Last month I made an appeal in this blog for developers to run a tool that would scan QVWs and contribute anonymous data to a study. Sixty one sites have submitted data so far. As promised, I’m making the results available for download and analysis here.
It turns out to be a pretty interesting data set — if you are into QV. The available data is limited by a few design points of the collection:
1. All returned data had to be absolutely anonymous. This required that counting be done in the scan. I could not bring back the actual expression text.
2. The scanner should not open the document. This means that script was not examined. Only data available in the QVW XML was used.
As a representative sample of the QV world, the sample size is rather small and probably suffers from a significant selection bias.
Please also note some specifics of the counting. Functions in variables were counted, but only once. “Sum()” in a variable was counted as 1 use of Sum, even if the variable was used in multiple charts.
There are some interesting outliers in the submissions, which are intriguing to look at on their own.
So don’t use this dataset it for your PHD thesis but have fun with it. Here are some insights I found.
I would have never guessed that the number #2 Function was “If”!
I used a couple of different ranking formulas throughout the analysis, but If() comes out a solid number #2 no matter how you look at it.
This led to another question (there’s that Data Discovery thing): Is there a relationship between the use if Set Analysis and the use of If()? Not surprisingly, there is a strong correlation.
I’m of the mind that Set Analysis can be overdone and make an app difficult to maintain, vs building function into the data model. I don’t have an exact figure, but I frequently suggest to students that a max of 30% expressions with SA feels about right. It’s a really rough number meant to illustrate the concept that good data modeling is core.
Let’s assume that Swedes are excellent QV developers. There were three submissions from Sweden:
They look pretty good to me. There were submissions from experienced developers whose work I was familiar with and their SA usage tended to fall in that 30%-40% range as well.
What about Objects? I found a couple of surprises there. 
I was surprised that Text Object beat out List Box for the top spot. But the biggest surprise to me was that Button made it in the top 5. I’d love to know what all those buttons are for. If you assume that many people use Text Objects with Actions as a Button substitute, there are even more “Buttons” out there.
There is lots to explore in the data. The analysis has popup and drill down detail so explore with your mouse. And of course create your own visualizations and post your insights here.
Sites that submitted data can select themselves for comparison by selecting the unique submitted QVD name.
I will update the Survey Results app when I receive more data and of course, when you point out my errors J
-Rob
