Let's Get Specific

As a Researcher for a university with multiple campuses, we grapple every day with how to maintain a unified identity between two very different campuses in two very different socio-economic areas. Each has its own constituency, strengths, weaknesses, challenges, and threats. While I am a fan of "big-picture" thinking, I do not find it to be as useful in analytics. In cases like this, sometimes data mining is more effective on a local level.

Variables that indicate affinity for one campus will not carry over to the other campus.
For example:
                                          Campus A                                     Campus B
1.  Population served       Immigrant/ blue collar pop.         Upper-middle class pop.
2.  Location                      Urban                                             Suburban
3.  Academic focus          Sciences                                         Liberal Arts & Business

Because the foci of these two campuses and the student populations are so different, so will be those factors that create a higher alumni affinity. If sports participation is an indicator of affinity at Campus B, it may not indicate affinity at Campus A.

This is not to say that a wholistic perspective is useless--quite the contrary. Creating an overall predictive model can help to inform your analytics efforts on the local levels and speed those processes along. But it is often easier to start with a smaller scale and build outwards, refining and redefining your models as you go. The process is longer and more drawn-out; however, with patience the results are worth it.

Data Mining ROI

Microsoft Word - Does Data Mining Work

View more documents from Tommy96

So Much Data So Little Time

Well, now that the Data Entry Manual is nearing completion (::APPLAUSE::), one of the (many) pressing thoughts on my mind is how to bridge the gap between how data is stored in RE and how we agreed to store data in RE. If all the components are not in place, end users will not follow the data entry instructions. So a good system for data entry problems and concerns is necessary.

After considering options, I have decided to review the RE Manual one Chapter at a time and compile a list of changes that need to be made. Though this is slower going, the end users will find it easier and less time-consuming to enter the data and thus be more likely to do it.

Priority Tabs:
Attributes
Relationships: Education
Volunteer
Prospect
Appeals
Honor/Memorial

To see the current draft copy, please go here: LIU RE Data Entry Policies and Procedures Manual

Work To Do:
Revisit LIU Affinity Score
Create revision list for RE
Look into Prospect Rating and Management Systems

Personal To Do:
Post on personal Blog
Register for Spring Library courses
Work on Resume
Finish Censorship Paper

Great Seminar on... Statistics!?

"The first one was too hot. The second one was too cold. The third one was JUST RIGHT."


I must admit, I was a bit doubtful when I boarded the LIRR at the Douglaston Station yesterday afternoon. I skipped lunch and left work early to attend a conference-style seminar on predictive analytics. I have been to many seminars of the like that were less than stellar. Many were either too basic or too theoretical to be of practical use to anyone in this field.


I have been dabbling in data mining and predictive analytics for 3 and a half years. My one achievement was being able to justify our need for data entry and data entry standards. This led to a 2 year project to create a Raiser's Edge Manual for data entry. While it is not the most intriguing project (and at times more tedious and frustrating than learning to knit). As far as actual analytics goes, I have been stuck in a rut of not understanding the NOVA chart output enough to weight a model. I was stuck in a flat binary world of 0's and 1's.



Isn't binary Sudoku a bit bland?

Well I'm glad to report that problem as solved and after David E. Robertson's seminar, "Zen and the Heart of Predictive Modeling." For information on David and the seminar go here. The world of predictive analytics is looking up for me once again.

I'm also glad to report having met another person with my first name and two other great APRA people who I can't wait to see at the holiday party in December. Woo hoo!