D, E and F will be weighted using the A weight
• If the user drags weight B into the analysis, B, D and E will be weighted by B, and A, C and F will become unweighted
• If the user drags weight C into the analysis, C and F will be weighted by C, but A, B, D and E will be unweighted
• If the user drags weight D into the analysis, D will be weighted by D, but A, B, C, E and F will be unweighted
• If the user drags weight E into the analysis, E will be weighted by E, but A, B, C, D and F will be unweighted
• If the user drags weight F into the analysis, F will be weighted by F, but A, B, C, D and E will be unweighted

When a weight from a parent is applied to its child, each child record is assigned the value of the parent weight

For example:

• If a record in A has 3 records in B, with 4 D records, and 5 E records, they will have the following weighted counts:
• A = 2 (unweighted = 1)
• B = 6 (unweighted = 3)
• D = 8 (unweighted = 4)
• E = 10 (unweighted = 5)

5. Relationship between Levels

When there are multiple child levels within a multi-level project, there is no direct relationship between the siblings. They each have a relationship between the parent, but not with each other. 

In a project that has a parent level of respondents with two child levels, e.g., campaigns (counting ad campaigns) and brands (counting brands), running the two child levels together will not use the respective units of count.
 

6. Low Sample Size Calculation

When calculating the low sample size in multi-level projects, Harmoni uses the unweighted respondent data when assessing the sample size. With counting set to the child level, e.g., occasion data, Harmoni uses the parent level data to determine the low sample size and not the response level. 

Learn more about low sample size warning.

Where to from here?

Learn more about analysis.

• Multi-Level Data Overview
• Total Responses (Share of Mentions)
• Selector - Interacting with your analysis selections