Leigh and Gans bring Groseclose and Milyo's method to Oz, but use public intellectuals rather than think tanks due to the paucity of the latter. Recall that the original Groseclose and Milyo method didn't benchmark to partisanship but rather to ideology: the US has ADA scores for representatives, so each representative is rank-ordered from liberal to conservative and an imputed ideological score is then given to each think tank. Robustness checks in Leigh and Gans include expert coding of front-page newspaper articles, both political and overall, and coding of newspaper editorials.
Norton says the ranking of intellectuals lacks face validity: that their left-right positions don't correspond to casual observation; moreover, public intellectuals tend to be ideological rather than party partisan so the method may not work.
I certainly don't know enough about Australian public intellectuals to know whether there's a lack of face validity in Leigh and Gans's ordering, nor do I know whether the public intellectuals tend to be party hacks or ideological hacks. But I'm pretty sure that Groseclose and Milyo were able to drop negative or reverse mentions from their analysis (things like a National MP saying, "Even Brian Easton agrees with X", for example) while Leigh and Gans weren't able to given their initial dataset. On first thought, so long as both sides tend to do this with equal frequency, it might just add noise to their estimates rather than much bias. On the other hand, I can imagine those reverse mentions happening a lot more for the tails of the distribution, which would greatly narrow the measured range of public intellectuals' opinions and could lead to a centrist-bias in the results. Leigh and Gans say that folks don't much cite public intellectuals for the purpose of bashing them (a truly negative mention), but I can imagine a fair number of the "Even prominent Labour supporter X agrees..." mentions. In the end, their robustness checks mostly give them the same answers: most papers are largely centrist in Australia - so I'm less worried.
One thing that they can't control for over their period is whether slight evidence of slant in favour of the Coalition government reflected incumbency bias or pro-Coalition sentiment. Presumably they'll be able to test for that in a replication after the next election.
Cool that Leigh and Gans have taken this on. I've told my classes that a New Zealand replication of Groseclose and Milyo is likely impossible given:
- Party line voting makes ADA scores impossible, so we can't benchmark against median voter ideology.
- A paucity of think tanks makes the mapping impossible.
Measuring partisan slant rather than ideological slant seems possible using Gans and Leigh's method. Congrats guys.