Venns

Tuesday, August 17 - Thursday, August 19

Song of the day: Worst Song Medley - Garfunkel & Oates

Venn Diagrams.

We all know what they are.

They're used to highlight differences and similarities in an easy to comprehend image.

Piece of cake. Child's play.

Not for this lone wolf.

I'm working with Jason on some of the data analysis for one of our clients.

It's actually great that I get to help because the more work I do increases the $ that comes.

Presenting findings and the data to support it is usually both boring and complicated.

Our job is to spell it out for our clients. Simplify the message and make it visually appealing.

Generally this means graphs and images in colors and font changes.

Jason asked me to find meaningful cross-sections between our 22 data fields and visually represent them as Venns. I was to use Microsoft Access for the first time.

Our data was currently in Excel but Access, as a relational database, was better for this type of work. I had created and captured our data without the back-end in mind; so instead of simple numbers or short answers; most of our work was in fragment sentences. Access doesn't like punctuation or contractions so commas, apostrophes and semicolons are the enemy. I was not aware of this now, but it would bite me in the ass 2 days later.

Jason taught me the basics to Access and slowly I began.

I needed to see understand which fields of data compared strongly to other fields (high %).

I would then take these percentages and find the % of total intersection and then create the Venn using the different percentages to show each overlay.

The idea is to show what ideas have the most commonalities and which pieces fit (or don't fit) together to tell the best story or identify a main theme, factor or cause.

For example;

By this picture, let's assume that this Venn only depicts men and that the sample proportion of Fedora Hat warers, Epic Adventure story tellers and Body Odors culprits are the same.

We can judge that by the approximate sizes of the 1x1 overlaps that they are similar in number and look to be about 12% of each population. The 1x1x1 overlap looks to be about 1/3 of the 12%. We can then estimate that if you see, hear and smell a fedora wearing, odor-licious, crazy adventure story teller, that it'll turn out to be your grandfather 4% of the time.

The math isn't necessary to explain because it's easy to grasp with the picture, right?

There's a small chance that it'll be your grandfather, but it's possible.

That's the quick punchline I needed to be able to tell.

I'll spare the details, but it took 1.5 days of struggling with Access to confirm that the punctuation was screwing with the data (I was getting answers that I knew were incorrect). Jason recognized the reason and quickly taught me to translate the "word" data into simple numbers and use a key so that Access wouldn't choke on the punctuation anymore.

Same verse, same as the first, a little bit louder and a little bit worse.

Day 3 begins but I'm not exactly starting from scratch (though it feels like it).

Now I understand Access a lot better and know the relationships I want to try and the ones that I want to use to try to disprove my theory (in order to kill any self-fulfilling prophecy bias). The goal is to see how everything overlaps first and then identify the major players.

Venn Diagram of Weirdos

Clearly this is way too much information for the client and you can use simple math and Access so you don't waste time graphing things unnecessarily, but it paints a picture for us.

It's pretty easy to see that "Cosplay" correlates strongly with a couple other Weirdo factors and "Fat People Who Like Anime" have smaller ties to numerous other Wierdo. These would be two of the factors that we would start to test heavily.

Another day passes and I have a strong case for 4-5 of our data fields. Once I identified the culprits and double checked the math, my Access work was complete. On to it's much more familiar cousin, Powerpoint. Now the task was to generate proportional circles to data sample size findings and estimate the overlaps based off their percentages. It wasn't quite as easy as I had hoped (I'll be honest, I got sassy after finally conquering Access). Interweaving 5 cross-sections was much harder to eyeball than I planned.

The picture is supposed to represent the math so it's highly critical that it "looks" right.

For example;

Based on the size of the circles and the sizes of the overlaps it's very obvious to see that Denzel has had the most movies with facial hair and the least movies with just hats and glasses.

At the end of day 4, I finally have three strong graphs and stories represented.

Sadly, they're only 15% of our presentation.

Jason knocked out the remaining 85% within the same time. DOH. LOL

However, I can now see this taking as little as 5-6 hours for the next time.

And I know I will get better.

Several big things has come out of this:

1) I'm learning Access and database relationships (which is awesome)

2) I' m going to continue to help Jason with analysis and presentation;which means more formulas, graph types, causation relations, etc

3) I'm getting better at not beating myself up when I don't immediately grasp a concept or do things right the first time.

The last is the hugest.

I've been putting myself on a roller coaster ride when I've felt inadequate or fell below expectations. I'm incrementally letting go of the stress and anxiety I heap on myself.

It's coming excruciatingly slow, but I'm making progress.

One of the final products :)