The cost of a real work test is much higher, both to the company, and to the applicant. Some companies take that approach (weebly for one), but it means that some of the best programmer will not apply, and it only makes worse the problem of having to aggressively pre-filter (based on credentials, or something else).
You are 100% correct that most research into interviewing has show it to be far less predictive than companies believe. However, I'll add that no one is doing the false negative studies that would really be required to show this. For example, Google has shown that among people who pass their interview GPA is not correlated with success. This does not at all mean that GPA is not correlated with programming ability. In fact, it almost certainly is. People with low GPA who pass google interviews have a lot of other things going for them. To really analyze this, you'd have to give a job to a random sample of applicants. No one has done this. My goal is for Triplebye to get large enough that I can.
You are 100% correct that most research into interviewing has show it to be far less predictive than companies believe. However, I'll add that no one is doing the false negative studies that would really be required to show this. For example, Google has shown that among people who pass their interview GPA is not correlated with success. This does not at all mean that GPA is not correlated with programming ability. In fact, it almost certainly is. People with low GPA who pass google interviews have a lot of other things going for them. To really analyze this, you'd have to give a job to a random sample of applicants. No one has done this. My goal is for Triplebye to get large enough that I can.