Even these regression analyses of official government data, however, may not be telling the whole story. For instance, there may be other inherent biases in the data that researchers simply cannot control for. For example, it could be the case that unions tend to organize workers in workplaces that pay higher than normal wages to begin with. And there’s some evidence to this effect: unions are more likely to organize in larger, more productive workplaces, which is certainly more attractive for unions, because they can collect more in dues from firms with a larger number of employees. On average, larger, more productive workplaces pay higher wages than smaller, less productive ones.[9] So, in this particular case, workers in these larger firms would have already been paid more than their counterparts in smaller firms, even if they weren’t unionized.

These limitations of government data leave researchers in a position of not being able to confidently answer the question about the average difference in wages between a union and nonunion worker. The data clearly suggest that the figures that unions themselves cite — 27 percent higher wages for union members — is far from correct, but they do not provide an opportunity to confidently estimate what the real average wage premium is.