My father and I along with his grad school buddy, Doug Southgate, have been working on a paper on return migrants in Mexico. That is Mexicans who come to the US than return home. Using a survey of Mexican households we found those who had returned from the US, in some regions, were less likely to be employed than those who never left. When we submitted it for publication the reviewers suggested that something other than the migration itself may be the cause of the employment differences. For example perhaps economic conditions are poor currently in some regions causing lower employment and higher migration (leading to more return migrants). To control for this we needed an instrument, that is something related to migration but not likely to be related to current employment. We found a paper , by Gordon Hanson and Christopher Woodruff where for a different study the authors had used 1950s State level migration to control for the possible bias. We utilize their data to help control for this fact and once we put it in our model the negative effect of being a return migrant was no longer there. Instruments can get even more extreme. The freakonomics blog, points to a new paper that utilizes 19th century rail road patterns to instrument for where people chose to live to explain segregation. In fact I later found out the authors who used the 1950s state level migration data later found data on late 19th century railroad access. They used it as an instrument which was linked to past migration and past migration helps future migration.
Instruments can get even crazier, but those are the kind of instruments economists play. Perhaps if we find more Mexican instrument we can start a marriachi band?