The link for this article was posted on another board.
Mathematical confusion in how Democrats, Republicans compare
By Scott Peterson
An engineer decides to do an experiment. Every day for a month, he sets his alarm for 4 a.m. and gets up. Every day without fail within two hours, the sun rises in the east. He concludes that his getting up at 4 a.m. causes the sun to rise.
We all know this is ridiculous, and in technical terms, is a confusion of the difference between correlation and causation. Yet it is just this type of mathematical/statistical/logical error that permeates Yagil Hertzberg's commentary comparing Democrats with Republicans. (To read the original, click here.)
Evaluating Hertzberg's editorial, I found examples of four statistical fallacies or analysis errors (there is not nearly enough space to enumerate every logical error). Those included:
1) Causation versus correlation -- Almost all of the analysis regarding family values falls into this basic error. For example, there is no mechanism that can suggest that voting for Party A leads to lower incidence of rape. More likely, states with higher crime rates might vote for Party B because they think that party will take care of the problem. The causal relationship implied by Hertzberg is more likely reversed. Just because two pieces of data are tied together does not mean that one piece causes the other.
2) Selective time periods -- Hertzberg chooses time periods for his economic evaluation that are beneficial to the conclusion he wants you to draw, not for any independent evaluation criterion. He conveniently leaves out the Obama administration when it comes to deficits and jobs, and if we go back to 1932 instead of 1960, the entire analysis of jobs gets turned around.
3) Missing independent variables -- In this type of error, the analyst leaves out significant variables that might have a significant effect on the outcome, and puts in only those that make a particular case. Hertzberg does this in almost every example he cites. For example, he quotes state by state statistics for things like household income, poverty, and all the family values categories that are much more likely to be influenced by state and local political policies, but only evaluates them against the national presidential election. Crime statistics, for example, are much more likely to be influenced positively by that particular state's willingness to let their law-abiding citizens own and carry guns (a Party B position) as opposed to who they voted for in a presidential election.
4) Aggregation -- In this error, the analyst takes long periods of time, and aggregates data across it without evaluating the changes of policy or condition that occurred during that time, and then claims the single factor at the top of the analysis was the cause. One glaring example of this is in the discussion of deficits. Consider that during the Clinton administration, increasing tax rates (or voting for Clinton) did not reduce the deficits for the first two years, instead they stayed flat. Two years in, the Republicans took Congress and slowed the rate of spending, and 5 years in, they reduced the capital gains tax rate (a typical Party B policy) which (along with the dot-com boom) led to significant economic activity and significant tax revenues. To claim that the election of a "Party A" president was the CAUSE of the reduced deficit is pure nonsense.
Nothing in Hertzberg's analysis passes the basic mathematical/statistical/logical tests for determining causation. Sadly, he employs a slick trick in trying to get you to believe that if you vote for Party A, you will get all the benefits in the categories he chose (notice he left out things like inflation and interest rates).
When you read such analyses, start to consider the kinds of errors I listed here. Otherwise, set your alarm for 4 a.m., and start to feel powerful that you can cause the sun to rise.
Read more: http://www.sfgate.com/cgi-bin/blogs/opinionshop/detail?entry_id=72534#ixzz0zzLWdUZg