Meanwhile, I have obtained some information on the Gruber estimates--sorry for not updating in the post, but we're having some technical difficulties that make it impossible to update cross-posted entries. Gruber says he started with the CBO's figure of $4700 (he says $5,000 but I assume he used the correct figure in his model) for a single policy, discounted it for inflation, and then multiplied it by 2.7 to get to the family numbers. I still can't make his numbers work unless I use different inflation rates for different types of policies, but I'm not sure how much that matters.
But I'm not comfortable with the decision to ignore the CBO's family numbers, which are considerably less rosy than his. The CBO presumably has reasons for picking the multiples it did, including expected changes in the composition of the family market--in fact, their estimate specifically states this:
Compared with family policies that are expected to be purchased in the exchanges, family policies purchased in the current-law nongroup market cover fewer dependents, on average. That difference largely explains why the ratio of single to family premiums differs across those settings.
Of course, covering more family members is presumably valuable for people. But the fact remains that they're going to have higher premiums, and no choice about paying them.
Update: I can get something very close to Gruber's numbers if I assume 6% annual inflation for 6 years, rather than the seven I was counting between 2009 and 2016, and that young, healthy workers get virtually no discount in the current system over what Gruber estimates they would receive in a mandated, community rated system. So mystery solved, except for the part where we have 6% annual inflation.
The problem with ignoring the CBO's family estimates remains, however.