As a dedicated SAS employee, I take advantage of as many offered perks as I can, and that includes attending the annual SAS Winter Party. My wife and I have not missed one since I started working here, which means that last weekend we attended for the 16th time.
Every year we make the funny observation that the winter party coincides with what feels like the coldest day of the year. I figure that the weather is somehow engineered by the coat check staff at the Sheraton hotel (party venue) to ensure a prosperous evening.
But does the winter party really coincide with below-average temperatures, or am I just whining? (Note: I would never actually complain about this. I love the party and would brave subzero weather to get there. I just wouldn't hang out around the ice sculpture so much.)
To check the answer, I grabbed some historical temperature data and then merged that with the dates of the winter party since 1995. This was simple SAS DATA step and PROC SQL, all generated in SAS Enterprise Guide. Then I created a plot with the SGPLOT procedure to see how many years we had to endure cold temperatures to see bands like Sleeping Booty.
As you can see from the plotted result above, we've had some cold party years. But we also had some mild years. My conspiracy theory about the coat-check-climatologists is not exactly solid.
Last week I wrote a blog post -- internal to SAS -- about this analysis. I received all sorts of helpful advice about how I could "improve" my plot to make my case more convincing. This set of suggestions came from Robert Allison:
- Throw out the 1995 data.
- Don't show the "high" temp bars.
- Start your y-axis higher than zero, so the bars look shorter/colder.
- Put a snowflake image in the background, to make people "feel" colder when they look at the graph.
- Add "hover text" charttips to each bar, which say 'xx degrees below average'.
Taking some of this advice, I produced a second plot that shows the same data, but with some misleading techniques applied. I converted the data to centigrade to make it seem lower (but left the "average temp" annotation in Fahrenheit, clearly labeled). I changed the max value on the Y axis to 30 degrees (which is really about 80 degrees Fahrenheit) so that the party day temps look closer to 0. And I did throw out the 1995 data, which was unseasonably warm. Finally, I post-processed the plot by adding a stylized image of my colleagues dressed in formal wear -- in the form of penguins.
If you care to see the code behind this overt slight of hand, I've posted it here.