User Features Guidelines

Rule 1: Single-Select Per Category

Principle: Each participant selects exactly ONE feature per category.

Implication: Features within a category must be mutually exclusive.

Validation: If two features seem like “both could apply”, they belong in different categories.

Rule 2: No Negation Features in Groups

Prohibition: Negations should be defined as separate yes/no categories.

Rationale: To avoid features overlapping within categories.

Implication: There is currently no implementation for a user to be auto-selected for their feature based on a parent category. They will have to manually select each category.

❌ BAD:

{"categoryName": "Industry", "featuresName": "Technology"}
{"categoryName": "Industry", "featuresName": "Finance"}
{"categoryName": "Industry", "featuresName": "Healthcare"}
{"categoryName": "Industry", "featuresName": "Non-tech"}

✅ GOOD: Use non-overlapping features instead.

{"categoryName": "Industry Technology", "featuresName": "Yes"}
{"categoryName": "Industry Technology", "featuresName": "No"}

Rule 3: No Catch-All Features

Prohibition: Features must NEVER be catch-all bins like “Other”, “Traditional”, “Miscellaneous”, “General”, or open-ended ranges like “55+”, “200k+”, “12+”.

Rationale: Catch-alls prevent future granularity and scalability. If a client wants to study the 300k–400k income segment but you already have “300k+”, users earning 300k–500k are trapped in that bin and cannot be re-segmented for granular analysis.

Scalability Benefit: Specific bounded ranges allow future studies to add more granular segments without invalidating existing user selections. Participants who cannot select an existing option can choose “Rather not say” to opt out temporarily until a valid feature exists.

❌ BAD:

{"categoryName": "Work Sector", "featuresName": "Traditional Industry"}
{"categoryName": "Income", "featuresName": "200k+"}
{"categoryName": "Age", "featuresName": "55+"}

✅ GOOD: Use specific, atomic features that don't subsume others.

{"categoryName": "Industry", "featuresName": "Technology"}
{"categoryName": "Industry", "featuresName": "Finance"}
{"categoryName": "Industry", "featuresName": "Healthcare"}
{"categoryName": "Industry", "featuresName": "Retail"}

Rule 4: Mutually Exclusive Numerical Ranges

Requirement: Numerical ranges must have NO overlapping boundaries.

Rationale: Overlapping ranges create confusion. Someone earning exactly $100k shouldn't have to guess between “75k-100k” and “100k-150k”.

Pattern: For continuous bins, use half-open ranges like “75k-<100k”, “100k-<150k”, “150k-<200k”.

❌ BAD:

{"categoryName": "Salary USD per Year (width 50k)", "featuresName": "50k-100k"}
{"categoryName": "Salary USD per Year (width 50k)", "featuresName": "100k-150k"}

✅ GOOD:

{"categoryName": "Salary USD per Year (width 50k)", "featuresName": "50k-<100k"}
{"categoryName": "Salary USD per Year (width 50k)", "featuresName": "100k-<150k"}

Rule 5: Consistent Bin Sizes

Requirement: Within each category, all numerical ranges should use consistent bin sizes.

Rationale: Inconsistent bin sizes create confusion.

Application: If inconsistent bin sizes are required in the study, leverage different category names, but avoid overlap due to multicollinearity.

❌ BAD:

{"categoryName": "Salary USD per Year", "featuresName": "0-<25k"}
{"categoryName": "Salary USD per Year", "featuresName": "25k-<100k"}
{"categoryName": "Salary USD per Year", "featuresName": "100k-<150k"}

✅ GOOD:

{"categoryName": "Salary USD per Year (width 25k)", "featuresName": "0-<25k"}
{"categoryName": "Salary USD per Year (width 25k)", "featuresName": "25k-<50k"}
{"categoryName": "Salary USD per Year (width 50k)", "featuresName": "0-<50k"}
{"categoryName": "Salary USD per Year (width 50k)", "featuresName": "50k-<100k"}
{"categoryName": "Salary USD per Year (width 50k)", "featuresName": "100k-<150k"}
{"categoryName": "Salary USD per Year (width 50k)", "featuresName": "150k-<200k"}

Rule 6: Use Measurable or Specific Definitions

Requirement: When an attribute can be measured, define it with explicit units and bounded values or ranges instead of subjective labels. When an attribute is not naturally numeric, use specific, atomic categories.

Rationale: The goal is not to force every feature into numbers. The goal is to avoid vague labels such as “high,” “medium,” “low,” or “frequent” that different people may interpret differently. Measurable ranges improve consistency for numeric attributes, while specific atomic labels preserve clarity for non-numeric attributes.

When a user feature represents an average, treat it as continuous and use half-open bins, even if the underlying quantity is a count. The implied averaging window is 1 month for daily or weekly values, and 1 year for monthly values.

Non-numerical attributes are also valid when they use specific, atomic categories such as “Industry,” “Diet Type,” or “Employment Status.”

❌ BAD:

{"categoryName": "Purchasing Behavior", "featuresName": "Frequent"}

✅ GOOD:

{"categoryName": "Online Purchases Average per Month (width 4)", "featuresName": "0-<4"}
{"categoryName": "Online Purchases Average per Month (width 4)", "featuresName": "4-<8"}
{"categoryName": "Online Purchases Average per Month (width 4)", "featuresName": "8-<12"}

Rule 7: Be Specific About What's Measured

Requirement: Category names must clearly indicate the attribute being measured and, where applicable, include the unit or time basis.

Rationale: “Purchasing Behavior” is vague because it does not say what is being measured, in what context, or over what time period.

❌ BAD: “Purchasing Behavior”, “Activity Level”, “Engagement”

✅ GOOD: “Online Purchases per Month (width 4)”, “Exercise Hours per Week (width 3)”, “Social Media Hours per Day (width 2)”

Rule 8: Context-Independent Naming

Requirement: Category and feature names must be self-explanatory without study context.

Rationale: Users don't see the study question when selecting their profile; names must stand alone.

Conciseness: No limit on length, but keep as short as possible while maintaining clarity.

Rule 9: Hierarchy Through Nomenclature

Principle: Start with general categories, then create more granular versions for follow-up studies.

Pattern: Use suffixes to indicate granularity level.

Example Hierarchy:

• Social Media Hours per Day (width 2) = “0-<2”, “2-<4”, “4-<6”, etc. (broad)
• Social Media Instagram Hours per Day (width 2) = “0-<2”, “2-<4”, “4-<6”, etc. (fine granularity for follow-up)

Workflow: Client sees high importance for “Social Media Hours per Day (width 2): 4-<6” in results → submits follow-up study with platform-specific categories to drill down to which social media site is most influential.