Monday, February 09, 2004

"A Massachusetts Liberal"
What does this Mean?  Is there something wrong with being a Massachusets Liberal?

Another lesson in interpreting statistics:

[Left brain stuff:] Today, on The Connection, an NPR broadcast hosted by Dick Gordon, Mr. Gordon cited a poll showed that showed 21% of potential voters would be disinclined to vote for John Kerry because he is from Massachusetts.  The implication is that there is a cultural significance to the label, "Massachusetts Liberal."  Further, that the Massachusetts Liberal label is pejorative.  I was not able to find the survey using Internet search engines.  I listened to the Internet stream of the broadcast a couple of times to try to catch the reference to the survey, but there was nothing specific enough for me to be able to find it. 

This 21% number, what does it mean?  In order to interpret this properly, we first have to know what number you would expect to get if the Massachusetts label had no connotation, positive or negative.  It is tempting to assume that it would be 0%; that is, that any deviation from zero is an indication that the label has a positive or negative connotation. 

Without knowing the methodology of the survey, we cannot really say.  But we can say that it is not safe to assume that the 21% means anything.  To illustrate, let's say that we find out how the survey was done, and we decide to repeat it.  Let's say that we get 21% again.  That would tend to confirm the precision of the survey, but it does nothing to confirm the accuracy.  Remember, precision refers to the replicability of a measurement.  Accuracy refers to how close the measurement is to the truth.  

Ok, so repeating the experiment doesn't really tell us if the 21% is a measure of a negative connotation.  Let's do another experiment.   We repeat the survey, using the same method, but this time we use a hypothetical candidate and we sample 50 times the number of people in the original sample.  2% are asked about a hypothetical candidate from Alabama, and then we repeat the process, specifying all 50 states at random.  What would happen?  Would you get 49 zeros and one state, Massachusetts, would come up with a 21% disinclination rate?  Of course not.  In fact, you probably would not get very many zeros. 

In The Connection, the panelists were asked if the term "Massachusetts Liberal" actually means something.  The statistic cited above was used to show that the label is a liability for a candidate.  But it very well could be that if you compared it with all other states, the 21% could be a low number relative to others. 

The take-home lesson is that you can't tell if a survey supports your hypothesis, unless you also know what the results of the survey would be if the hypothesis were false. 

[Right brain stuff:]  Why take up valuable bandwidth to make some kind of esoteric point about statistics?  After all, hardly anyone cares about statistics.   Reagan proved that statistics don't matter.  Statistics show that most blog readers care about politics.  I care about politics.  I want everyone of voting age to be an informed voter.  I want everyone to know how to understand news reports, think about them critically, and draw valid conclusions. 

An often-expressed concern about the news media is that they might influence a political process by amplifying an early trend -- which could be meaningless -- and turing it into an inevitability.  If we all are able to interpret news reports carefully, there will be little risk of the media having such an influence. 
(post #036)