At the household level, we can create dimensions such as "Number of Teenage Mothers In a Household", or "Percentage of Income Contributed by Hispanic Occupants", or (in the telephone industry) "Amount spent calling Phoenix from NYC between the hours of 8 PM and 10 PM". We can create user defined dimensions at the household level based on nearly any criteria that can be conceived.

It may be helpful to understand how our system creates these dimensions, in order to open up new possibilities for users who have never had this capability. We can collect any (un)restricted volume (e.g. [un]weighted person count, incomes, expenses, simulations), and then apply a recode against them to convert them from continuous variables to dimensions (e.g. 1-100 is one category, 101-500 is the next, etc.). It is also possible to collect a variety of different (un)restricted volumes and create an expression involving them. The result of the expression is then recoded to create a dimension. The "Percentage of Income Contributed by Hispanic Occupants" above is an example. To create that dimension, we:

• Collect the income from all household occupants (an unrestricted volume)
• Collect the income from Hispanic occupants in the same household (a volume restricted to Hispanics)
• Create an expression from the two collections, which results in the percentage: (HispanicIncome*100)/AllIncome
• Recode the result of the expression to create a household level dimension (0-10%, 11-20%,...91-100%)

If it is preferable to create a dimension based on the presence or absence of a condition, we can create a TRUE/FALSE dimension. If, for example, we wanted to dimensionalize households based upon whether or not more than 50% of the income was from Hispanics, we would collect the same volumes as above, but instead of creating the expression and recoding the result, we would create an expression like this: "HispanicIncome*2 > AllIncome". Each household will then be dimensionalized as "T" or "F".

We can also use multiple types of restrictions in the same volume collection. In the "Number of Teenage Mothers In a Household" example above, the volume collected was "unweighted person count", and it was restricted to people who:

• were females and
• had indicated that they had given birth and
• were less than 18 years old

At the transaction level, one example of a user defined dimension that might be used in the telephone industry is "Average Rate Per Minute", which is calculated by dividing the total price for the call by the number of minutes (duration) of the call. Since this is a continuous variable, it would then be recoded however the user wants, in order to create a dimension. In census data, it might be desirable to create a dimension based upon some relationship between income and education, or income and commuting time, or various types of incomes, or by recoding the result of a simulation

Other examples at the transaction level are found in the Indian problem, where we created one user defined dimension to demonstrate whether or not a person was an Asian Indian (according to Census Bureau rules), and another user defined dimension to correct misclassified ancestry data.

If you have a need to create a dimension, but it is not readily apparent how to do it, please tell us what you are trying to do. We have often been able to produce solutions that are not immediately obvious.