14 - Organizational Intelligence

A TPS at a university may be the payroll system for the faculty and staff or the student grade system. For each system you would need to determine the number of entries per year, and the size of the fields contained in each entry. For instance, consider a simple payroll system where the data generated each pay period are just the social security number and the amount paid per employee. The social security number is a 9 digit field and the amount paid might be an 8 digit field. Given these dimensions and an employee count of 2,000, the amount of data generate each year would be :

1. (9+8) * 2000 * 26 = 884,000

This is a very abbreviated look at a system, other fields would be included in most systems, and it is worth revisiting the data model on page 173 to get some ideas of these other fields.

The university may need a variety of data for decision making, coming from within the university and from external sources.

For instance, to plan for incoming students, demographic information would be important, as would drop out rate (internal data), dorm capacity (internal rate), and the faculty/staff availability.

To hire new faculty, the university may need information about average salaries for a particular field, current department size, the teaching load, number of students enrolled in that college, and average salary of current faculty members in that college.

There are many more decision making areas for the university and each will have its own needs for data (both internal and external).

Dirty data that may be found in a university's database system include:

• Duplicate student records (one in the billing department, one in the grading system, and perhaps one in payroll);
• Duplicate faculty records (one in the payroll and one in the grading system);
• Multiple entries for the same student (for instance, if they changed their name or they took some time off and re-enrolled in the university);
• Misuse of data fields (a database may have been designed prior to the widespread use of e-mail and used a non-populated field such as fax number for the e-mail addresses).

A university may not need to revise its data warehouse frequently. Perhaps once term would be adequate.

Examples of verification questions:

• What is the average GPA for MIS students?
• What is the average age of a university student?
• How many students have a 4.0 GPA?
• How many students have a home address from the same state in which the university resides?
• What is the average number of class hours a student registers for in a term?

Examples of data discovery questions:

• What is the best predictor of GPA?
• What is the best predictor of dropping out?
• What is the profile of the best teachers in the university?
• What is the best predictor of a student becoming a philosophy major?
• What is the best predictor of the timeliness of student tuition being paid?

Apply the basic prompts

It is possible that data mining will uncover relationships that are not causal relationships. For instance, it may find that half of all the people who bought Spam also live in a specific zip code. The purchase of Spam may have nothing to do with zip code but instead be related to dwelling type (i.e. apartments). Therefore, targeting a specific zip code would not be as effective as targeting a dwelling type. To avoid such pitfalls, the data needs to be scrutinized by the analysts to ensure the conclusions made about data patterns are causal and not just coincidental. Also, do data mining on a portion of the database and test the findings against the remaining data to test that the relationship still holds.