Sunday, April 13, 2008

media bloopers - (mis)counting voters

Here is a serious media blooper. The Financial Times of all places is having problems with numbers. Yesterday the FT had yet another of these stories on the Pennsylvania primary coming up. See: 'Reagan Democrats' Switch their allegiance.

The core of the article is a bar chart showing how the numbers of registered Democrats and the number of registered Republicans has flipped in just the last 6 years. That chart looks like this:

The problem is not that the numbers are just wrong by a little, but the 2002 numbers are backwards. It says that as recently as 2002 the number of Republicans in the state exceeded the number of Democrats by more than half a million. It's the other way around! Check the numbers yourself: here is the table of data for 2002. It clearly says 3.2 million Republicans and 3.8 million Democrats. That chart and the text of that article are based on it being the other way around.

I can guess how this happened. If you go through the data for each year and what you will see is that in 2002 Republicans were listed the first column while Democrats were listed in the second column. In all susequent years Democrats were listed in the first column. Somebody didn't notice the switch and assumed the 2002 data had Democrats listed in the first column. I noticed this when I made a similar chart of the time series for registration data which I posted earlier. I actually extracted the data not realizing the columns switched, but when plotted it was immediately obvious that the big flip in registration was a data issue. It's not really a question of fact checking, it's an issue of applying the eyeball test.

So rework the bar chart in your head and reread the article to see how much of it still makes sense. Yet, I bet most will internalize the headline and I will occassionally hear the false factoid that registration recently flipped in the state for years to come. Maybe they will issue one of those corrections nobody reads.


Post a Comment

<< Home