- For many years, the New York City police engaged  in stop-and-frisk, a policy of police officers stopping citizens on the street and searching  for hidden weapons, drugs or other items. Stop-and-frisk disproportionately targeted  racial minorities and escalated a sense of racial injustice of New Yorkers. Despite  claims to end Stop-and-frisk, former New York city mayor Michael Bloomberg was convinced  that the policy was needed to reduce crime. And Bloomberg was not alone: public  officials in favor of stop-and-frisk would leverage data on crime rates to back  up their claim that stop-and-frisk worked. Now we know that stop-and-frisk doesn't work and  does more harm than good. How could policy makers be misled by the data to the point of maintaining  a policy that does more harm than good? To understand how these mistakes can happen, we must master a crucial concept:  confounding variables. The defense of stop-and-frisk was flawed because it  failed to account for confounding variables. By completing this video, you will be able to  identify the criteria of confounding variables, propose a confounding variable  when presented with a hypothesis, and justify your proposed confounding variable  using the criteria of confounding variables. Suppose we collect data and find that countries  that invest more in clean energy research pollute more than countries that invest less in clean  energy research. In other words, there is a positive correlation between investment in clean  energy research and carbon dioxide emissions. Should we conclude that investing in clean  energy research makes countries pollute more? No, we shouldn't because as it turns out, richer  countries can afford to invest more money in clean energy research than poor countries. Richer  countries can also afford to consume more energy from fossil fuels, which leads to more carbon  dioxide emissions. So richer countries can invest more in clean energy research  and emit more carbon dioxide. This means that wealth is a confounding variable.  So there are three criteria that allow us to identify a confounding variable. These criteria  are: they cause change in the dependent variable; they are correlated with the independent variable;  and they causally prior to the independent variable. This means a confounding variable  comes before the independent variable, not after. So notice how, in our example, the variable  "wealth" satisfies all three conditions. First wealth causes change in the dependent  variable. That is, being wealthier, makes a country emit more carbon dioxide. Second, wealth  is correlated with the independent variable. That is, wealthier countries can invest  more in clean energy research. And finally, wealth is causally prior to investment  in clean energy research. In other words, being wealthier comes before investing  in clean energy research, not after. To be clear, pointing out a confounding variable  may give us reasons to question a hypothesis. In our example, think about the following  hypothesis: investing in clean energy research causes countries to pollute more. If  we didn't know about confounding variables, we could have thought that the evidence was  in favor of this hypothesis. Knowing that wealth is a confounding variable in this case,  is what makes us skeptical about the hypothesis. We can see from this example that when  we propose a hypothesis, we must think about the confounding variables. If we didn't  know that wealth causes both investment in clean energy research and carbon dioxide emissions, we  could have been tricked into stopping investment in clean energy research, and that would  make climate change even more disastrous. Thankfully, we know that wealth is a confounding  variable in this example, but this is not always the case. Public officials may overlook  confounding variables and implement bad policies. That is exactly what happened with stop-and-frisk. When confronted with claims to end stop-and-frisk,  New York city public officials would defend their stance by showing how crime had decreased after  the police department implemented the policy. So we can see in this graph that indeed  crime rates fall after stop-and-frisk starts. However, if we expand the time span of  this graph, we can see that crime rates have been falling since the early 1990s -  long before stop-and-frisk was implemented. In fact, crime was falling for  reasons other than stop-and-frisk. One of the variables known to affect crime  rates is the number of employment opportunities, which have increased since the early 1990s. We  can see that employment is a confounding variable because it meets all the criteria. So first,  employment causes change in the dependent variable (employment causes crime rates to decline).  Second, employment is correlated with the independent variable (when employment was low,  there was no stop-and-frisk and when employment was high, there was stop-and-frisk).  And finally employment is causally prior to the independent variable. So the rise  in employment came before stop-and-frisk. The last two steps to identify a confounding  variable are crucial. In my experience as a TA, I have seen many students talk about confounding  variables by looking only at the relationship between the confound and the dependent variable.  Remember that for a variable to be a confound, it must be correlated with the independent variable  and it must come before the independent variable. So to sum up, the objective of this video  is to teach you to identify the criteria of confounding variables, propose a confounding  variable when presented with a hypothesis, and justify your proposed confounding variable  using the criteria of confounding variables. People who don't master these skills  will be subject to being misled by data, just like public officials who  said that stop-and-frisk worked. So mastering the concept and application  of confounding variables is crucial for social scientists who want  to contribute to a better world.