IDENTIFICATION PROBLEMS IN ECONOMICS. Kids, a rough way of looking at identification problems is to consider the problem of trying to determine the effect of more than one variable in causing something to happen. A good experiment, as explained by Richard Feynman, will proceed to eliminate all alternative explanations so that there is only one variable which can have a causal effect. In a randomized controlled experiment, controls are used to changes in any variable other than the one that is being studied.
I found an explanation from an economist’s point here. It has a good explanation of the standard example in economics—how do you tell whether a change in price was mainly caused by supply side factors or by demand side factors? However, the other example in the post is perhaps more familiar: “Suppose you want to know the effect of an extra year of school on wage rates. You could collect information on individualsâ€™ earnings and years of schooling and estimate the relationship between the two. But thereâ€™s a problemâ€”the people who got a lot of schooling are not a random sample of all workersâ€”in general, people with more schooling stay in school for a reason. And one reason is that school may be a better investment for ‘high ability’ workers than for ‘low ability’ workers. If thatâ€™s the case, then the premium that workers with high levels of schooling receive over others will reflect both the effect of additional schooling as well as higher average ability levelsâ€”in other words, those workers may have earned more even if they had stayed in school no longer than the low-schooling group.
So how do you disentangle the effects of ability on wages from the effect of schooling? How do you solve the identification problem?”