Good analysis (not light reading) - that the numbers vary so widely depending on what range of years is chosen indicates that linear regression is the wrong tool to determine the relationship (if there is one).
Not to mention that even if the correlation was correct, there are obvious explanations other than 'H-1Bs cause more jobs' - for instance that both are signs and consequences of general economic health.
Such a study would make us think that bringing in job seekers at a time of high unemployment would be a good thing. Not.
Even if it were true, there would not be an immediate turn around. It would probably take years and years.
So, the time to bring in immigrant job seekers is in a thime of high EMPLOYMENT. They would threaten no one’s job, and they could possibly extend a rising market.
Putting an American out of work in a time of high unemployment doesn’t make sense. Nor does bringing in an unemployed immigrant.
Not to mention that even if the correlation was correct, there are obvious explanations other than 'H-1Bs cause more jobs' - for instance that both are signs and consequences of general economic health.
Good reading! You sound like a math major or someone familiar with this kind of analysis. I agree that the wide variance of results depending on the range of years suggest that it's a very poor model. It's funny how the study just happened to pick the range of years with the best result!
Also, you're very much right that "correlation does not imply causation". When I communicated with Zavodny, she even said that she did "prefer to not use strong causal language" although her study did in places and the parroters of her numbers definitely do. I wrote a section on the correlation issue at this link. As I quote at the end of that section, "[correlation] is also one of the most abused types of evidence, because it is easy and even tempting to come to premature conclusions based upon the preliminary appearance of a correlation".