It is a well known fact in economics that statistical correlation does not imply causation. If in the data two variable increase or decrease approximately at the same time then correlation is positive and if they go in the opposite directions approximately at the same, correlation is negative.
Probably the most common example of significant correlation that does not say anything about causality is positive correlation between wage and education in the data. One person can say that people get higher wages because they know a lot and it benefits them when they are hired. Another person can say that higher wages allow people to get better education. And both would be right. However, there can be a third person who would look even deeper into the problem and say that there is another factor that influences both previous ones such as IQ or inborn smartness. It is easier for a person who is smart to learn (for this reason we might see higher education levels). At the same time employers might not care about education per se, but about how smart you are and how much you can contribute to the company. See the graph below (please forgive my drawing skills).
OK, so where is the problem? The problem is that the data on how smart people are is not available. Moreover, it is very hard to measure and involves measurement errors that would equally screw the results. This is what the largest oval is about – we observe what is inside it but we do not observe what is outside it, in this example – data on IQ. The part of the graph outside the largest oval is part of the theory behind the relationship between wage and education. In fact, the third guy’s statement is the most difficult to come up with. It is always easier to think in terms of variables that you already have in the data, but IQ data is not a part of it. Definitely, the theory between wage and education without IQ would not be complete and it is possible to end up with a wrong result.
Great, but how is it related to housing problems? In one of my previous posts I posted two pictures: interstate mobility declines over 2005-2006 and drops by approximately 35% (pic). At the same time volatility of unemployment increases a lot in 2008 (another pic from the same post). Here is a new picture that depicts correlation between unemployment rates across states and number of houses that are “underwater” (Value of your house is less than you owe for it in loans). Sources are BLS (for unemployment rates) and CoreLogic (for equity reports).
Nevada has 62.6% houses underwater in the first quarter of 2011 and 12.9% unemployment rate…So it means that if we delay foreclosures on those houses we increase unemployment in this state relative to others by for example reducing labor mobility? Right? Former governor of CA got an answer ( youtube video with sound). Not quite, even if there was a real decline in labor mobility in the states with a lot of people underwater that won’t mean that one caused the other. It might not be foreclosure delays per se that lock people in their house and thus they should not be implemented. Moreover, the HARP intervention was poorly designed and resulted in high redefault rates. But this information is not in the data, it is “the third guy’s logic”.