Monday, May 11, 2015

Q&A - 11/5


Pollsters failed to predict the UK election results.

Their sample must be biased

Calling people up and collecting answers is only one way of making this prediction. I am sure other models could have done a better job. I haven't applied Time for Change model for the UK election (and it is not advisable to apply a model as-is, with the same variables, from one country to the next [litegeek], if one chooses to however the coefficients / weights of the variables should be recalculated as they would now be based on different data points [/litegeek], but let's say UK and US are similar in that they are both old democracies so the variables are the same (not in value) for both countries; then... Cameron was running for a second term, he gets no punishment for two-term'ness for the incumbent. GDP growth, anemic, but there, at least not negative after a huge crisis. Net approval, around -1 percent, you don't love him, you don't hate him - but not a Bush-low which was -37 in 2008. In summary, UK electorate did not feel the need for change. According to the TfC model the election is always about the incumbent party.


What do you mean with bias?

An example

During WWII, statistician Abraham Wald was asked to help the British decide where to add armor to their bombers. The picture he looked at was roughty like below, black parts are where the plane returning from fights had the most bullet holes.

After looking at this, Wald suggested the military armor the white sections, the parts with least bullet holes, because the data was biased. He was only looking at the planes which returned, suggesting the parts with bullet holes are where planes could afford to get hit at.


You say patent system does not work but I know examples X,Y,Z when it did!

It will work if people don't push it too hard

Patents are an example of a system that can function if it is not used all that much. But when all actors in it are combative, trying to use it, exploit it, and break it all at the same time, then we'll have problems. Now there are too many  users of the system who are parsing lines, meanings of patent descriptions, reading between the lines of the texts, and some, ironically, are trying to be vague so the patent can apply as broadly as possible. This is not an optimal setting.

And this is precisely what is happening today, as economies are becoming more knowledge-based, more people with diverse interests are applying for patents which is overwhelming the system. As every software architect knows, there is a huge difference between a system built for 100 users and one built for 100,000.  Increased load can break a system in weird ways. "Bugs come out of the woodwork" would say a senior architect friend of mine whenever one of his projects started to conduct load testing (simulating a lot concurrent users on a system). A system more suitable for the 21st century is needed. "Better patent lawyers, better judges" won't fix the problem, just like asking for "more and better teachers in classrooms" won't fix the education problem. The purported solution is a practical impossibility that suggests a new approach must look into the structure of the system, not simply trying to potato-stamp the existing order.